No you haven't, the main reason for this video is probably that GTX680/770 (same card) can be found for relative cheap and offer decent performance in nowadays crazy GPU price world.
Hey HU! For the next 'Throwback Thursday' benchmark, I would like to see an in-depth analysis of the 280x and R9 285 in terms of texture quality. This would be an eye-opener to AMD's improvements with tessellation support and memory compression. Thanks!
@@stanisawszczypua9076 Exactly what i have done now in 2021, got a Gainward 680 2gb for 60 euros. Working like a charm.... Not being updated anymore in driver terms, but still doing its job
Pro Tip: Those that scream the loudest and say you did it wrong etc tend to be wrong 95 out of 100 times. Glad you did this to quell the misinformation that always wallows around. great video Steve thanks for all your hard work.
I find those screamers not even reading descriptions on YT not to mention doing a proper research. Many people these days are just to F'n lazy and yet they demand everything be their way.
Heya, Steve! I can confirm, I own both an EVGA GTX 680 SC 2GB and a EVGA GTX 680 Classified 4GB. Performance isn't much different, however to play games with DSR and newer games it does make a difference, for example, on COD WWII (at least in the launch version) I was getting really bad stuttering, I had to settle for 720P with my 2GB 680, while my friend (using my 4GB 680) could play it comfortably at 1080p high settings, same could be said when I swapped out my GTX 680 2GB for a Sapphire HD 7950 3GB. Another example was when I was trying to play GTA V at 1440p on my 680 SC, I could play fine but after a while, it would start stuttering very hard, going from 30ish fps to 8 and at some point it would just hard crash I believe. GTX 680s 4GB come at a premium and they're only worth buying if you're on a tight budget and not upgrading anytime soon, but if they're over 120$, you might as well go for a GTX 1050 Ti as in most cases, it performs better and has better driver support anyway. Cheers for the video Steve, good on ya.
Yeah COD WWII was a massive memory pig upon release, they have fixed it now. I was having issues with 4GB and even 6GB cards when I first tested it and 16GB's of system memory was a must. Thanks for the feedback though mate, appreciate it.
I had the same issue with my GTX 980 running out of vram in UT4 pre alpha. In the fully meshed maps it often filled the full 4GB and tanked in performance when it did so with massive lag and juddering.
It only affects performance when there's not enough space in vram for "immediate scene", and when what's left for cache in vram is so small that the driver have to drop and reload the resource between ram and vram frequently, which mostly result in stuttering when quickly rotating the view or moving really fast. It also cause a temp rise in vram for the swapping process. These phenomenon is also observed in Bethesda games Fallout4 and Skyrim, that a absurd amount of oversize texture are loaded in by mods, overflow the vram. Or when driving on a highway in GTAV, that the limited cache in vram can't keep up the scene change, that gpu is forced to wait for the memory swap. ie, a open world game with diversitized texture set is more prone to be vram bound. Also a good way to check these thing is with field of view change, greater the FOV, the more space it need for the immediate scene, when frustum culling is applied.
I noticed the 4GB RX 580 was matching the 8GB model in your farcry 5 tests (even at 4K), I think a test of a few more demanding games at 4K/1440p with the two RX 580 cards would be interesting.
@@Time_Traveling_Lesbian It's the same even for games in 2022, and that's because it lacks memory bandwidth to utilize EFFECTIVELY more than XGB of VRAM, now the 680 should have been a 3GB card for sure. The reason GPUs had a higher VRAM models back in the day was Crossfire/SLI, which do NOT stack VRAM, but mirror it, so you'd need double the VRAM to get the effect of a doubled VRAM buffer. The moment they all but dropped multi-GPU support in games and thus SLI/Crossfire support in 2019, they mostly stopped making GPUs with higher and lower amount of VRAM.
I held on to my 780 for a while before upgrading and that thing was still quite powerful. Only problem I had was DX12 titles ran like absolute shit. So long story short now I have a 1080 Ti. :)
DirectX12 runs like crap (compared to Directx11) on most games with all NVIDIA cards. GTX 780 is a bit more powerful than a GTX 680 though. Congratulations on your GTX 1080 ti. That one Beast.
Throwback Thursday BABY! I love this segment, and this topic especially. Being a system builder, with a specialty in gaming, all I ever hear is "get the one with the highest vram version of a card, you'll regret it if you don't"....sigh....this is no news to me or you Steve, but the general masses out there really need to know that the difference is pretty small all told, you certainly won't really notice a thing in real life. Example: I like to use AMD in my customer builds simply because I can get them Freesync monitors and lower overall cost. Of late, of course, getting any AMD card is a nightmare. I have been finding a few RX560's at my local stores/online, though, but the majority of them are the 2GB versions. I have a personal 4GB version on hand and show my customers that the difference in most all titles, especially e-sports, is almost non existent. There will be some more memory usage overall, but I compensate by using faster/overclocked ram, piece of cake really. Anyone who says otherwise is a fanboi/big mouth who really knows nothing at all about how this works. As always Steve, thanks for the great ideas you come up with for videos, especially the Throwback segments, you really deserve over a million plus subscribers!
There isn't a speed difference, yes, but the higher VRAM version of GPUs will age better as games require more and more VRAM to run. For example, the GTX 1060 3GB won't last more than 3 or 4 years as VRAM requirement increases, while the 6GB version will last a lot longer.
I'm suprised the 680 2GB compresses the texture quality to lessen VRAM utilisation. NVIDIA IS CHEATING AGAIN!!!! PITCHFORKS! jks. I'd like to see what happens on AMD hardware because memory and texture compression started to become the norm from the release of the 285 onwards? (Correct me if I'm wrong.)
the 2gb doesnt run as good because it compresses the textures. on a gpu the vram is kind of like a cache that only holds the textures that the card needs right thten to render. the textures you are not looking at are swapped out to system ram. since you are looking at different things all the time the card has to swap the textures back and forth more on a card with less vram. so if you got high speed 3200 ram on a pcie gen 3 or 2 bus then they will swap pretty quicly and it doesnt matter most of the time. now on an older system with 1333 or 1600 ram maybe you only bought one stick of ram so its in single channel and its a second gen i5 which only has pcie gen 2 and maybe for some reason your board is only giving the card 8 lanes (like if you had sli or using the other slot for raid or an ssd) then the swapping takes longer and the game stutters like mad. a graphics card doesnt have enough vram most of the time to hold all the textures. i mean think about it when you download a game the textures are most of the download. how are you goin to fit a 70 gigs of textures into even a 4 or even 8 gig card. vram is just a buffer with the ram holding less. thats why turning down texutres also helps system ram too
maz199 texture compression isn't what's going on here... both and and nvidia use what's called color delta compression which is far more effective and surprisingly how much data color delta uses and how useless it can be frame to frame this why usually compressed.. The reason some games on nvidia hardware more often than AMD look nastier at times is down to either culling engine or the heavy handed use of negitive LOD biasing.. or maybe but unlikely nvidia maybe using some biasing in AF filtering causing certain. Things at oblique angles to look messy if outside of the render queue.. I think in prey example its negitive LOD biasing too much.. it's a slightly cheatsy way nvidia used for ages now to gain a few percent of preformance.. most times the way it handles it you shouldn't really notice too much if at all.. however some games where can't fully hide the fact.. or this could be the culling engine mistaking the texture for being obscured and this not fully rendering it at oblique angles or if its flush with surfaces.. last but least likely theory is that the antisotric filtering is just garbage at times, maybe exaggerated but other things going on tricking it.. Nvidia is known for pulling slightly cheatsy moves.. some of which I don't mind as a concept, like LOD biasing and culling as a concept I'm ok with however sometimes (gameworks titles specially) they over do it so its punishing to AMD.. AMD really needs to start helping build more games like they did with farcry 5 to show that no need to add cheatsy optimizations.. only potential tidally cheatsy thing AMD did with far cry 5 was use compute engine for certain. Things and that's one reason Vega cards is so up in the charts hanging with 1080 and the 1080Ti.. even that isn't too cheatsy since not exclusive and think about how a GPU needs to do half of its work involves complex math thus computing..
bobhumplick I don't have any games that show issues with poor LOD biasing other than new game far cry 5 and it has issues across all GPUs..AMD and nvidia with certain poppin and fill in.. I wonder how much performance loss could be had if we can disable it entirely? I may test it on my 1080Ti tomorrow and see if clamping all my games will knock a few frames off or improve some visuals.. We have to ask our benchmark god at hardware unboxed I guess..?
