If possible, please try to make more videos like this together in the future, this type of video is very enjoyable to watch. Both informative and entertaining.😃
I logged into Gordon's webcam and it's garbage! Nothing but a bunch of naked women fondling computer components. Not a benchmark in sight. Complete scam. Will pay again.
I think some of the most helpful things I've seen on benchmarks lately from GN is having much older generations on there just as a comparison. A lot of places don't do that and it helps understand the upgrades more if you've waited a few years. Some people only show the current top end cards/processors.
@@TigonIII Last time I checked Steams hardware survey, it looked like people generally have worse hardware (and less storage) than they used to (I'm guessing there's a chance that Steam has had an influx of lower budget gamers).
top ten for gfx cards are rather recent releases, actually: 1 - NVIDIA GeForce GTX 1650 2 - NVIDIA GeForce GTX 1060 3 - NVIDIA GeForce RTX 3060 4 - NVIDIA GeForce RTX 3060 Laptop GPU 5 - NVIDIA GeForce RTX 2060 6 - NVIDIA GeForce GTX 1050 Ti 7 - NVIDIA GeForce RTX 3060 Ti 8 - NVIDIA GeForce RTX 3070 9 - NVIDIA GeForce RTX 3050 10 - NVIDIA GeForce GTX 1660 SUPER (total of these is about 40% of the entire market when combined, the remain 60% is heavily stratified) That said, I think even today including 900-series is warranted, given how amazingly popular the 970 was and how it can be a cultural touchstone - and many people who buy a particular series are not coming from the most previous one.
7:10 Gotta congratulate Gordon for not just saying DLSS outright One of the things I respect hardware reviewers to do is to run no-DLSS tests for the apples to apples comparisons. If at least alongside the DLSS-on tests, which are explicitly marked. GN bears that sort of standard rather solidly.
Yep, always respect GN for doing comparisons without DLSS or AMD FSR. It gives a baseline of raster performance, which is the MAIN goal of benchmarking a GPU against another. Afterward it can be interesting to incorporate some of the extras into a chart, but the general consensus around those technologies is that there's going to be *some* benefit when they're enabled (sadly, normally at the cost of artifacting or blurriness) so those become extremely subjective For example, DLSS 3.0 adds "fake" frames, so the framerate will technically be higher at the cost of worse image quality. If the smearing doesn't bother you, and some game that you may want to play supports it, then it might be for you. In my opinion, the main point of a comparison in a tech review is basically to compare measurable statistics that are as objective as possible. The optional or generally superfluous stuff is personal preference. No one is going to compare Nvidia, AMD, and Intel gpus specifically on the basis of Nvidia's ULMB 2, because there's nothing to compare to from the other vendors to get any sort of result relative to them. However, if you're looking for general performance numbers, it's ideal to compare them based on rasterization--afterword it can become personal preference if you specifically wanna run a game with ULMB 2 or other niche feature. That said, it pays to do research on feature sets based on purpose. Planning around stuff like AV1 for content creation, or avoiding Intel's offerings for a build that is meant to focus on pre-DX11 gaming at high framerates, should be the decision maker. TLDR; A GPU is a tool and it's all about a tool's purpose, which for gaming is most related to a GPU's raster chops. So, Gamers Nexus goes with their name sake and tests for that.
The "problem" with these benchmarks is they're scientific with controls and variables; the benchmarks provide information users can extrapolate for their systems and software. There is nothing wrong with the benchmarks. People are lazy and want to be told what to buy or justify their purchase decisions. You can't help people that don't want to help themselves.
Yup. If someone for whatever reason depends on a very specific use case to make a decision on what to buy, chances are that there're a few forums dedicated to that specific scenario. I mostly play flight and racing sims on VR, a niche within a niche and guess what? There are several sites and forums dedicated to that. Even if one can't extrapolate from the usual sources, the internet is a vast place.
I think Gordon needs to make an OnlyFans account, but instead of the usual stuff that goes on there, he'd just do hardware tests as suggested by paying customers. Not a bad idea at all!
My go to channels for reviews. Props to the others, but some have too little data to be relevant or it isn't details enough and some prioritize things that I personally don't care for. It's also good on GN's side because they test different APIs and even GTA V so it's easy to see how it handles those APIs. For example, I play a lot of DX9 games so that's important for me. I also like that they have started adding older parts so that way it's easier to contextualize the improvements when viewed from the perspective of someone that's running something a few generations old Reviews with fixed memory speeds makes sense because if the viewer knows that one benefits more from higher speeds then the viewer can make the adjustment. Testing in a vacuum and isolating the variables is a great methodology. I also agree on the FSR and DLSS thing about testing with them off. I don't even play games that support those so a review with only FSR and DLSS would be completely useless to me. Benchmarks from Intel, AMD, nvidia etc I don't like because obviously each company would insist on something that they benefit from and it's optimized for their product. That's also why benchmarking several titles is important. The only thing that the benchmarks from companies are worth is checking to see if their claims are true and figuring out what they did to achieve their numbers (which are most probably misleading)
This is why we like PCWorld, even though they are very much mainstream media they are honest. Gordon is there supporting a truly 'independent' journalist. I put it in quotation marks because obviously independent journalism is impossible, at the end of the day somebody pays the bills so it really is about who pays the bill: advertisers, a state or regular people. What we like to call independent journalism (yes, incredibly important!) is basically journalism funded by regular people, directly and not indirectly (advertising or state).
Much of GamersNexus bills are paid through their online store. While sponsored they really push their own products for this very reason... independent journalism
I agree with Steve at ~11:50 on just benchmarking both ways as this adds an interesting element: seeing how big of a difference memory speeds can make on different CPU's. Although this adds work to benchmarkers, it does add value for viewiers. I do think everyone needs to realize that equal doesn't always mean fair, or rather fair doesn't always mean equial. Love seeing Gordon and Steve together. Hope you have more collaboration going forward!
Charge 0.99 a minute for valet benchmarks and a minimum block of 60 minutes...50$ nonrefundable administrative fee...plus taxes. Lower rates for public valet benchmarks (think Groupon/massdrop where the benchmarks are chosen in advance).
It is fair to test ram at the same speed because it should be the same ram regardless. It should be the exact same stick if possible. If it costs more for a faster kit of ram then the total platform cost is higher for that performance to be realized which is absolutely a fair point of comparison.
