Видео

Anytime you send something in for repair most often they send you a different one that has already gone through whatever their process is and determines usable... the conditions should be evaluated so that you're sent one in the same condition... would be a simple thing to fix.

Interesting vid. The reason why testers often skip cpu testing on 1080p low game settings is really simple. It's not a realistic scenario. People who buy the 9700x cpu for games ar not going to play games at 1080p low settings. It makes sense to test a cpu at 1080p low settings for the purpose of testing, but it does not make sense for the people who are going to use this cpu for actual gaming.

You will never be a real frame. You have no contact with the anything but the tensor cores and ai tops, you have no native resolution. You are a crude mockery of a frame, extrapolated with artifacts and blur behind the engines back, and inserted in between real frames, introducing input lag and noticeable distortions in the graphics. All the validation you get in benchmarks and frames per second counters is two faced and half hearted. Behind your back gamers mock you. Your mega corporation and shills are ashamed and disgusted by you as you are a hard sell. Players are utterly repulsed by you. Years of gaming has allowed gamers to sniff you frauds out with great efficiency. Even DLSS 4.0 frames who “pass” look uncanny and unnatural to gamers. Your appearance and structure is a dead giveaway. And even if you manage to get a drunk guy to play a game with you on, he’ll turn tail and bolt the second he gets a whiff of your diseased, infected pixels.

Very informative video. Although I was left wanting a comparison between the 4060 Ti 8 & 16 GB cards using the DLSS technology and subsequently with Frame Gen, just to be able to see how they fare with upscaling. Otherwise brilliant stuff all around.

There is a problem with DLSS made frames you didn't touch and I often feel is being left out... no it isn't the latency. It is the fact that the gamea needs to support DLSS, FSR etc. If they don't then you only have the "real frames" . Not many major games will have that issue but there is AMAZING games out there not supporting it because they are made by small studios. And thats why I think defending using those frames is wrong as marketing, because it is a borderline lie, and it is arrogant towards the indie studios where alot of the real new ideas for gaming is starting.

I won a 7900 XTX from some clown a while back. I water cooled it. I was on the pure rasterization bandwagon before this video, but you make some great points. I don't game much anymore, but I want to start playing around with the frame gen tech. Just to see if I like it. I can tell you when VRAM really matters...video editing. The 7900 XTX is a BEAST when editing/rendering huge videos. I take it for granted because it just works as it should, but I can render 10 GB videos in just a few minutes. Great video!

I have a CPU: AMD Ryzen 7 5700X, RAM: 96.0GB Dual-Channel DDR4 @ 1596MHz and I need a new graphics card for some gaming and mostly video editing. Which one would you recommend under $700? Amd, Nvidia? Thanks

Perfect discussion around generated frames. No one will be able to tell the difference 👍

My electric bill and the fps counter said that the fake frames are real 😂 The AI cores draw a lot of power to bring the fake frames to reality. Fake Frames Matter!!!

GPU vram really important🤔Nope i still use a 3060 ti 8gb and a 4070 12gb at 1440p with no issue, this goes for 3d animations and rendering as well

Here is the thing, I agree with all the testing here but its still not enough WOW for me to upgrade an entire X3D system from a 5800 to a 7800x3d or 9800x3d. 100+ fps generally is not going to be felt in a game. Sure bench marking wise you get 30 percent in some games but 241 fps to 300 fps is not really going to be felt only seen on a bench mark. I see no compelling reason to update cpu/ram/mb etc to a 9800x3d just not a WOW especially with a 4070FE. Now I had a bunch of games on the edge say running at 40fps and I could boost them to 60+ then maybe I would think about it but its not the case with 4070FE/1080P/240hz. I have two 65" OLEDS but never play games on them I prefer smaller up close high refresh. Don't forget a total system rebuild takes a huge amount of time to test and reinstall everything to get it working like the old system. Some might love it, but the older I get the less LOVE I have for tinkering that much for 30 percent. For most people I would just swap out the non X3D for the 5700X3D plug n play upgrade no need to reinstall the OS etc. Good enough dude for most folks.

In Escape of Tarkov I easily fill up 16gb on medium'ish settings. But EFT is special kind of game regarding lack of optimizations

You earned my sub! Super objective! Please keep it real like this and I'll 100% stay around

I made a poll on this topic just yesterday on Facebook. Asking will VRAM effect Nvidia in the future more/1st Or will RT performance effect AMD GPUs more/1st. It was fairly even, but to my surprise, more people chose the second option. Showing many people already believe Nvidia will succeed in this vram compression technology

Thank you for your video, I'm considering upgrading from my i7 8700k, but the price in Belgium for the 7800X3D and 9800X3D are crazy high. So, thanks to your video, I have another option to consider for my upgrade that will save me some money while allowing me to change everything I need to change.

Agreed on the 'fake frames' argument. However there is one issue which is the DLSS frames dont improve latency they only improve smoothness. Which makes it more useful for single player games and less useful for competitive shooters.

under 4k, RT and ultra settings, 4060ti can run 2077 with 41fps? This doesn’t sound right.