@meehhhe Of You ive had it set on all my games for a while now and havent really seen any difference i perf. just wondering if that was what the negative lod bias clamp was for. i set everything to quality and go from there
I think i paid like 75$ one year ago for mt 680 4gb phantom , so no, its not overprised. Performance is mostly on par with the 1050. If you dont care about power consumption.
Its actually amazing how the HD79XX and GTX670/680/770 still hold up today. Got one pc with the 670, one with 680, and onewith 970. And theres not one single game so far that does not run on all three computers with some settings adjusted :) Gonna do some FarCry 5 Co-op soon hopefully :)
Why are you surprised. The base consoles are 560 Ti/6950 performance for PS4 and slower for Xbox One. If not VRAM starved, of course the 670 and up are still able to play anything running on these consoles still.
This might be nothing but it might be everything but it's very important. There is this thing i've noticed about Nvidia drivers for quite some time now. Back in the old days when VRAM wasn't enough games would crash either on loading or after a while. You don't see this phenomenon today no matter how high you set resolution and textures etc etc. From personal experience i've noticed that even when i tried to maxed out my old's GTX 660 VRAM, the damn thing would never crash on anything(tried Witcher 3 and Rise of the Tomb Raider and many other games at 4k everything on max), the fps would drop to like 1-4 but the game would never ever crash or completely freeze. There seems to be some kind of optimization workaround which is more likely to be activated by the driver to keep games from crashing and obviously but most importantly keeping it as much as possibly playable. Why do i mention this? Because when you swap GPUs on a system that the hardware like this is so identical it is probably going to keep the optimizations ON since the system doesn't even try to change the driver or you manually not doing a full clean re installation of them and in this case this might be affecting the card that has more memory in a negative way. I have kinda confirmed this, when i got my GTX 1060 6 GB, noticing the same games on the same settings consuming way more VRAM and it made me very skeptical about it. So keep in my mind that things like this could end up screwing any benchmark. Also judging by how Nvidia treats their older hardware and because the 4 GB 680 is more rare than the regular 2GB variant one could assume they might not even bother taking the time to deactivate the optimizations i mentioned to fit the 4 GB variant in order for it to reach its max theoretical performance + the missing textures you mentioned is only making my theory more plausible. I think we need more testing on this though just to be sure.
Using system memory to help GPU is something that existed for a while. So i'm not so sure why it was crashing before. Other than that yeah having more ram always mean more ram will be used. It's the way ram work by nature. It never tell how many ram is needed by the software or the game. The more place is available the more stuff will be kept in the ram just in case. It's meant to reduce loading time and also because deleting stuff from memory take time and resources so it's done only when really required. By example stutter that happen when we do not have enough memory are not caused by loading new stuff ... But it's caused by garbage collection it's kinda more CPU intensive to figure out what can be removed from the memory than loading new stuff. Loading new stuff it's easy because the system know what is required. But when it's time to clean it's not obvious what to keep and what to remove. Removing something important by mistake would cause a load delete loop and it would not be a good scenario. So even if we had 128GB of ram and 32 of GPU ram ... the system would actually fill it over time.
But.. New questions arise.. How would the comparison look with less system memory? And slower memory (like what was available/usual at the time)? What I'm wondering is how well a typical 6 year old computer with the 680 4GB would age compared to the same machine with just a 2GB version. Would it have made sense to go for the 4GB version after all, if one was to keep the machine for 6 years? Also, I'm a little curios about the 0,1% frametimes. Especially with less system memory..
whether its a crime or not, im still rocking it with my gtx680 4gb, and up until now with latest aaa games i can see its age creeping up. Nevertheless has served me well so far for 6 years great vid btw making the gtx 680 great again :D
Good video as always, Steve! :) My opinion has always been "More VRAM = better textures" and not "More VRAM = better performance" as many would like to think. Yes, sure, if the VRAM caps out you will most likely experience stutters and slowdowns. Simple fix is to lower the settings that affects VRAM such as Anti-Aliasing, textures, resolution, draw distance, particles(bandwidth) and so on. But. All the settings will affect the GPU itself more often than not, but things like Anisotropic Filtering and textures affects VRAM more as in capacity and bandwidth than what raw GPU performance could help with. As long people buy graphics cards with GDDR5 or better and stay the hell away from garbage like DDR3/4, then yes, more memory will help achieve better fidelity even though the GPU is low/mid-tier in today's standards. While things such as calculations goes with the amount of stuff rendered on-screen might be suffering when the GPU isn't fast enough, textures and AF will stay sharp as long there's memory capacity (and bandwidth) to spare. Of course, the resolution must be decent to even gain an advantage in fidelity. 1024x768 will just butcher anything in its way for example. ;)
it wont affect fps per se, but you will notice the odd stutter here and there with less vram AND provided your textures are exceeding that vram in the first place
People REALLY need to learn how RAM works: Maxing out the GPUs RAM does not then cap the fps or any crap like that, it means that it then offloads any more storage to system RAM, and this typically doesn't affect performance much unless a significant amount is offloaded.
It seems to depend on the game it possibly the API. Most games can offload over half the memory demand to the system memory and still be fine. Wolf 2 or possibly Vulcan may be the exception. It really hates offloading memory to the system ram.
Problem is the GTX 680 doesn't have the horsepower to make use of the extra memory to it's fullest potential. But the 4GB buffer does give nice headroom to relief stress off the RAM limit.
Why all the medium settings and 70+ FPS? VRAM is more likely to make a difference when the card really has some work to do, especially with the highest texture settings. So yeah, not much of a difference under these conditions, but the test settings were not really aiming at filling the VRAM.
Project Cars 2 was maxed out, Rise of the Tomb Raider used max textures, Assassin's Creed used max textures, Battlefield 1 used maximum texture quality....
I got lucky on release day purchasing a Zotac GTX 680 2 gig GPU. The card holds up very well in gaming. I still replaced it with GTX 1050ti for lower power usage and more V ram space in newer games. I would easily mount this GTX 680 2 gig model to a 720p display and call it a day. Thanks for Retro performance review Steve.
This really makes me want to replace the 128mb vram chips on my GTX 690 with 256mb ones and flash the bios so it thinks it's two 4gb 680s. But this isn't terrible for a 6 year old card surprisingly. I just hope next gen has better availability and prices than the current market. Great work as always!
Im still glad I paid extra and waited for the 4GB FTW 680 when it came out. GTA5 uses 3.5GB of my 4GB. I would like to see GTA5 tested. At 1080p maxed I use 3.5GB vram, 10GB of system ram and 12GB of pagefile and still manage 45+ fps constant.
Love you shutting down the misinformed people !! Great vid as always Steve ♥ ps. Steve or anybody who knows, what's that monitoring software that shows fps: current, avg, 1% low and 0.1%?? I can't get RivaTuner to show that. Is it NZXT CAM? Haven't used that in a while and unsure if they updated it
Knew it!! I still remember when some youtubers were recommending the 270x and gtx 960 4gb versions instead of a 280x because it only had 3gb of Vram (yet it was 30% faster lol)- one of them being Techsource i think? The same guy also said Xeon's are bad for gaming lmao
I still have an old HOF 680 in my media PC that I bought in 2012 runs at 1345mhz it's an absolute beast of a card. The cooler is loud but the frame increase over a stock 680 is insane. It runs 1293mhz out the box and a further OC gets a little more out of the chip. Galaxy (Galax) weren't lying when they said they bin the chips.
appreciate your effort, but please differenciate: as far as i am aware, more vram basically never(depending on engine) affect performance, as missing textures will often not be reloaded but just be dropped, resulting in BETTER fps numbers(less render work, in case of tessalation), but worse image quality. this was correctly stated by you in the last part of the video, but the fps part just dont make any sense to me. aslo, as textures are often backed up in the system ram, more vram usage results in more system ram usage(bigger backup), as the lower vram is noticed and high res textures are often not even loaded into the system ram.
You will get massive stutter. For example, I go from 60 fps Ultra at 4k down to 1 fps when I set AA to 2x just to hit the Vram limit (about 5-10 seconds of input lag, like no joke, hit a button, hold it for a couple seconds, let off, and then wait patently to eventually see a response). This was with R9 Fury on NFSMW, the 2x AA is actually pretty massive impact, but its not a 60x impact. I also own the Titan Xp, and with AA its also unplayable, so you end up running the same settings and you can't tell what card you're running, but the 4gb Fury is hit more due to Vram when using AA.