It's fair in one sense but it isn't in another, the real issue is it isn't realistic, if you buy something like a 13900K you are not going to buy a slower RAM kit to go with it "because it's fair", if you are buying a part like that you are probably going to try and get all the performance you can out of it. If one part can't run the RAM at that speed that's a limitation (in theory) of the part and should be taken into account imo. Also, you might not know this if you've haven't built a PC recently ( I didn't till I built mine a couple of weeks back, AMD BTW so I'm not shilling for Intel...😛) but RAM kits are sold as "XMP optimized" and "EXPO optimized" these days, and while they sometimes work anyway, sometimes they don't and you are obviously going to buy the one for your CPU brand so using the exact same RAM isn't realistic either. As for the cost analysis, that should really just be left to the user, you're going to do it anyway, so it's kinda superfluous.
It used to be quite common that you did log into a webcam as those were cameras connected to the internet. Some where public, but many had (and still have) passwords. Perhaps Steve is confusing webcams to usb cameras, which are confusingly called webcams even though they dont have any web connectitivy themselves but require you to run some app to share the images/video to others over the internet.
My favorite one is "You were so worried about benchmarks that you completely neglected to catch the Voltage issues on the latest X3D chips and failed to do a proper review and be the actual first line of defense for consumers so they are informed. You failed"
I’ll forever love GN for pioneering the mainstay usage of including frame times/ percentile FPS. In terms of gaming, there’s so much that can affect your FPS, that ‘average’ is almost arbitrary. If GN has a video about benchmarking in general, it’s always going to be the best thing you can watch.
justice for gordon! the first web cam was developed so scientists could check if there was coffee in the breakroom before they left their offices at the University of Cambridge, and they did log into the portal to see what the camera saw!
I think because so many reviewers us the same games not alot of engines are getting looked at for performance. When you have multiple games using Unreal or Engines that have historically had biases its hard to get a sense for the majority to see the true capabilities of hardware or even in regards to real world industry software, like CAD/3d Studio Max/Unreal Engine Editor/Unity etc...
I college I saw them "benchmark" an IBM 360 against a Univac 1108. I learned that you can tune each system so that when you run the "benchmark" you can get the results you want. Or you can "interpret" the results the way you want. A standard "benchmark" is not sufficient.
What we need is to include older benchmarks to show performance improvements across generational leaps. Yes, outdated now, but as an example the speed600 was great for its time to show how perf leaped from 286 to pentium... Steve keeping older benchmarks for this comparison is incredibly valuable
Steve would you mind sharing your game list and settings you use as benchmarks? I've been using synthetic benchmarks at work but I too believe that real world programs like games are far better than synthetic benchmarks. I barely play games anymore as I'm spending more time working.
1980s Gordon appears on Computer Chronicles with Gary Kildall, the real deal, among other legends. Fast forward to 2023, Gordon hangs out with NG and Wendall and all them wannabes.
One thing i wasnt seeing from most outlets recently was performance comparisons between different PCIE generations which is a totally valid audience concern with these recent GPU releases and their pcie 4x8 interface while gen 3 is still so prevalent.
I've always thought that when it comes to testing CPUs you should test whatever JEDEC speed they're rated at and whatever the vendor recommends for at minimum the first launch of an architecture. Memory scaling is something that does not get enough coverage these days amongst tech press.
It doesn't fair to turn off Intell quick sync completely. You can test both ways with and without but you can't just ignore it because it works well for many video editors and new users must know it.
Having no idea what "the incident" was, nor even knowing there was a GPU "circle of trust", I find myself respecting Gordon even more than before. Hopefully the story will be shared when the time is appropriate, in the same vein as Tom Snyder saying he'd tell the audience all about -- insert subject here -- during the last show.
Some annoyances with gaming benchmarks is multiple reviewers: 1-all running the same game/application set 2-refusing to run benchmarks in hyper popular games (apex, fortnite, pubg, call of duty, etc etc) 3-showing the new (4000/7000 series) and current generation (3000/6000 series) GPU's for AMD/Nvidia and....1-3 for old generatiom (2000 series)....and...well...nothing older.
There are some useful things to be said for improving benchmarking. Like how for games like Civilization, Stellaris, Total Warhammer 3 campaign, and other grand strategies, end turn timer and load time is a big deal.
Something that I feel like should be mentioned more: there's a LOT of benchmarking channels out there who do exactly the thing that commentors like the ones mentioned by GN want. They specifically test cards using older CPUs, do A/B comparisons with cards, test cards in games that are often overlooked by reviewers, test different technologies like DLSS or resizable bar on/off, and more scenarios I can't even think of. Watching them not only helps smaller creators but will likely give the data people are looking for specifically. Just whatever you do, _please_ don't use userbenchmark and give people the watchtime who deserve it and even go above and beyond sometimes for specific commentors who request it.
ya’ll should do a stock for stock comparison and max tuned comparison for new hardware to show how much headroom each platform has with just a 360aio and the best ram kit for each platform.
somehow, you missed the entire point of this video. Can *you* explain why should they, exactly? What benefit do *you* think that running such a complex benchmark would achieve that is somehow different or better than what GN, for example, does already? What are you actually even talking about, temperature headroom, performance headroom? "Just" a 360 aio? Like, what?? How do you define "the best ram kit" for each platform? Take a step back and realize your ideas of performance are not universal and that these reviewers are trying to provide data that is relevant and useful to the majority of consumers. This kind of cross-sectional analysis and max limit stuff you are talking about the top 1% of the top 1% of all performance metrics possible here. Not realistic or useful. If you want to see that kind of performance you can get those kinds of parts and do it yourself, ya?
@@archaelia The point is if you can extract more value out of old gen parts because they have more headroom, it gives a better idea of how much you will actually gain from upgrading. As for the best ram kits, its pretty straightforward-whats the best frequency each platform is stable with good timings.
Tbh as a programmer a CPU/RAM benchmark run more geared towards compilation performance would be golden. Eg, in the past while evaluating upgrades we just ran them through a frozen state UE4 compile cycle and scaled whichever came out the fastest. Setting that up every time for a one-off upgrade is kinda overkill tho 😅
It should be a new standard when textures don’t load in, they get a 0 for fps, because it’s not actually working. Fps means nothing without the game working, and not having textures load in is defeating the point of fps.