I remember during the 700 series that the higher VRAM models were only useful if planning for sli in the future. On that note, planning for sli upfront was a fools errand anyway as getting a better GPU was more worthwhile than buying more VRAM. Did rock gtx 760 4gb sli for awhile for 1440p tho, before selling for a single 980.
I had the GTX690 from when it launched up untill quite recently when it unfortunately died. I love this card. And while I wish they made the 3GB per gpu (so 6GB) version if you lowered AA and textures it was still quite capable. For example when I was playing the NFS MW 2012 on triple 1080p with max settings no AA, the game would be locked at 60 all the time with no drops. When I enabled AA the framerate dropped to mid 20s. But hey it kept up even in newer titles at 1080p.
What people often tend to forget these days is that developers no longer develop games for top end hardware or to be played at ultra settings. In matter of fact most games will look decent at medium and going beyond high usually isn't even worth performance impact. Same goes for hardware. They don't develop just for latest and greatest. They want to include as many potential customers in and that means they don't wan't to exclude someone who has graphic card like GTX680. Meaning they will always optimize things to do well, even if they have 2GB of VRAM. While you definitely won't rock ultra, here we come back to what I said earlier, most games are designed to be playable and look decent at medium. And while you will definitely find games like Prey, which will have issues with low VRAM cards. I still feel you are mostly fine with 2GB, if you don't expect too much. Another thing people tend to often forget. Graphic card performance is sum of many things, not just amount of VRAM. I used to have RX480 8GB. And you know what? 8GB didn't do anything for me, because I was 99% of the time limited by what GPU can do and not how much room I have with VRAM. Hell, I even burnt myself by buying cheap 8GB card which was loud instead of better 4GB RX480. But real issue of VRAM quantity fights is really simple. It's oversimplification. People love to have things simple. I mean we saw it before when people took frequency as performance metric for CPUs, despite lower clocked AMD CPUs at the time outperforming higher clocked Intel CPUs. Then came multicore and people were like more cores equals better performance and here tables turned and Intel CPUs with higher IPC beat AMD CPUs with more cores. And same is with graphic cards, VRAM is simple, it is just one number, easy to compare and as wrong as all previously mentioned metrics. That rabbit hole unfortunately is just deeper. And while as mentioned, if it is close enough in price, you definitely should go for more, just more VRAM won't perform miracles either.
Devs don't optimize for mid range PC hardware, they optimize for consoles. You can thank consoles for smooth performance in a number of games, and you can also blame their resonating success for the myriad awful PC ports as well.
I noticed the 4Gb model consistently ran 7-12 degrees higher. How Does temperature affect performance on that generation of cards? Great video Steve! Keep up the great information.
Back when they launched I said it was a waste of money, and I'm happy to see I was right. My argument back then was that the bandwidth of the memory was so poor (as was evident by the massive performance scaling of the 670/680 with memory overclocking) that even if it did dip over 2GB, it would always be limited by the bandwidth. And because of this, due to the fact that a larger memorycapacity means more memory latency, the 2GB was consistently slightly faster.
This is great, absolutely awesome. Though I'd liked to see results from a DDR3 based system: Sandy, Ivy and/or Haswell. Just to see if your findings on your Coffee Lake test bed compares. Thanks for the good content!
back on 2013 i got the 2gb gtx 770 with the reasoning that by the time 4+ will really be needed the card will be at thevery least at the end of it's life .... i still have my gtx 770 but only because of the mining craze the scheduled update for me was last july ...anyway prices setling down but i think to prolong the retirement date till nvidia release the new cards ... i mean few more months will not be the end of the world :P
A video explaining some of the graphical settings in game would be awesome so we could have a better understanding and tweak the settings to get the most out of the graphics card without giving up too much eye candy and fps
Most games now include a VRAM meter so you can monitor how much impact settings have, that said it's mostly textures that impact memory usage the most.
The GTX 680 was first released as a 2Gb card if I'm not mistaken. So I wouldn't be surprised if the GPU architecture wasn't really optimized to take advantage of that extra memory.
Alright. So it might be because you run out of gpu power before the memory issues become visible. There's a framerate limit (determined by the gpu's raw processing power) you can't go past even if you increase the memory capacity and bandwidth. When you run out of Vram the gpu switches to system ram, which is slower, but if the gpu is already running @100% it is less likely that it is wasting cycles by "waiting" after data coming from the memory, so even with the slower system ram you barely notice any difference. While on a faster GPU you could easily see a significant impact as it becomes bandwidth deprived.
The 4GB version of the GTX 680 was meant for SLI to obtain higher resolutions and or higher settings. You should try 2GB GTX 680 SLI vs 4GB GTX 680 SLI at 1440p (2560x1440) resolution. That is where you should see the difference in performance.
Despite the other card have twice as less Vram, you can use the system ram as a buffer, end result as I originally expected all along, the difference was basically nil at best. Try 1440p resolution next time, would love to see the difference then. Question, on your osd, it didn't show the gpu speeds on both gpu's, were they set to identical speeds?
One thing I'm curious about is temperature differences and the lack of core clock monitoring. It seemed like the 4gb model was often running at around 80 when the 2gb model was running at around 70. Is it possible that the higher temperature caused lower gpu boost clock speeds? Also, do you know what caused the difference in temperatures? Thanks for another great video mate, enjoyed seeing this revisited
They are different models mate, as I said the 2GB model would often boost up to 2% higher, obviously because it's running cooler. The 2GB model is from Gainward, 4GB from Palit.
I went looking for videos about the 680 because I’m being given a system for free tomorrow and all I know about it, is the cpu is liquid cooled and the radiator is massive and outside the case, free standing and plugs into the wall, it has a OCZ 650w power supply, 2 sticks of ram with massive heat sinks on them, and an EVGA GTX 680. I don’t know what I’m gonna do with this computer yet, but I’ve gotta fix it first. Sounds like the hard drive died. I’ll find out for sure this weekend. I’m dying to know what CPU is in it, the amount of ram, and if it’s a 2gb or 4gb card. I’m also curious about that massive exterior radiator/pump.
I notice the quality settings are often low or medium on many tests. I thought more memory was needed with higher quality settings, usually? I'm assuming the quality settings were lowered to get playable framerates...but wasn't sure if high/ultra quality settings would have much of an impact. This is from a guy with a 7790 with 1GB...and it still (barely) gets the job done. I don't have any experience, but thought I'd ask. Thanks! Full disclosure, I bounced around watching this, so apologize if I missed the explanation in video.
I see that the later tests have higher quality or higher quality textures. I think that took care of my reservations. Should have known you'd cover all the bases as usual!!! Thanks for the great content and fun watch!
I replaced my 680 2gb a year and a half ago, and up until that point I didn't really have an issue with performance in games, except for one; Lords of the Fallen. Set the quality to high at first to see how it ran, but performance was abysmal. Single digit fps. This was with an overclocked 3570k. I started adjusting graphical settings to see what would help, and after trying pretty much anything, there was only a single setting that brought fps into the playable range: texture quality. If it was set to anything but the bare minimum, the gpu would max out vram and grind to a halt, but as soon as it didn't max out, loads of fps! Lords of the Fallen was (is?) an unoptimised mess so that might be it. Rocking an 8gb 1070 now, hopefully that will last me as long as my 680 did.
Just a small thing I noticed, the Assassin Creed FPS chart of GTX 1060 3GB vs 6GB, the 1060 3G doesnt only has less 3GB of VRAM it only has less Cuda Cores etc
Just 2cards? What outrage! Where is the justice in the world? 😂 Nice content as always. I remember being jealous of someone buying this second hand for 10k INR while this was retailing for 22k INR at the time.
Judging from the title of this vid alone, the answer would be probably 'no'. But it seems the 2GB 680 is holding up pretty well. However, in heavily modded games with special texture packs, you might find that 2GB doesn't cut it. For all the rest of games, you'd probably be fine, except from Prey it seems. Thanks for this video.
I see what you did there. you used the medium preset on the games that mattered to offset the VRAM as to make the 2gb look better. For the love of just use the ultra setting on all the games that why you can see the actual difference of the cards.
Did you miss all the side-by-side testing using the maximum possible texture settings? The benchmark results are from previous videos so I merely updated the graphs with the 4GB model.