Purpose of benchmarks is to have the same process with only a single variable to demonstrate value. It's for the person to make purchase decisions or evaluate their current parts so they know when to upgrade. GN was the first to clamp down on the fixed values, making the gpu/cpu the only variable. Which changed the way consumers would purchase computer parts. The "value" of a product gets shown "Relatively". "The 3080 was an awesome card because the next step up is 500 more bucks for x% gain. The 3080 was an awesome card because it was same price as a 2080ti from last generation for an immense x% gain." If you did not SEE these statements in charts, or statements, you would have no grounded way to evaluate the actual value of a 3080. It would just exist in a volatile, subjective world. Because we have great sites like GN, we've gained lots of consumer info. Because of the objective values, you can glean insight into how the companies are actually treating their consumers. Because of objective benchmarks, we know exactly what Nvidia thinks of us when they released the 4060ti. In Gordon's defense, I log in to my surveillance webcams. That's prolly where his idea came from lol.
There is no "You're benchmarking is wrong" This is where the viewers of any media need to do just a LITTLE bit of thinking. For instance a channel that benchmarks NVMe drives will show different benchmarks. To me about 1/3 of the benchmarks are meaningless because of what they are SUPPOSED to represent. Trying to say for instance that one drive will be better in light loads when the comparison is gen4 drives don't make sense because LIGHT loads for NVMe when one drive beats another by maybe as much as 30% probably won't show up because in LIGHT loads an NVMe is blazingly fast. You'd have to have one give a TOTAL beat down to the other to notice. On the other hand a test I want to see but most channels won't do it as moving a large percentage of data to and from that drive. To do this you need a source/destination for that data that's faster than the drives you're testing such as an NVMe RAID controller. This will tell me a lot about how the drive is made, caching and how much, etc.......... For people dealing with large files this is important. For CPUs, benchmarking games is a WIDE variation and I'll only use one for instance. Because AMD makes X3D parts with a very large amount of L3 cache, you have to REALLY THINK about what happens during gameplay, and does a certain type of benchmark really show what happens during gameplay AND show where the extra L3 is giving benefit. I can point to Steve from Hardware Unboxed and say confidently, oftentimes that's a hard NO. For instance when benchmarking SOTTR, he runs through a town area and says this area of the game is more demanding on the CPU which is why he uses it. OK that may be but running through a town area is not typical game play. For testing Vcache parts this REALLY MATTERS. You have to understand the way cache works. Running through a town and not interacting with game assets/characters doesn't imitate real game play. For cache, a core pulls data first from L1, if not there then L2, if not there then L3, if not there then memory, DRAM. If it needs data/instructions that has to go to L1 for the core to use it. L1 runs at the speed of the cores. Instructions get pulled from L1 to get loaded into a pipeline of instructions, that instruction logic uses to do operations. L1 is small, data/instructions will be pushed out pretty quickly unless you're running a small algorithm over and over again, but that's not gameplay. Data/instructions that gets kicked out of L1, will be moved to L2 or there may be a copy of it already in L2. L2 can also only hold a limited amount of data/instructions so the core would also be pushing stuff out of L2 and putting it in L3, unless it's already in L3. In other words, you get a trickle down of data/instructions to L3, and if L3 is already full then stuff has to be kicked out of L3. During gameplay the benefit of L3 is when you STAY in an area to interact with assets in that area or you're fighting characters, or there's simply a VERY large amount of assets/characters in the area. The CPU needs to know about the objects that your character can interact with along with a lot of other data about where you are, movement, weapons, etc..... When you simply run through an area you aren't really using the data that gets pushed down to L3 very much because you keep loading in more game assets in your CURRENT area. You get the biggest benefit from being in a specifc area where there's lots of game assets/characters that you're interacting with, which is to say the cores need access to a very large amount of data, at the same time, and you need to keep reusing that data. So I could say Steve from HDWU doesn't benchmark right and have some justificaction when he's dealing with open world games where you would be interacting with game assets much more so than what he does, which is none at all. If I want numbers that are more representative of real game play I need to watch videos that show real game play and there are plenty of channels that do this, but they aren't the main channels that do product reviews, etc.......... These are the channels that do CPU or GPU comparisons like an XXXX vs. XXXX video. But some of these channels don't show game play either. For the ones that do they pick an area to deal with the game assets in that location and that's going to give me an idea of how much Vcache is benefiting game play since I keep needing to access the same data and that's EXACTLY what cache is for, or at least L2 and L3. On the other hand there are plenty of games where the data load for the CPU is less. For me to make a blanket statement that you ONLY represent real game play as far as how fps is going is when you are interacting with game assets would be wrong. In some games the variation in FPS may be about the same whether I run through an area or interact with the game assets. It took me a lot just to describe testing in games especially in relation to CPUs that have a very large L3 cache. At the beginning of this comment I said people have to think about what it is they're seeing with benchmarks and does it really represent what THEY do. Without thinking about such things you're watching a video that is meaningless to YOU personally, other than maybe beating your AMD or Intel chest because you're a fanboy. For me personally I know I benefit the most from CPUs that use less power and right now in game loads which affects me, something like the 7800X3D is DESTROYING anything that competes with it for power consumption and that matters to me. I rarely need to watch a review because I can go to Techpowerup's website and look at what they show which gives me more detail with issues like power consumption across multiple workloads. It's a rare review that gives more than one power stat, and it's usually in a way I don't use my PC. YOU need to learn to think for yourself and understand what you do and how some benchmark you see means ANYTHING in relation to what you do. If a channel is continually showing benchmarks for video editing/rendering tasks it doesn't mean anything to me because I don't edit videos. But reading the comments sections you'd think everyone IS editing videos. I doubt it's more than 5% of PC users.
While it sounds very nice to pick games in different genres to give an idea how a GPU or CPU performs in different situations, the fact is there can be huge variations _within_ the same genre (i.e., one FPS game can perform a lot better on Nvidia GPUs while another performs better on AMD GPUs). So, when you test game X, you're effectively testing game X, not "games of the same type as game X". I'd say it's a lot more important to test games that use different _engines_ (which Steve also mentioned) than games of different genres (that might use the same engine). Although the best games will tend to have loads of custom optimisations, so even then you're often still "testing game X" more than you are "testing engine Y".
calling the design flaw in the 12vhp cable user error, that's why people say steve has sold out these days. looks like selling out to me, if it's not designed to stay plugged in that is not user error.