*Cutler Cycles* Did you even bother to watch the video or just paused and looked at the graphs? As Steve said: "Project Cars 2 was maxed out, Rise of the Tomb Raider used high textures, Assassin's Creed used high textures, Battlefield 1 used max texture quality...."
Now with the 680 being a cheap second hand card, and ending up in budget builds, would it be possible to retest with only 8gb of ram and the games stored on a mechanical drive? Or maybe even 1 stick of ram? Would love to see these now budget cards working in a similar environment to what they actually will end up in
The only game I ever noticed any performance issue with my old GTX-680 (2GB) was the 2016 version of Doom. Definitely ran better once I upgraded to an Rx-480 (8GB) a couple of years ago...
One suggestion: As people who uses those older cards, or 2 / 4 GB models usually cannot afford too much RAM, specially in these days with high prices, I think it would be nice to test both with only 8GB of RAM, which is the reality for most people (including) me. Because when the framebuffer of the VGA starts to fulfil, it uses the system RAM, but when there is not too much RAM available, these 2GB models uses to stutter like crazy in some titles.
Now that's what I don't understand... one bought a 4GB (once) high-end video card in a hope that investment will become justified one day in the future, yet now you bring them down to the ground in this test by settling just for the "High" or "Medium" texture quality (with the exception of BF1), when a 4GB card can easily manage the full texture quality in most games (for example in RotTR, where the "Very High" Textures are optimized exactly for 4GB cards in 1080p, and that's precisely the setting where the problems start)... A bit not too objective comparison overall, isn't it? However, I like that you at least showed _one_ of the advantages of investing more money in the first place by showing how texture streaming in Prey works in the end of the video. Still, it would be much nicer to have more objective comparisons like that. Especially in a comparison which title says _"More is always better, right?"._
Thanks man. Curious if productivity apps have different results versus games. Agreed that games are high performance, however I assume their code is designed to priority to performance. Productivity apps, i.e. 3D animation, editing, etc, tend to number crunch just to grind through a project. Maybe the extra memory there may make a difference.
Not entirely sure why you would not load the highest quality textures possible throughout? If the mem pool is large enough to handle it that is. Testing at such low settings just seems a tad off.
About Prey: I've been testing it with 1 or 2 HD7850s, each with 2GB vram in CF. I can confirm the game desperately needs vram. In CF vram usage is peaking from the start at 3.9GB+ with medium textures which only drops to 3.8GB with low textures. In single GPU config the vram also hits the ceiling at 2GB, but also continuously loading from the drive the game is on, during gameplay. Sure, this would be different on a single card, but still. There is one other game that does this. Fallout 4. Bethesda anyone?
I think 2gb or vram is enough for all games except for some maybe so the 10603gb will be good for most games in 2018 but if u are into modding like increasing texture quality static mesh overhaul of environment and stuff on games that support it dont bother buying a card under 4gb (nvidia 1060 6gb and above and amd rx470 4gb and above) cos once u r into modding on such 4gb vram will get filled up or use up.
I was gonna go to patreon and show you some love for this video alone. But now I see you didn't show any of my comments any love.. So why should I?.. :P
Why in this benchmark the R9 280X exceeds the 1050 Ti in battlefield 1, but in your other video ruclips.net/video/BfSi-Z8r12M/видео.html the 7970 GHZ which is exactly the same GPU, loses against the 1050 Ti ?. Good video by the way.
Normally games tend to chock. I remember arguing with you about the 780 GTX 3 vs 6GB models. I wonder if it could have anything to do with them being newer games and not being optimized around Kepler arch? Just like how Nvidia Drivers are EOL for kepler for optimal performance? Shame games like Sleeping dogs don't even come close to 4 gig's of vram since it came out when Kepler had really good optimization.
It depends on quite a few things but people often forget we can only see memory allocation, not usage (despite the term 'usage' often used). So unless we can measure the difference (fps performance for example) it's hard to know exactly what's going on.
Yeah that is pretty spot on. I guess it's been a long time since I used a GPU with 2 gig's. Because I always was used to seeing stuttering / hitches ect ect when going over the vram usage.
goes to show what a decent gpu the 1050 ti really is, it's actually quite beastly when you consider it's msrp...which is 50% of what it currently sells for :( (bought mine for msrp, when i sold my rx480 for 200% profit xD) - that being said, i hope the gpu marked relaxes soon a bit so we can all enjoy some midrange gpus for decent prices.. although i actually enjoy those kind of videos :D
Yay!! :D Steve answering the questions no-one is asking! :P Well, except for geeks like us! :P I've been wondering about if the 680 4GB version had enough HP to best the 2GB ever since that test with a 4GB 1050ti. And now I know!! Thanks Steve!! :D
It causes stuttering in some games, was a common issue a few years ago. Was just talking to someone yesterday about it. Games like Just Cause 3 and Arkham Knight were unplayable on his 2GB 770... but fine on my 4GB 760. Maybe take a glance at those two? Because so many people complained about stuttering.
Those games are fine now, maybe there were issues upon release OR you're just assuming the difference in VRAM solved the issue, it really could have been a good many things.
Someone else mentioned speed and quantity of system memory. Are you using it with a newer CPU and 16GB+ of DDR 3200 or something? People would have mostly had 4-8GB DDR3 1600 at the time.
Hardware Unboxed Sorry then. After looking closer I had misinterpreted some of the poorly rendered textures of the 2GB model watching on my phone screen.
![How much RAM do gamers need? 4GB vs. 8GB vs. 16GB vs. 32GB [2017]](http://i.ytimg.com/vi/0t4uLh2ZK04/mqdefault.jpg)
I saw this video in my feed and had to check the date to make sure I hadn't accidentally traveled back in time to 2012.
No you haven't, the main reason for this video is probably that GTX680/770 (same card) can be found for relative cheap and offer decent performance in nowadays crazy GPU price world.
It's 'Throwback Thursday' mate, each week we fire up the time machine for a retro benchmark adventure :D I know I need to get out more.
Hey HU!
For the next 'Throwback Thursday' benchmark, I would like to see an in-depth analysis of the 280x and R9 285 in terms of texture quality. This would be an eye-opener to AMD's improvements with tessellation support and memory compression.
Thanks!
@@stanisawszczypua9076 Exactly what i have done now in 2021, got a Gainward 680 2gb for 60 euros. Working like a charm.... Not being updated anymore in driver terms, but still doing its job
Pro Tip: Those that scream the loudest and say you did it wrong etc tend to be wrong 95 out of 100 times. Glad you did this to quell the misinformation that always wallows around. great video Steve thanks for all your hard work.
Yep!
Empty thesis based on cognitive bias
I find those screamers not even reading descriptions on YT not to mention doing a proper research. Many people these days are just to F'n lazy and yet they demand everything be their way.
*Cries in 1.5 GB GTX580*
Great video as always Steve!
TheNextGenNation
For real?
*Screams in 3.5 GB GTX970*
Laughs in RX460 4GB
*cries in 2gb GTX 560m
Blin cri in GeForce 8400 gs
More is always better, 58% of the time.
Heya, Steve!
I can confirm, I own both an EVGA GTX 680 SC 2GB and a EVGA GTX 680 Classified 4GB. Performance isn't much different, however to play games with DSR and newer games it does make a difference, for example, on COD WWII (at least in the launch version) I was getting really bad stuttering, I had to settle for 720P with my 2GB 680, while my friend (using my 4GB 680) could play it comfortably at 1080p high settings, same could be said when I swapped out my GTX 680 2GB for a Sapphire HD 7950 3GB. Another example was when I was trying to play GTA V at 1440p on my 680 SC, I could play fine but after a while, it would start stuttering very hard, going from 30ish fps to 8 and at some point it would just hard crash I believe.
GTX 680s 4GB come at a premium and they're only worth buying if you're on a tight budget and not upgrading anytime soon, but if they're over 120$, you might as well go for a GTX 1050 Ti as in most cases, it performs better and has better driver support anyway.
Cheers for the video Steve, good on ya.
Yeah COD WWII was a massive memory pig upon release, they have fixed it now. I was having issues with 4GB and even 6GB cards when I first tested it and 16GB's of system memory was a must. Thanks for the feedback though mate, appreciate it.
I had the same issue with my GTX 980 running out of vram in UT4 pre alpha. In the fully meshed maps it often filled the full 4GB and tanked in performance when it did so with massive lag and juddering.
Yeah but 1050tis are still overpriced. I mean 200 dollars for one seriously?
Storm - Dawntion Depends on where you live, I can buy them here for 159-179.
It only affects performance when there's not enough space in vram for "immediate scene", and when what's left for cache in vram is so small that the driver have to drop and reload the resource between ram and vram frequently, which mostly result in stuttering when quickly rotating the view or moving really fast. It also cause a temp rise in vram for the swapping process.
These phenomenon is also observed in Bethesda games Fallout4 and Skyrim, that a absurd amount of oversize texture are loaded in by mods, overflow the vram.
Or when driving on a highway in GTAV, that the limited cache in vram can't keep up the scene change, that gpu is forced to wait for the memory swap.
ie, a open world game with diversitized texture set is more prone to be vram bound.
Also a good way to check these thing is with field of view change, greater the FOV, the more space it need for the immediate scene, when frustum culling is applied.
I noticed the 4GB RX 580 was matching the 8GB model in your farcry 5 tests (even at 4K), I think a test of a few more demanding games at 4K/1440p with the two RX 580 cards would be interesting.
Donovan O'Leary that could just be down to the fact that Far Cry 5 refuses to use over 4GB of memory at any setting and resolution.
@@Time_Traveling_Lesbian It's the same even for games in 2022, and that's because it lacks memory bandwidth to utilize EFFECTIVELY more than XGB of VRAM, now the 680 should have been a 3GB card for sure. The reason GPUs had a higher VRAM models back in the day was Crossfire/SLI, which do NOT stack VRAM, but mirror it, so you'd need double the VRAM to get the effect of a doubled VRAM buffer. The moment they all but dropped multi-GPU support in games and thus SLI/Crossfire support in 2019, they mostly stopped making GPUs with higher and lower amount of VRAM.
Just replaced my 2gb 680 a few months ago, she was a solid card.
Moved to a rx580 as no 3rd party Vega 56 cards showed up.
GTX 680 still kick-ass :D
Well should be comparable with a GTX 1050 ti. So yeah, still extremely powerful card.
I held on to my 780 for a while before upgrading and that thing was still quite powerful. Only problem I had was DX12 titles ran like absolute shit. So long story short now I have a 1080 Ti. :)
DirectX12 runs like crap (compared to Directx11) on most games with all NVIDIA cards.
GTX 780 is a bit more powerful than a GTX 680 though.
Congratulations on your GTX 1080 ti. That one Beast.
Except when comparing to the 7970 which back in the day it was beating!
Had to swap it out for a 1070 cause it sucked so bad in 2016 :D
I think you've already answered this question for us, Steve. More is ALWAYS better. Would I rather have 5 benchmarks or 50?
Michael Gusevsky 50 is no sweat for the King of Benchmarks
Throwback Thursday BABY! I love this segment, and this topic especially. Being a system builder, with a specialty in gaming, all I ever hear is "get the one with the highest vram version of a card, you'll regret it if you don't"....sigh....this is no news to me or you Steve, but the general masses out there really need to know that the difference is pretty small all told, you certainly won't really notice a thing in real life.
Example: I like to use AMD in my customer builds simply because I can get them Freesync monitors and lower overall cost. Of late, of course, getting any AMD card is a nightmare. I have been finding a few RX560's at my local stores/online, though, but the majority of them are the 2GB versions. I have a personal 4GB version on hand and show my customers that the difference in most all titles, especially e-sports, is almost non existent. There will be some more memory usage overall, but I compensate by using faster/overclocked ram, piece of cake really. Anyone who says otherwise is a fanboi/big mouth who really knows nothing at all about how this works. As always Steve, thanks for the great ideas you come up with for videos, especially the Throwback segments, you really deserve over a million plus subscribers!
i just picked up a 6GB 1060 though because nV gimped the 3GB version :-(
There isn't a speed difference, yes, but the higher VRAM version of GPUs will age better as games require more and more VRAM to run. For example, the GTX 1060 3GB won't last more than 3 or 4 years as VRAM requirement increases, while the 6GB version will last a lot longer.
the 560 is so slow I think games that need more than 2 gigs of VRAM would be limited by the graphics core anyway
Benchmark-boss strikes yet again (Y)
I always thought that the one factor that had the most impact when talking about VRAM was texture detail anyway.
Correct, well it's one of the big ones yes.
I'm suprised the 680 2GB compresses the texture quality to lessen VRAM utilisation. NVIDIA IS CHEATING AGAIN!!!! PITCHFORKS! jks.
I'd like to see what happens on AMD hardware because memory and texture compression started to become the norm from the release of the 285 onwards? (Correct me if I'm wrong.)
the 2gb doesnt run as good because it compresses the textures. on a gpu the vram is kind of like a cache that only holds the textures that the card needs right thten to render. the textures you are not looking at are swapped out to system ram. since you are looking at different things all the time the card has to swap the textures back and forth more on a card with less vram. so if you got high speed 3200 ram on a pcie gen 3 or 2 bus then they will swap pretty quicly and it doesnt matter most of the time. now on an older system with 1333 or 1600 ram maybe you only bought one stick of ram so its in single channel and its a second gen i5 which only has pcie gen 2 and maybe for some reason your board is only giving the card 8 lanes (like if you had sli or using the other slot for raid or an ssd) then the swapping takes longer and the game stutters like mad. a graphics card doesnt have enough vram most of the time to hold all the textures. i mean think about it when you download a game the textures are most of the download. how are you goin to fit a 70 gigs of textures into even a 4 or even 8 gig card. vram is just a buffer with the ram holding less. thats why turning down texutres also helps system ram too
maz199 texture compression isn't what's going on here... both and and nvidia use what's called color delta compression which is far more effective and surprisingly how much data color delta uses and how useless it can be frame to frame this why usually compressed..
The reason some games on nvidia hardware more often than AMD look nastier at times is down to either culling engine or the heavy handed use of negitive LOD biasing.. or maybe but unlikely nvidia maybe using some biasing in AF filtering causing certain. Things at oblique angles to look messy if outside of the render queue..
I think in prey example its negitive LOD biasing too much.. it's a slightly cheatsy way nvidia used for ages now to gain a few percent of preformance.. most times the way it handles it you shouldn't really notice too much if at all.. however some games where can't fully hide the fact.. or this could be the culling engine mistaking the texture for being obscured and this not fully rendering it at oblique angles or if its flush with surfaces.. last but least likely theory is that the antisotric filtering is just garbage at times, maybe exaggerated but other things going on tricking it..
Nvidia is known for pulling slightly cheatsy moves.. some of which I don't mind as a concept, like LOD biasing and culling as a concept I'm ok with however sometimes (gameworks titles specially) they over do it so its punishing to AMD..
AMD really needs to start helping build more games like they did with farcry 5 to show that no need to add cheatsy optimizations.. only potential tidally cheatsy thing AMD did with far cry 5 was use compute engine for certain. Things and that's one reason Vega cards is so up in the charts hanging with 1080 and the 1080Ti.. even that isn't too cheatsy since not exclusive and think about how a GPU needs to do half of its work involves complex math thus computing..
@meehhhe Of You does the negative lod clamp stop this on nvidia cards? just curious.
bobhumplick I don't have any games that show issues with poor LOD biasing other than new game far cry 5 and it has issues across all GPUs..AMD and nvidia with certain poppin and fill in.. I wonder how much performance loss could be had if we can disable it entirely? I may test it on my 1080Ti tomorrow and see if clamping all my games will knock a few frames off or improve some visuals..
We have to ask our benchmark god at hardware unboxed I guess..?
@meehhhe Of You ive had it set on all my games for a while now and havent really seen any difference i perf. just wondering if that was what the negative lod bias clamp was for. i set everything to quality and go from there
Either way, they both cost more than they are worth.
I think i paid like 75$ one year ago for mt 680 4gb phantom , so no, its not overprised. Performance is mostly on par with the 1050. If you dont care about power consumption.
Same with me, got a 4gb phantom for 50$. Then again that was a REALLY good deal
Its actually amazing how the HD79XX and GTX670/680/770 still hold up today. Got one pc with the 670, one with 680, and onewith 970. And theres not one single game so far that does not run on all three computers with some settings adjusted :) Gonna do some FarCry 5 Co-op soon hopefully :)
Why are you surprised. The base consoles are 560 Ti/6950 performance for PS4 and slower for Xbox One. If not VRAM starved, of course the 670 and up are still able to play anything running on these consoles still.
This might be nothing but it might be everything but it's very important.