Back in the Adobe CS5 days intel had a quick sync plug-in, for years adobe did not use quick sync by default. Was a good benchmark for intel but most people did not instal the plug-in, then when Adobe added in the plug-in to the program by default they advertised it as 'look how much faster the new version of PP is'.. Adobe being adobe Talking about the new Nvidia GPU's Pugetsystems benchmarks show regression in pro apps with the 4060 TI and 4070 etc. PS Nice to know Gordon will set up 'onlyBenchmarks', a new way for people to spend there money.
I am baised and happy to admit it. I am biased towards whichever vendor will give me the best products at a reasonable price with the most useful features.
i don't personally see a problem with a Chart having Asterisks. i think it's suitable to represent Hardware based on what it can generally do, not like pushing the edge, but also not running a configuration that it could easily run faster than just because reasons. development of Hardware is in more than just one thing. if say, a later Generation of part has a better Memory Controller or I/O interconnect or whatever, those too are improvements/innovations that can benefit the performance of the system. so what if something older could not, that's part of what you're getting from upgrading / Buying that newer thing vs something else in the past. that's a part of it. if CPU A would struggle to run above like 2666 or 5600 or whatever, that's what that Hardware is capable of. if something else several Years later is capable of like 3800 or 7000 or whatever, **that's a part of what the Product improved during the Years between**. however, if separate data is made to also include those "this can generally do this config" as well as some fixed settings irregardless of what the Hardware is capable of - then, that is okay since both are being presented.
One of the things that I would hope more reviewers did, was to turn certain settings on the game off that are known to; A) Add nothing to the game. B) Have massive effect on performance. I know this is something that you generally shouldn't do to review how a GPU performs. But it is something to consider if you're looking at how the game performs in a real-life scenario. In the case of Cyberpunk as an example, Screen Space Reflections Quality have a massive impact on the performance and overall LOOK of the game. In that it's better to turn it off for gameplay. And it removes the pixelization on surfaces. Monster Hunter World for example had "Volume Rendering Quality" which literally boosted FPS in the game by 10-15fps if you turned it off. And it made the game look better by removing alot of the volumetric fog the game had. TLDR; It would be great if there was a "custom settings" portion for couple of the games that people really like to tinker with.
"why don't you test everything?" ...looks at day before shipping on parts to test with day before that embargos. Seriously I would LOVE professional app benchmarking but 90% of channels focus on gaming because of viewership. These small/tight review windows have to stop. I'd also ask for more "why did x company impliment this feature" interviews along with reviews for a more comprehensive "feel" of how successful was the part at accomplishing what the company set out to do. Example you did a whole motherboard lineup without thunderbolt, why? May be some good learning to be had there.
On the memory clocks I used to think if a CPU has better IMC, then it merits using faster ram, however IMC quality is a variable, intel e.g. still only support 3200mhz on DDR4, and two out of three intel CPU's I was not able to get my b die to run stable at over 3200mhz. Those two reasons alone in my opinion validate the modest ram clock benchmarking ion my view. DDR5, I dont know so much about, but I think applying the same logic there is reasonable, even though Intel seem to be ahead of the game on DDR5, I bet there is poor binned chips that struggle to get above AMD memory speeds out there.
Unfortunately while GN does a pretty good job of catching driver updates and retesting as necessary, sometimes gn stats suffer from their style becuase older cards/chips that may have gotten perf boost thru drivers isn't accurately shown in data.
Not sure I follow. Generally when they perform a driver update they retest all their previous cards that they usually include on roundups with the new driver. Only when they specify otherwise is that not the case.
Observation: For some bizarre reason, Gordon rubs me the wrong way when he's solo, but get him on camera with Steve, and he's god mode. How TF does that even work?
An aggressive Gordon is best handled with a mild dose of internet! And log into his webcam!
If possible, please try to make more videos like this together in the future, this type of video is very enjoyable to watch. Both informative and entertaining.😃
I logged into Gordon's webcam and it's garbage! Nothing but a bunch of naked women fondling computer components. Not a benchmark in sight. Complete scam. Will pay again.
I'd be furious too if i'd have a webcam that takes coins! :)
Great video once again!
Given how many hardware companies want to transition to hardware-as-a-service, logging into your webcam could soon become an actual thing.
Will he be oiled up in a thong and high heels during the testing?
Gordon needs to start every video with "Welcome to my webcam" now.
100%
-Adam
Steve and Gordon is always a treat to watch. And Gordon can be aggressive in an open area.
Passion is great. It helps inspire. And after some decades, Gordon is only more immersed.
Gordon can have a little agression as a treat.
He prefers to be aggressive at restaurants!!
"You are already losing, man..." 😂
I think some of the most helpful things I've seen on benchmarks lately from GN is having much older generations on there just as a comparison. A lot of places don't do that and it helps understand the upgrades more if you've waited a few years. Some people only show the current top end cards/processors.
Especially when the Steam Hardware Survey says that many people are still gaming on GTX 1060 and GTX 1650.
I like comparisons with old tech as well.
@@TigonIII Last time I checked Steams hardware survey, it looked like people generally have worse hardware (and less storage) than they used to (I'm guessing there's a chance that Steam has had an influx of lower budget gamers).
top ten for gfx cards are rather recent releases, actually:
1 - NVIDIA GeForce GTX 1650
2 - NVIDIA GeForce GTX 1060
3 - NVIDIA GeForce RTX 3060
4 - NVIDIA GeForce RTX 3060 Laptop GPU
5 - NVIDIA GeForce RTX 2060
6 - NVIDIA GeForce GTX 1050 Ti
7 - NVIDIA GeForce RTX 3060 Ti
8 - NVIDIA GeForce RTX 3070
9 - NVIDIA GeForce RTX 3050
10 - NVIDIA GeForce GTX 1660 SUPER
(total of these is about 40% of the entire market when combined, the remain 60% is heavily stratified)
That said, I think even today including 900-series is warranted, given how amazingly popular the 970 was and how it can be a cultural touchstone - and many people who buy a particular series are not coming from the most previous one.
Notice Steve does these things with Gordon in large open spaces so he has plenty options for Escape Routes.
7:10 Gotta congratulate Gordon for not just saying DLSS outright
One of the things I respect hardware reviewers to do is to run no-DLSS tests for the apples to apples comparisons. If at least alongside the DLSS-on tests, which are explicitly marked. GN bears that sort of standard rather solidly.
Yep, always respect GN for doing comparisons without DLSS or AMD FSR. It gives a baseline of raster performance, which is the MAIN goal of benchmarking a GPU against another. Afterward it can be interesting to incorporate some of the extras into a chart, but the general consensus around those technologies is that there's going to be *some* benefit when they're enabled (sadly, normally at the cost of artifacting or blurriness) so those become extremely subjective For example, DLSS 3.0 adds "fake" frames, so the framerate will technically be higher at the cost of worse image quality. If the smearing doesn't bother you, and some game that you may want to play supports it, then it might be for you.
In my opinion, the main point of a comparison in a tech review is basically to compare measurable statistics that are as objective as possible. The optional or generally superfluous stuff is personal preference. No one is going to compare Nvidia, AMD, and Intel gpus specifically on the basis of Nvidia's ULMB 2, because there's nothing to compare to from the other vendors to get any sort of result relative to them. However, if you're looking for general performance numbers, it's ideal to compare them based on rasterization--afterword it can become personal preference if you specifically wanna run a game with ULMB 2 or other niche feature. That said, it pays to do research on feature sets based on purpose. Planning around stuff like AV1 for content creation, or avoiding Intel's offerings for a build that is meant to focus on pre-DX11 gaming at high framerates, should be the decision maker.
TLDR; A GPU is a tool and it's all about a tool's purpose, which for gaming is most related to a GPU's raster chops. So, Gamers Nexus goes with their name sake and tests for that.
The "problem" with these benchmarks is they're scientific with controls and variables; the benchmarks provide information users can extrapolate for their systems and software. There is nothing wrong with the benchmarks. People are lazy and want to be told what to buy or justify their purchase decisions. You can't help people that don't want to help themselves.
you give them something for free and they still complain
@@GewelReal
You just can't win
Yup. If someone for whatever reason depends on a very specific use case to make a decision on what to buy, chances are that there're a few forums dedicated to that specific scenario. I mostly play flight and racing sims on VR, a niche within a niche and guess what? There are several sites and forums dedicated to that. Even if one can't extrapolate from the usual sources, the internet is a vast place.
this is why toms hardware cpu and gpu hierarchies are so important to me
I think Gordon needs to make an OnlyFans account, but instead of the usual stuff that goes on there, he'd just do hardware tests as suggested by paying customers. Not a bad idea at all!
@OnlyBench
Always a joy seeing Gordon and Steve together on screen! Gordons "intermetiness" really balances out Steves professionalism.
I love how you guys interact. Please make more of such videos!
G'day Gordon & Steve,
Always love your get togethers as they are both insightful & so funny
My go to channels for reviews. Props to the others, but some have too little data to be relevant or it isn't details enough and some prioritize things that I personally don't care for.
It's also good on GN's side because they test different APIs and even GTA V so it's easy to see how it handles those APIs. For example, I play a lot of DX9 games so that's important for me. I also like that they have started adding older parts so that way it's easier to contextualize the improvements when viewed from the perspective of someone that's running something a few generations old
Reviews with fixed memory speeds makes sense because if the viewer knows that one benefits more from higher speeds then the viewer can make the adjustment. Testing in a vacuum and isolating the variables is a great methodology. I also agree on the FSR and DLSS thing about testing with them off. I don't even play games that support those so a review with only FSR and DLSS would be completely useless to me.
Benchmarks from Intel, AMD, nvidia etc I don't like because obviously each company would insist on something that they benefit from and it's optimized for their product. That's also why benchmarking several titles is important. The only thing that the benchmarks from companies are worth is checking to see if their claims are true and figuring out what they did to achieve their numbers (which are most probably misleading)
This is why we like PCWorld, even though they are very much mainstream media they are honest. Gordon is there supporting a truly 'independent' journalist. I put it in quotation marks because obviously independent journalism is impossible, at the end of the day somebody pays the bills so it really is about who pays the bill: advertisers, a state or regular people. What we like to call independent journalism (yes, incredibly important!) is basically journalism funded by regular people, directly and not indirectly (advertising or state).
Much of GamersNexus bills are paid through their online store. While sponsored they really push their own products for this very reason... independent journalism
Has there ever been a collab with Steve and Gordon that wasn't fucking awesome!?
Such great chemistry.
I agree with Steve at ~11:50 on just benchmarking both ways as this adds an interesting element: seeing how big of a difference memory speeds can make on different CPU's. Although this adds work to benchmarkers, it does add value for viewiers. I do think everyone needs to realize that equal doesn't always mean fair, or rather fair doesn't always mean equial.
Love seeing Gordon and Steve together. Hope you have more collaboration going forward!
Charge 0.99 a minute for valet benchmarks and a minimum block of 60 minutes...50$ nonrefundable administrative fee...plus taxes.
Lower rates for public valet benchmarks (think Groupon/massdrop where the benchmarks are chosen in advance).
It is fair to test ram at the same speed because it should be the same ram regardless. It should be the exact same stick if possible. If it costs more for a faster kit of ram then the total platform cost is higher for that performance to be realized which is absolutely a fair point of comparison.
It's fair in one sense but it isn't in another, the real issue is it isn't realistic, if you buy something like a 13900K you are not going to buy a slower RAM kit to go with it "because it's fair", if you are buying a part like that you are probably going to try and get all the performance you can out of it. If one part can't run the RAM at that speed that's a limitation (in theory) of the part and should be taken into account imo.
Also, you might not know this if you've haven't built a PC recently ( I didn't till I built mine a couple of weeks back, AMD BTW so I'm not shilling for Intel...😛) but RAM kits are sold as "XMP optimized" and "EXPO optimized" these days, and while they sometimes work anyway, sometimes they don't and you are obviously going to buy the one for your CPU brand so using the exact same RAM isn't realistic either. As for the cost analysis, that should really just be left to the user, you're going to do it anyway, so it's kinda superfluous.
The intro alone was gold.
It used to be quite common that you did log into a webcam as those were cameras connected to the internet. Some where public, but many had (and still have) passwords. Perhaps Steve is confusing webcams to usb cameras, which are confusingly called webcams even though they dont have any web connectitivy themselves but require you to run some app to share the images/video to others over the internet.