There is this thing i've noticed about Nvidia drivers for quite some time now.
Back in the old days when VRAM wasn't enough games would crash either on loading or after a while. You don't see this phenomenon today no matter how high you set resolution and textures etc etc. From personal experience i've noticed that even when i tried to maxed out my old's GTX 660 VRAM, the damn thing would never crash on anything(tried Witcher 3 and Rise of the Tomb Raider and many other games at 4k everything on max), the fps would drop to like 1-4 but the game would never ever crash or completely freeze. There seems to be some kind of optimization workaround which is more likely to be activated by the driver to keep games from crashing and obviously but most importantly keeping it as much as possibly playable.
Why do i mention this? Because when you swap GPUs on a system that the hardware like this is so identical it is probably going to keep the optimizations ON since the system doesn't even try to change the driver or you manually not doing a full clean re installation of them and in this case this might be affecting the card that has more memory in a negative way. I have kinda confirmed this, when i got my GTX 1060 6 GB, noticing the same games on the same settings consuming way more VRAM and it made me very skeptical about it.
So keep in my mind that things like this could end up screwing any benchmark. Also judging by how Nvidia treats their older hardware and because the 4 GB 680 is more rare than the regular 2GB variant one could assume they might not even bother taking the time to deactivate the optimizations i mentioned to fit the 4 GB variant in order for it to reach its max theoretical performance + the missing textures you mentioned is only making my theory more plausible.
I think we need more testing on this though just to be sure.
Using system memory to help GPU is something that existed for a while. So i'm not so sure why it was crashing before.
Other than that yeah having more ram always mean more ram will be used. It's the way ram work by nature. It never tell how many ram is needed by the software or the game.
The more place is available the more stuff will be kept in the ram just in case. It's meant to reduce loading time and also because deleting stuff from memory take time and resources so it's done only when really required.
By example stutter that happen when we do not have enough memory are not caused by loading new stuff ... But it's caused by garbage collection it's kinda more CPU intensive to figure out what can be removed from the memory than loading new stuff. Loading new stuff it's easy because the system know what is required. But when it's time to clean it's not obvious what to keep and what to remove. Removing something important by mistake would cause a load delete loop and it would not be a good scenario.
So even if we had 128GB of ram and 32 of GPU ram ... the system would actually fill it over time.
But.. New questions arise.. How would the comparison look with less system memory? And slower memory (like what was available/usual at the time)? What I'm wondering is how well a typical 6 year old computer with the 680 4GB would age compared to the same machine with just a 2GB version. Would it have made sense to go for the 4GB version after all, if one was to keep the machine for 6 years? Also, I'm a little curios about the 0,1% frametimes. Especially with less system memory..
Gtx 660 squad here :dd,anyway great vid Steve i love these old gfx comparisons!
whether its a crime or not, im still rocking it with my gtx680 4gb, and up until now with latest aaa games i can see its age creeping up. Nevertheless has served me well so far for 6 years great vid btw making the gtx 680 great again :D
Good video as always, Steve! :)
My opinion has always been "More VRAM = better textures" and not "More VRAM = better performance" as many would like to think.
Yes, sure, if the VRAM caps out you will most likely experience stutters and slowdowns. Simple fix is to lower the settings that affects VRAM such as Anti-Aliasing, textures, resolution, draw distance, particles(bandwidth) and so on.
But.
All the settings will affect the GPU itself more often than not, but things like Anisotropic Filtering and textures affects VRAM more as in capacity and bandwidth than what raw GPU performance could help with.
As long people buy graphics cards with GDDR5 or better and stay the hell away from garbage like DDR3/4, then yes, more memory will help achieve better fidelity even though the GPU is low/mid-tier in today's standards. While things such as calculations goes with the amount of stuff rendered on-screen might be suffering when the GPU isn't fast enough, textures and AF will stay sharp as long there's memory capacity (and bandwidth) to spare.
Of course, the resolution must be decent to even gain an advantage in fidelity. 1024x768 will just butcher anything in its way for example. ;)
it wont affect fps per se, but you will notice the odd stutter here and there with less vram AND provided your textures are exceeding that vram in the first place
Dishonored 2 was barely unplayable on my 2GB card due to stutter caused by a full VRAM.
Thank you for making this video and pointing out the biggest difference (that being an occasional difference in texture quality).
People REALLY need to learn how RAM works:
Maxing out the GPUs RAM does not then cap the fps or any crap like that, it means that it then offloads any more storage to system RAM, and this typically doesn't affect performance much unless a significant amount is offloaded.
Correct.
It seems to depend on the game it possibly the API. Most games can offload over half the memory demand to the system memory and still be fine. Wolf 2 or possibly Vulcan may be the exception. It really hates offloading memory to the system ram.
Problem is the GTX 680 doesn't have the horsepower to make use of the extra memory to it's fullest potential. But the 4GB buffer does give nice headroom to relief stress off the RAM limit.
Why all the medium settings and 70+ FPS? VRAM is more likely to make a difference when the card really has some work to do, especially with the highest texture settings. So yeah, not much of a difference under these conditions, but the test settings were not really aiming at filling the VRAM.
Project Cars 2 was maxed out, Rise of the Tomb Raider used max textures, Assassin's Creed used max textures, Battlefield 1 used maximum texture quality....
Alright, thanks for the clarification.
Great video Geeza! Top notch content 👌👍
I got lucky on release day purchasing a Zotac GTX 680 2 gig GPU. The card holds up very well in gaming. I still replaced it with GTX 1050ti for lower power usage and more V ram space in newer games. I would easily mount this GTX 680 2 gig model to a 720p display and call it a day. Thanks for Retro performance review Steve.
It's not the size that matters mate, it's how you use it.
That's what all the low-end GPUs say.
This really makes me want to replace the 128mb vram chips on my GTX 690 with 256mb ones and flash the bios so it thinks it's two 4gb 680s. But this isn't terrible for a 6 year old card surprisingly. I just hope next gen has better availability and prices than the current market. Great work as always!
Im still glad I paid extra and waited for the 4GB FTW 680 when it came out. GTA5 uses 3.5GB of my 4GB.
I would like to see GTA5 tested. At 1080p maxed I use 3.5GB vram, 10GB of system ram and 12GB of pagefile and still manage 45+ fps constant.
What about 2gb 680 sli vs 4gb 680 sli? possibly testing at higher resolutions
Love you shutting down the misinformed people !! Great vid as always Steve ♥
ps. Steve or anybody who knows, what's that monitoring software that shows fps: current, avg, 1% low and 0.1%?? I can't get RivaTuner to show that. Is it NZXT CAM? Haven't used that in a while and unsure if they updated it
Knew it!! I still remember when some youtubers were recommending the 270x and gtx 960 4gb versions instead of a 280x because it only had 3gb of Vram (yet it was 30% faster lol)- one of them being Techsource i think? The same guy also said Xeon's are bad for gaming
lmao
I still have an old HOF 680 in my media PC that I bought in 2012 runs at 1345mhz it's an absolute beast of a card. The cooler is loud but the frame increase over a stock 680 is insane. It runs 1293mhz out the box and a further OC gets a little more out of the chip. Galaxy (Galax) weren't lying when they said they bin the chips.
As usual, great video. Now pop those x2 680s in sli and run some BMs please... That's what I'm still running on without any problems tbh.
appreciate your effort, but please differenciate:
as far as i am aware, more vram basically never(depending on engine) affect performance, as missing textures will often not be reloaded but just be dropped, resulting in BETTER fps numbers(less render work, in case of tessalation), but worse image quality. this was correctly stated by you in the last part of the video, but the fps part just dont make any sense to me.
aslo, as textures are often backed up in the system ram, more vram usage results in more system ram usage(bigger backup), as the lower vram is noticed and high res textures are often not even loaded into the system ram.
You will get massive stutter. For example, I go from 60 fps Ultra at 4k down to 1 fps when I set AA to 2x just to hit the Vram limit (about 5-10 seconds of input lag, like no joke, hit a button, hold it for a couple seconds, let off, and then wait patently to eventually see a response). This was with R9 Fury on NFSMW, the 2x AA is actually pretty massive impact, but its not a 60x impact. I also own the Titan Xp, and with AA its also unplayable, so you end up running the same settings and you can't tell what card you're running, but the 4gb Fury is hit more due to Vram when using AA.