'gordon, we're in a restaurant' was an all timer from steve lol
Gordon is onto something. Wont be long before we get webcam dlc
My favorite one is "You were so worried about benchmarks that you completely neglected to catch the Voltage issues on the latest X3D chips and failed to do a proper review and be the actual first line of defense for consumers so they are informed. You failed"
I’ll forever love GN for pioneering the mainstay usage of including frame times/ percentile FPS. In terms of gaming, there’s so much that can affect your FPS, that ‘average’ is almost arbitrary. If GN has a video about benchmarking in general, it’s always going to be the best thing you can watch.
justice for gordon! the first web cam was developed so scientists could check if there was coffee in the breakroom before they left their offices at the University of Cambridge, and they did log into the portal to see what the camera saw!
You guys are one of the best PC/tech channels around. Always happy to see Gordon and Steven do their thing. Live free everyon
That Cutress video sounds interesting, but I can only find the RISC-V one.
Adam pleading with these folks to get back on topic is hilarious
Now i want to see webcam benchmarks in order to get the most out of my minutes of 9.99 i am going to spend.
Love the chemistry between these two legends 😅 Always entertaining and educational
I think because so many reviewers us the same games not alot of engines are getting looked at for performance. When you have multiple games using Unreal or Engines that have historically had biases its hard to get a sense for the majority to see the true capabilities of hardware or even in regards to real world industry software, like CAD/3d Studio Max/Unreal Engine Editor/Unity etc...
Gotta love the sponsor being sff power supply after his coupe remarks.
I'm glad you get the reference :P
-Adam
The out-takes at the start was good lol
damn i cant wait to log into gordons webcam to benchmark
these really are the moments that make computex and other event like it truly shine its not the cool new hardware
I college I saw them "benchmark" an IBM 360 against a Univac 1108. I learned that you can tune each system so that when you run the "benchmark" you can get the results you want. Or you can "interpret" the results the way you want. A standard "benchmark" is not sufficient.
What we need is to include older benchmarks to show performance improvements across generational leaps.
Yes, outdated now, but as an example the speed600 was great for its time to show how perf leaped from 286 to pentium...
Steve keeping older benchmarks for this comparison is incredibly valuable
Steve would you mind sharing your game list and settings you use as benchmarks? I've been using synthetic benchmarks at work but I too believe that real world programs like games are far better than synthetic benchmarks. I barely play games anymore as I'm spending more time working.
1980s Gordon appears on Computer Chronicles with Gary Kildall, the real deal, among other legends.
Fast forward to 2023, Gordon hangs out with NG and Wendall and all them wannabes.
One thing i wasnt seeing from most outlets recently was performance comparisons between different PCIE generations which is a totally valid audience concern with these recent GPU releases and their pcie 4x8 interface while gen 3 is still so prevalent.
I've always thought that when it comes to testing CPUs you should test whatever JEDEC speed they're rated at and whatever the vendor recommends for at minimum the first launch of an architecture. Memory scaling is something that does not get enough coverage these days amongst tech press.
It doesn't fair to turn off Intell quick sync completely. You can test both ways with and without but you can't just ignore it because it works well for many video editors and new users must know it.
Love the collaborations, thank you! Can you elaborate on this "incident" or was Steve just busting Gordon's chops? Kinda funny either way. Cheers!
the day we found out Gordon is straight up hacking his way into webcams. Intrusive Gordon. Investigative journalism.
Having no idea what "the incident" was, nor even knowing there was a GPU "circle of trust", I find myself respecting Gordon even more than before. Hopefully the story will be shared when the time is appropriate, in the same vein as Tom Snyder saying he'd tell the audience all about -- insert subject here -- during the last show.
Tech Buddah and Tech Jesus is always the best combo.
Gordon is a national treasure and must be protected at all costs.
Some annoyances with gaming benchmarks is multiple reviewers:
1-all running the same game/application set
2-refusing to run benchmarks in hyper popular games (apex, fortnite, pubg, call of duty, etc etc)
3-showing the new (4000/7000 series) and current generation (3000/6000 series) GPU's for AMD/Nvidia and....1-3 for old generatiom (2000 series)....and...well...nothing older.
There are some useful things to be said for improving benchmarking. Like how for games like Civilization, Stellaris, Total Warhammer 3 campaign, and other grand strategies, end turn timer and load time is a big deal.
I bet logging in to that webcam is very very spicy
Something that I feel like should be mentioned more: there's a LOT of benchmarking channels out there who do exactly the thing that commentors like the ones mentioned by GN want. They specifically test cards using older CPUs, do A/B comparisons with cards, test cards in games that are often overlooked by reviewers, test different technologies like DLSS or resizable bar on/off, and more scenarios I can't even think of. Watching them not only helps smaller creators but will likely give the data people are looking for specifically. Just whatever you do, _please_ don't use userbenchmark and give people the watchtime who deserve it and even go above and beyond sometimes for specific commentors who request it.
I'd love to be an observer if Steve and Gordon went on a night out. 😅
Dang it, I need to put more effort into laundry. Because my GamersNexus PC components shortage shirt doesn't look nearly that vibrant anymore.
those opening subtitles lol
Gordon will always hold a special place in my heart for being the first person to sign my Angelfire guestbook
Is this true??
-Adam
.... he's that old...?
Absolutely love seing you 2 together, it always get a lil bit wild.
ya’ll should do a stock for stock comparison and max tuned comparison for new hardware to show how much headroom each platform has with just a 360aio and the best ram kit for each platform.
somehow, you missed the entire point of this video. Can *you* explain why should they, exactly? What benefit do *you* think that running such a complex benchmark would achieve that is somehow different or better than what GN, for example, does already? What are you actually even talking about, temperature headroom, performance headroom? "Just" a 360 aio? Like, what?? How do you define "the best ram kit" for each platform? Take a step back and realize your ideas of performance are not universal and that these reviewers are trying to provide data that is relevant and useful to the majority of consumers. This kind of cross-sectional analysis and max limit stuff you are talking about the top 1% of the top 1% of all performance metrics possible here. Not realistic or useful. If you want to see that kind of performance you can get those kinds of parts and do it yourself, ya?
@@archaelia The point is if you can extract more value out of old gen parts because they have more headroom, it gives a better idea of how much you will actually gain from upgrading. As for the best ram kits, its pretty straightforward-whats the best frequency each platform is stable with good timings.
Tbh as a programmer a CPU/RAM benchmark run more geared towards compilation performance would be golden. Eg, in the past while evaluating upgrades we just ran them through a frozen state UE4 compile cycle and scaled whichever came out the fastest. Setting that up every time for a one-off upgrade is kinda overkill tho 😅
It should be a new standard when textures don’t load in, they get a 0 for fps, because it’s not actually working. Fps means nothing without the game working, and not having textures load in is defeating the point of fps.
Purpose of benchmarks is to have the same process with only a single variable to demonstrate value. It's for the person to make purchase decisions or evaluate their current parts so they know when to upgrade. GN was the first to clamp down on the fixed values, making the gpu/cpu the only variable. Which changed the way consumers would purchase computer parts. The "value" of a product gets shown "Relatively". "The 3080 was an awesome card because the next step up is 500 more bucks for x% gain. The 3080 was an awesome card because it was same price as a 2080ti from last generation for an immense x% gain." If you did not SEE these statements in charts, or statements, you would have no grounded way to evaluate the actual value of a 3080. It would just exist in a volatile, subjective world.
Because we have great sites like GN, we've gained lots of consumer info. Because of the objective values, you can glean insight into how the companies are actually treating their consumers. Because of objective benchmarks, we know exactly what Nvidia thinks of us when they released the 4060ti.
In Gordon's defense, I log in to my surveillance webcams. That's prolly where his idea came from lol.
brb just gonna log in to my webcam real quick 😂
Steve is going to have to bring a shirt next time to pay Gordon.
There is no "You're benchmarking is wrong"
This is where the viewers of any media need to do just a LITTLE bit of thinking.
For instance a channel that benchmarks NVMe drives will show different benchmarks. To me about 1/3 of the benchmarks are meaningless because of what they are SUPPOSED to represent. Trying to say for instance that one drive will be better in light loads when the comparison is gen4 drives don't make sense because LIGHT loads for NVMe when one drive beats another by maybe as much as 30% probably won't show up because in LIGHT loads an NVMe is blazingly fast. You'd have to have one give a TOTAL beat down to the other to notice.
On the other hand a test I want to see but most channels won't do it as moving a large percentage of data to and from that drive. To do this you need a source/destination for that data that's faster than the drives you're testing such as an NVMe RAID controller. This will tell me a lot about how the drive is made, caching and how much, etc.......... For people dealing with large files this is important.
For CPUs, benchmarking games is a WIDE variation and I'll only use one for instance. Because AMD makes X3D parts with a very large amount of L3 cache, you have to REALLY THINK about what happens during gameplay, and does a certain type of benchmark really show what happens during gameplay AND show where the extra L3 is giving benefit. I can point to Steve from Hardware Unboxed and say confidently, oftentimes that's a hard NO.
For instance when benchmarking SOTTR, he runs through a town area and says this area of the game is more demanding on the CPU which is why he uses it. OK that may be but running through a town area is not typical game play. For testing Vcache parts this REALLY MATTERS. You have to understand the way cache works. Running through a town and not interacting with game assets/characters doesn't imitate real game play. For cache, a core pulls data first from L1, if not there then L2, if not there then L3, if not there then memory, DRAM. If it needs data/instructions that has to go to L1 for the core to use it. L1 runs at the speed of the cores. Instructions get pulled from L1 to get loaded into a pipeline of instructions, that instruction logic uses to do operations. L1 is small, data/instructions will be pushed out pretty quickly unless you're running a small algorithm over and over again, but that's not gameplay. Data/instructions that gets kicked out of L1, will be moved to L2 or there may be a copy of it already in L2. L2 can also only hold a limited amount of data/instructions so the core would also be pushing stuff out of L2 and putting it in L3, unless it's already in L3. In other words, you get a trickle down of data/instructions to L3, and if L3 is already full then stuff has to be kicked out of L3.
During gameplay the benefit of L3 is when you STAY in an area to interact with assets in that area or you're fighting characters, or there's simply a VERY large amount of assets/characters in the area. The CPU needs to know about the objects that your character can interact with along with a lot of other data about where you are, movement, weapons, etc..... When you simply run through an area you aren't really using the data that gets pushed down to L3 very much because you keep loading in more game assets in your CURRENT area. You get the biggest benefit from being in a specifc area where there's lots of game assets/characters that you're interacting with, which is to say the cores need access to a very large amount of data, at the same time, and you need to keep reusing that data.
So I could say Steve from HDWU doesn't benchmark right and have some justificaction when he's dealing with open world games where you would be interacting with game assets much more so than what he does, which is none at all. If I want numbers that are more representative of real game play I need to watch videos that show real game play and there are plenty of channels that do this, but they aren't the main channels that do product reviews, etc.......... These are the channels that do CPU or GPU comparisons like an XXXX vs. XXXX video. But some of these channels don't show game play either. For the ones that do they pick an area to deal with the game assets in that location and that's going to give me an idea of how much Vcache is benefiting game play since I keep needing to access the same data and that's EXACTLY what cache is for, or at least L2 and L3.
On the other hand there are plenty of games where the data load for the CPU is less. For me to make a blanket statement that you ONLY represent real game play as far as how fps is going is when you are interacting with game assets would be wrong. In some games the variation in FPS may be about the same whether I run through an area or interact with the game assets.
It took me a lot just to describe testing in games especially in relation to CPUs that have a very large L3 cache. At the beginning of this comment I said people have to think about what it is they're seeing with benchmarks and does it really represent what THEY do. Without thinking about such things you're watching a video that is meaningless to YOU personally, other than maybe beating your AMD or Intel chest because you're a fanboy.
For me personally I know I benefit the most from CPUs that use less power and right now in game loads which affects me, something like the 7800X3D is DESTROYING anything that competes with it for power consumption and that matters to me. I rarely need to watch a review because I can go to Techpowerup's website and look at what they show which gives me more detail with issues like power consumption across multiple workloads. It's a rare review that gives more than one power stat, and it's usually in a way I don't use my PC.
YOU need to learn to think for yourself and understand what you do and how some benchmark you see means ANYTHING in relation to what you do. If a channel is continually showing benchmarks for video editing/rendering tasks it doesn't mean anything to me because I don't edit videos. But reading the comments sections you'd think everyone IS editing videos. I doubt it's more than 5% of PC users.