I remember during the 700 series that the higher VRAM models were only useful if planning for sli in the future. On that note, planning for sli upfront was a fools errand anyway as getting a better GPU was more worthwhile than buying more VRAM. Did rock gtx 760 4gb sli for awhile for 1440p tho, before selling for a single 980.
I had the GTX690 from when it launched up untill quite recently when it unfortunately died. I love this card. And while I wish they made the 3GB per gpu (so 6GB) version if you lowered AA and textures it was still quite capable. For example when I was playing the NFS MW 2012 on triple 1080p with max settings no AA, the game would be locked at 60 all the time with no drops. When I enabled AA the framerate dropped to mid 20s. But hey it kept up even in newer titles at 1080p.
people dismiss the 1050, that little card is a monster.
OctoMan PC's it is working great for me!:)
Thank you,i was curious about this.
What people often tend to forget these days is that developers no longer develop games for top end hardware or to be played at ultra settings. In matter of fact most games will look decent at medium and going beyond high usually isn't even worth performance impact. Same goes for hardware. They don't develop just for latest and greatest. They want to include as many potential customers in and that means they don't wan't to exclude someone who has graphic card like GTX680. Meaning they will always optimize things to do well, even if they have 2GB of VRAM. While you definitely won't rock ultra, here we come back to what I said earlier, most games are designed to be playable and look decent at medium. And while you will definitely find games like Prey, which will have issues with low VRAM cards. I still feel you are mostly fine with 2GB, if you don't expect too much.
Another thing people tend to often forget. Graphic card performance is sum of many things, not just amount of VRAM. I used to have RX480 8GB. And you know what? 8GB didn't do anything for me, because I was 99% of the time limited by what GPU can do and not how much room I have with VRAM. Hell, I even burnt myself by buying cheap 8GB card which was loud instead of better 4GB RX480.
But real issue of VRAM quantity fights is really simple. It's oversimplification. People love to have things simple. I mean we saw it before when people took frequency as performance metric for CPUs, despite lower clocked AMD CPUs at the time outperforming higher clocked Intel CPUs. Then came multicore and people were like more cores equals better performance and here tables turned and Intel CPUs with higher IPC beat AMD CPUs with more cores. And same is with graphic cards, VRAM is simple, it is just one number, easy to compare and as wrong as all previously mentioned metrics. That rabbit hole unfortunately is just deeper. And while as mentioned, if it is close enough in price, you definitely should go for more, just more VRAM won't perform miracles either.
Devs don't optimize for mid range PC hardware, they optimize for consoles. You can thank consoles for smooth performance in a number of games, and you can also blame their resonating success for the myriad awful PC ports as well.
Bought the Gigabyte 4GB GTX680 OC Edition, at 2013. It still goes strong at most of the games combined with an i5 2500.
you test some games on medium or low settings, i did the same and it use 2gb vram , so try games with ultra textures and you will see difference.
I noticed the 4Gb model consistently ran 7-12 degrees higher. How Does temperature affect performance on that generation of cards?
Great video Steve! Keep up the great information.
Doesn't impact performance, neither throttled. They just use different coolers as they are different brands. Thanks BTW :)
Back when they launched I said it was a waste of money, and I'm happy to see I was right. My argument back then was that the bandwidth of the memory was so poor (as was evident by the massive performance scaling of the 670/680 with memory overclocking) that even if it did dip over 2GB, it would always be limited by the bandwidth. And because of this, due to the fact that a larger memorycapacity means more memory latency, the 2GB was consistently slightly faster.
maybe the 2gb was clocking higher due to the lesser heat output of the memory?
This is great, absolutely awesome. Though I'd liked to see results from a DDR3 based system: Sandy, Ivy and/or Haswell. Just to see if your findings on your Coffee Lake test bed compares. Thanks for the good content!
back on 2013 i got the 2gb gtx 770 with the reasoning that by the time 4+ will really be needed the card will be at thevery least at the end of it's life .... i still have my gtx 770 but only because of the mining craze the scheduled update for me was last july ...anyway prices setling down but i think to prolong the retirement date till nvidia release the new cards ... i mean few more months will not be the end of the world :P
A video explaining some of the graphical settings in game would be awesome so we could have a better understanding and tweak the settings to get the most out of the graphics card without giving up too much eye candy and fps
Most games now include a VRAM meter so you can monitor how much impact settings have, that said it's mostly textures that impact memory usage the most.
The GTX 680 was first released as a 2Gb card if I'm not mistaken. So I wouldn't be surprised if the GPU architecture wasn't really optimized to take advantage of that extra memory.
Not how it works mate. It's just designed to use a certain type of memory. It's the games that dictate how much you might need.
Alright. So it might be because you run out of gpu power before the memory issues become visible. There's a framerate limit (determined by the gpu's raw processing power) you can't go past even if you increase the memory capacity and bandwidth. When you run out of Vram the gpu switches to system ram, which is slower, but if the gpu is already running @100% it is less likely that it is wasting cycles by "waiting" after data coming from the memory, so even with the slower system ram you barely notice any difference. While on a faster GPU you could easily see a significant impact as it becomes bandwidth deprived.
This is why the 7950 3gb is still a great choice, it is £70 on ebay in the UK and performs really well.
The 4GB version of the GTX 680 was meant for SLI to obtain higher resolutions and or higher settings. You should try 2GB GTX 680 SLI vs 4GB GTX 680 SLI at 1440p (2560x1440) resolution. That is where you should see the difference in performance.
Despite the other card have twice as less Vram, you can use the system ram as a buffer, end result as I originally expected all along, the difference was basically nil at best.
Try 1440p resolution next time, would love to see the difference then.
Question, on your osd, it didn't show the gpu speeds on both gpu's, were they set to identical speeds?
One thing I'm curious about is temperature differences and the lack of core clock monitoring. It seemed like the 4gb model was often running at around 80 when the 2gb model was running at around 70. Is it possible that the higher temperature caused lower gpu boost clock speeds? Also, do you know what caused the difference in temperatures? Thanks for another great video mate, enjoyed seeing this revisited
They are different models mate, as I said the 2GB model would often boost up to 2% higher, obviously because it's running cooler. The 2GB model is from Gainward, 4GB from Palit.
Hi Steve, I think it's better to benchmark games with mod to show the impact on the VRAM. Most of the issue related to VRAM mostly comes from mods.
What is the configuration of the computer you used with gtx 680 in testing this video?
I went looking for videos about the 680 because I’m being given a system for free tomorrow and all I know about it, is the cpu is liquid cooled and the radiator is massive and outside the case, free standing and plugs into the wall, it has a OCZ 650w power supply, 2 sticks of ram with massive heat sinks on them, and an EVGA GTX 680. I don’t know what I’m gonna do with this computer yet, but I’ve gotta fix it first. Sounds like the hard drive died. I’ll find out for sure this weekend. I’m dying to know what CPU is in it, the amount of ram, and if it’s a 2gb or 4gb card. I’m also curious about that massive exterior radiator/pump.
I notice the quality settings are often low or medium on many tests. I thought more memory was needed with higher quality settings, usually? I'm assuming the quality settings were lowered to get playable framerates...but wasn't sure if high/ultra quality settings would have much of an impact.
This is from a guy with a 7790 with 1GB...and it still (barely) gets the job done. I don't have any experience, but thought I'd ask. Thanks!
Full disclosure, I bounced around watching this, so apologize if I missed the explanation in video.
I see that the later tests have higher quality or higher quality textures. I think that took care of my reservations. Should have known you'd cover all the bases as usual!!! Thanks for the great content and fun watch!
I replaced my 680 2gb a year and a half ago, and up until that point I didn't really have an issue with performance in games, except for one; Lords of the Fallen.
Set the quality to high at first to see how it ran, but performance was abysmal. Single digit fps. This was with an overclocked 3570k.
I started adjusting graphical settings to see what would help, and after trying pretty much anything, there was only a single setting that brought fps into the playable range: texture quality. If it was set to anything but the bare minimum, the gpu would max out vram and grind to a halt, but as soon as it didn't max out, loads of fps!
Lords of the Fallen was (is?) an unoptimised mess so that might be it.
Rocking an 8gb 1070 now, hopefully that will last me as long as my 680 did.
Just a small thing I noticed, the Assassin Creed FPS chart of GTX 1060 3GB vs 6GB, the 1060 3G doesnt only has less 3GB of VRAM it only has less Cuda Cores etc
Just 2cards? What outrage! Where is the justice in the world?