👍 I predict there will be a day when that young man is paid whatever he wants to stand there and talk. Steve rules!
Background shadow is nicely raytraced
gmu and techjeebus man this is gold
I look forward to these two talking more than the actual products when computex rolls around
While it sounds very nice to pick games in different genres to give an idea how a GPU or CPU performs in different situations, the fact is there can be huge variations _within_ the same genre (i.e., one FPS game can perform a lot better on Nvidia GPUs while another performs better on AMD GPUs). So, when you test game X, you're effectively testing game X, not "games of the same type as game X".
I'd say it's a lot more important to test games that use different _engines_ (which Steve also mentioned) than games of different genres (that might use the same engine).
Although the best games will tend to have loads of custom optimisations, so even then you're often still "testing game X" more than you are "testing engine Y".
Log into my cam web. It was a phase. We used this phase in the early 2000s. 😂
Steve from GN and Steve from HU are the two most respected reviewers in the business.
One of the Best Videos I've watched shits hilarious. Login!
Early 2000's Nvidia cutting stuff / changing how benchmarks ran was a BIG no no, now its a feature.
I'm never seen Gordon smoother !
Y’all are great together
calling the design flaw in the 12vhp cable user error, that's why people say steve has sold out these days. looks like selling out to me, if it's not designed to stay plugged in that is not user error.
What’s a “login” guys? Does that mean I need to hire a lumberjack to access my confuser?
Back in the Adobe CS5 days intel had a quick sync plug-in, for years adobe did not use quick sync by default. Was a good benchmark for intel but most people did not instal the plug-in, then when Adobe added in the plug-in to the program by default they advertised it as 'look how much faster the new version of PP is'.. Adobe being adobe
Talking about the new Nvidia GPU's Pugetsystems benchmarks show regression in pro apps with the 4060 TI and 4070 etc.
PS Nice to know Gordon will set up 'onlyBenchmarks', a new way for people to spend there money.
If I don't see a PC world MFC or camsoda account within the next 3 months I'm going to be profoundly disappointed
There's some serious moneys going on here. More or less to the knees. You guys have swimming sleeves, don't you?
The two areas, I personally feel, are missing from benchmarks are,
1)Sound/Audio
2)Network throughput
Steve acts like a teenager girl besides Gordon ... so lovely :)
what are the credentials for your webcam
I'm kinda dissapointed that Tech Jesus can't just add a few more hours to a day.
One option would be to benchmark at the officially rated RAM speed.
I'm trying to log into the webcam but I can't find the coin slot? Now what??
I couldn’t watch this video because I wasn’t logged into my webcam :(
I am baised and happy to admit it. I am biased towards whichever vendor will give me the best products at a reasonable price with the most useful features.
i don't personally see a problem with a Chart having Asterisks. i think it's suitable to represent Hardware based on what it can generally do, not like pushing the edge, but also not running a configuration that it could easily run faster than just because reasons.
development of Hardware is in more than just one thing. if say, a later Generation of part has a better Memory Controller or I/O interconnect or whatever, those too are improvements/innovations that can benefit the performance of the system. so what if something older could not, that's part of what you're getting from upgrading / Buying that newer thing vs something else in the past. that's a part of it.
if CPU A would struggle to run above like 2666 or 5600 or whatever, that's what that Hardware is capable of. if something else several Years later is capable of like 3800 or 7000 or whatever, **that's a part of what the Product improved during the Years between**.
however, if separate data is made to also include those "this can generally do this config" as well as some fixed settings irregardless of what the Hardware is capable of - then, that is okay since both are being presented.
One of the things that I would hope more reviewers did, was to turn certain settings on the game off that are known to;
A) Add nothing to the game.
B) Have massive effect on performance.
I know this is something that you generally shouldn't do to review how a GPU performs.
But it is something to consider if you're looking at how the game performs in a real-life scenario.
In the case of Cyberpunk as an example, Screen Space Reflections Quality have a massive impact on the performance and overall LOOK of the game.
In that it's better to turn it off for gameplay. And it removes the pixelization on surfaces.
Monster Hunter World for example had "Volume Rendering Quality" which literally boosted FPS in the game by 10-15fps if you turned it off.
And it made the game look better by removing alot of the volumetric fog the game had.
TLDR; It would be great if there was a "custom settings" portion for couple of the games that people really like to tinker with.
"why don't you test everything?" ...looks at day before shipping on parts to test with day before that embargos. Seriously I would LOVE professional app benchmarking but 90% of channels focus on gaming because of viewership. These small/tight review windows have to stop. I'd also ask for more "why did x company impliment this feature" interviews along with reviews for a more comprehensive "feel" of how successful was the part at accomplishing what the company set out to do. Example you did a whole motherboard lineup without thunderbolt, why? May be some good learning to be had there.
I have been inserting coins into my webcam since this video released yet you wont run the clock speeds I want
what gives?
On the memory clocks I used to think if a CPU has better IMC, then it merits using faster ram, however IMC quality is a variable, intel e.g. still only support 3200mhz on DDR4, and two out of three intel CPU's I was not able to get my b die to run stable at over 3200mhz. Those two reasons alone in my opinion validate the modest ram clock benchmarking ion my view.
DDR5, I dont know so much about, but I think applying the same logic there is reasonable, even though Intel seem to be ahead of the game on DDR5, I bet there is poor binned chips that struggle to get above AMD memory speeds out there.
Thanks for transcript!
Gordon's a blast 😊
Gordon is the spirit animal of the internet. I love Steve and Gordon videos!
thanks for what you're doing !
Unfortunately while GN does a pretty good job of catching driver updates and retesting as necessary, sometimes gn stats suffer from their style becuase older cards/chips that may have gotten perf boost thru drivers isn't accurately shown in data.
Not sure I follow. Generally when they perform a driver update they retest all their previous cards that they usually include on roundups with the new driver. Only when they specify otherwise is that not the case.
Observation: For some bizarre reason, Gordon rubs me the wrong way when he's solo, but get him on camera with Steve, and he's god mode. How TF does that even work?
Hello from Singapore, love the content you guys make.
What I don't like the most is that GN uses an ancient methodology for PC case testing. Good overall case review but meaningless tests.
3:13 OnlyBench? :D