😂
Nice content as always. I remember being jealous of someone buying this second hand for 10k INR while this was retailing for 22k INR at the time.
Gtx 1050 ti is still impressive, I wonder how Nvidia's 11 series equivalent will perform
A 1150 Ti of sorts? Probably GTX 780/OG Titan performance levels.
Judging from the title of this vid alone, the answer would be probably 'no'. But it seems the 2GB 680 is holding up pretty well.
However, in heavily modded games with special texture packs, you might find that 2GB doesn't cut it. For all the rest of games, you'd probably be fine, except from Prey it seems. Thanks for this video.
Perhaps you should try and get your hands on the HD 7970 6GB?
Why did the 4GB model run way hotter than the 2GB model?? Was it the cooler?
I see what you did there. you used the medium preset on the games that mattered to offset the VRAM as to make the 2gb look better. For the love of just use the ultra setting on all the games that why you can see the actual difference of the cards.
Did you miss all the side-by-side testing using the maximum possible texture settings? The benchmark results are from previous videos so I merely updated the graphs with the 4GB model.
Your beef was with 3GB models which almost everyone knows are fine for 1080p gaming.
*Cutler Cycles* Did you even bother to watch the video or just paused and looked at the graphs? As Steve said: "Project Cars 2 was maxed out, Rise of the Tomb Raider used high textures, Assassin's Creed used high textures, Battlefield 1 used max texture quality...."
Still a proud user of ASUS GeForce GTX 680 DirectCU II TOP 2GB ^_^
I still play on my GTX 660Ti, though I feel that they missed the opportunity to call it the GTX 666 being in between a 660 and a 670 you know :D.
So funny
Now with the 680 being a cheap second hand card, and ending up in budget builds, would it be possible to retest with only 8gb of ram and the games stored on a mechanical drive? Or maybe even 1 stick of ram?
Would love to see these now budget cards working in a similar environment to what they actually will end up in
The only game I ever noticed any performance issue with my old GTX-680 (2GB) was the 2016 version of Doom. Definitely ran better once I upgraded to an Rx-480 (8GB) a couple of years ago...
One suggestion: As people who uses those older cards, or 2 / 4 GB models usually cannot afford too much RAM, specially in these days with high prices, I think it would be nice to test both with only 8GB of RAM, which is the reality for most people (including) me. Because when the framebuffer of the VGA starts to fulfil, it uses the system RAM, but when there is not too much RAM available, these 2GB models uses to stutter like crazy in some titles.
You deserve way more subscribers
Now that's what I don't understand... one bought a 4GB (once) high-end video card in a hope that investment will become justified one day in the future, yet now you bring them down to the ground in this test by settling just for the "High" or "Medium" texture quality (with the exception of BF1), when a 4GB card can easily manage the full texture quality in most games (for example in RotTR, where the "Very High" Textures are optimized exactly for 4GB cards in 1080p, and that's precisely the setting where the problems start)...
A bit not too objective comparison overall, isn't it?
However, I like that you at least showed _one_ of the advantages of investing more money in the first place by showing how texture streaming in Prey works in the end of the video.
Still, it would be much nicer to have more objective comparisons like that. Especially in a comparison which title says _"More is always better, right?"._
Thanks man. Curious if productivity apps have different results versus games. Agreed that games are high performance, however I assume their code is designed to priority to performance. Productivity apps, i.e. 3D animation, editing, etc, tend to number crunch just to grind through a project. Maybe the extra memory there may make a difference.
can you also do CPU benchmark from older to new generation. Does it effect gaming performance and is it necessary to keep updating CPU
Is this the same case for the gtx 770 2gb and 4 gb
Not entirely sure why you would not load the highest quality textures possible throughout? If the mem pool is large enough to handle it that is. Testing at such low settings just seems a tad off.
Honestly doesnt vram only matter to a point when it comes to resolution and details used?
Nice content on this channel but please change the outro music
Thanks. We like the outro music though and most viewers seem to as well.
About Prey:
I've been testing it with 1 or 2 HD7850s, each with 2GB vram in CF. I can confirm the game desperately needs vram.
In CF vram usage is peaking from the start at 3.9GB+ with medium textures which only drops to 3.8GB with low textures.
In single GPU config the vram also hits the ceiling at 2GB, but also continuously loading from the drive the game is on, during gameplay.
Sure, this would be different on a single card, but still.
There is one other game that does this. Fallout 4. Bethesda anyone?
I have one problem why is 4gb version about 10 degree hotter than 2gb
I think it's less a case of "won't more memory help" and more "do I have enough GPU power to begin with".
I think 2gb or vram is enough for all games except for some maybe so the 10603gb will be good for most games in 2018 but if u are into modding like increasing texture quality static mesh overhaul of environment and stuff on games that support it dont bother buying a card under 4gb (nvidia 1060 6gb and above and amd rx470 4gb and above) cos once u r into modding on such 4gb vram will get filled up or use up.
Welp, now we gotta await the gtx 770 2gb vs 4gb video now.
the 680 and 770 is the same chip
Many games are still optimized for 2gb VRAM b/c of the number of people with that amount, as seen in Steam user surveys.
Got a gtx 680 2gb a few months ago for 60$, couldn't be happier
680 2 GB performs way fresher than 4 GB. Is that possible that both cards perform equal because the 4 GB thermal throttled ?
I was gonna go to patreon and show you some love for this video alone. But now I see you didn't show any of my comments any love.. So why should I?.. :P
its funny that in 2012 you needed 7970 ghz editon to beat 680
and now even 7950 is faster
Why in this benchmark the R9 280X exceeds the 1050 Ti in battlefield 1, but in your other video ruclips.net/video/BfSi-Z8r12M/видео.html the 7970 GHZ which is exactly the same GPU, loses against the 1050 Ti ?. Good video by the way.
I had a GTX680 for a few years and I never knew there was a 4GB version of the card.
Now i haven't seen the video yet but i am pretty sure the difference is almost nothing between the two. Gonna be exciting to see if i'm right or not.
Normally games tend to chock. I remember arguing with you about the 780 GTX 3 vs 6GB models. I wonder if it could have anything to do with them being newer games and not being optimized around Kepler arch? Just like how Nvidia Drivers are EOL for kepler for optimal performance? Shame games like Sleeping dogs don't even come close to 4 gig's of vram since it came out when Kepler had really good optimization.
It depends on quite a few things but people often forget we can only see memory allocation, not usage (despite the term 'usage' often used). So unless we can measure the difference (fps performance for example) it's hard to know exactly what's going on.
Yeah that is pretty spot on. I guess it's been a long time since I used a GPU with 2 gig's. Because I always was used to seeing stuttering / hitches ect ect when going over the vram usage.
So its good only for SLI exactly like the gtx960 4gb which doesnt make sense for one card also...
What a pleasant surprise!
goes to show what a decent gpu the 1050 ti really is, it's actually quite beastly when you consider it's msrp...which is 50% of what it currently sells for :( (bought mine for msrp, when i sold my rx480 for 200% profit xD) - that being said, i hope the gpu marked relaxes soon a bit so we can all enjoy some midrange gpus for decent prices.. although i actually enjoy those kind of videos :D
Yay!! :D Steve answering the questions no-one is asking! :P Well, except for geeks like us! :P I've been wondering about if the 680 4GB version had enough HP to best the 2GB ever since that test with a 4GB 1050ti. And now I know!! Thanks Steve!! :D
Especially considering the time these cards came out in, it would probably have made sense to test with only 8 GB of system memory.
How about older games like Crysis 3?
It causes stuttering in some games, was a common issue a few years ago. Was just talking to someone yesterday about it. Games like Just Cause 3 and Arkham Knight were unplayable on his 2GB 770... but fine on my 4GB 760. Maybe take a glance at those two? Because so many people complained about stuttering.
Those games are fine now, maybe there were issues upon release OR you're just assuming the difference in VRAM solved the issue, it really could have been a good many things.
Someone else mentioned speed and quantity of system memory. Are you using it with a newer CPU and 16GB+ of DDR 3200 or something? People would have mostly had 4-8GB DDR3 1600 at the time.
All this was discussed in the video mate.
I'll rewatch, it's 5am here and I have insomnia so I must have missed something. Sorry lol
Seems like the captions were reversed for Prey with the missing textures.
No, they aren't.
Hardware Unboxed Sorry then. After looking closer I had misinterpreted some of the poorly rendered textures of the 2GB model watching on my phone screen.