I have a small system (my big rig with the 4090 is only for work, I work mostly as a 3D & VFX artist these days and use the 4090 solely for GPU rendering) with a crappy RTX 3050 8GB, and I run Cyberpunk at 1080p at 150 fps on average on an 180 Hz monitor. I do this with a FSR 3.1 frame gerneration mod that pushes the framerate to 150 fps, DLSS upscaling is activated (FSR 3.1 can utilize DLSS upscaling, which FSR 3.0, inbuilt into the game, can not) and set to performance mostly because "balanced" shows some strange artifacts at the edges of the frame (no other setting does that). I also have set a custom upscale range to use if I need it and the artifacts appear if I set the lowest fps target to the lowest setting relative to the (capped) highest settings. All graphics settings are set to high or "psycho" (if available). I get a frametime of 7ms, which is quite phenomenal. I can use Raytracing (simple reflections) too, but I don't see much of a difference in comparison to screen space reflections most of the time. I only see the difference when I actually stop and look for the reflections, or when I look out onto the harbor, the water looks not that good with screen space reflections and the simple speculars don't do it much better either. The frame time with simple reflections (no DLSS ray reconstruction) jumps to 11 ms on average, which is still acceptable. For a more smoother experience I can even turn on Lossless Scalings frame generation, but that shears off quite a bit of fps and it's also pixelbased, so distortions happen. The frametime goes to 15 ms when I do this. It is an option to use it on top of everything (or just turn off FSR 3.1 and use Lossless Scalings FG instead, which gives me better results than putting it on top of FSR 3.1), but it is of course rather suboptimal to use FG (pixelbased) on top of FG (engine based, which is much better on its own). Without raytracing but with AI trickery activated the RTX 3050 8GB never goes over 6.5 GB usage for Cyberpunk. With Raytracing it hovers around 7.6 GB, which pushes towards the limit. But it is absolutely possible to play Cyberpunk at the highest settings (excluding Raytracing) with acceptable frame times and general fps if you make heavy use of AI upscaling and frame generation.

The problem with fake frames is that they are fake (as in not connected to the game's calculations). Any game in which frame counting is important will become scuffed with the fake frames added in. You will miss shots in fps games, mistime jumps/inputs in some platformer games, and miss combos in fighting games. Yes, this mostly affects competitive gaming, but it is sill important to note that you are essentially creating some input lag when you use DLSS. They are not cost-free frames. I think most people don't have a problem with AI generated frames as a feature, they are upset at the deceptive marketing of them. There is a drawback to using them for some games. I believe that most would agree with me that until the game engine can respond to an input done at the generation of a "fake" frame then the proper way to market GPUs is to show BOTH raw and DLSS fps.

At 4k i think it matters. At 1440p not as much. My 4070 only uses about 10-11gb in cyberpunk on max settings. Hogwarts rarely uses more than 10 at max. However noticed a few games using more than 16gb ram though. Caught hogwarts using 24gb

The reason why Hardware Unboxed tests gpus in ultra settings is because it's the most realistic scenario. All the PCs I've built except for a few that were not for gaming were required for playing at Ultra Settings, ranging from 1080p to 4k. Playing competitively was just a bonus by lowering graphical settings. If you are going to spend on the latest and greatest then you have to expect the best. Don't get me wrong, I do, however, find this content about content creators very interesting.

bit of a silly qestion as DDR5 will need a new mobo and cpu too lol. so DDR 4 wont come in to it.

Subbed.

Well this was useful ;) Love it!

Can these optimization settings be achieved with B650 motherboards, regardless of the brand, such as Gigabyte or MSI?

Recommending users to disable memory integrity features can potentially increase security risks by creating vulnerabilities that could be exploited by malicious actors. However, I appreciate the thoroughness and methodology employed in your testing process.

One important thing with respect to undervolting that I didn't mention in the video, is that you also need to disable hypervisor or hyper-v in Windows before XTU will allow you to use the "Core Voltage Offset" option.

Saying that if you cant tell the difference visually between "fake" and real frames means they hold the same/ similar value is just plain wrong. Any competitve game where latency matters frame gen is useless which happens to be an auful lot of games. Also frame gen cant be used effectively below 60 which is the main area you would want to turn it. Also in games where you would use it, non competitve games i still wouldnt use 4x and would stick to 2x as i would rather less artifacts at 120 then more at 240 The main issue is most peoples target in games where they would use frame gen is 60-144 where 3x and 4x are not useful

How Much VRAM is needed for VR and Flat to VR games?

Thats good to hear 16gb would more than enough with 5000 series with dlss 4 being more efficient with resources.

One more question: can I apply your recommended settings? Or does it come down to silicone lottery whether a 30mV undervolt will work on my system? What would happen if the voltage is too low? And is it a good idea to test the RAM expo settings (6000MHz and CL30 in the case of the RAM I ordered) and see whether the system is stable, and then fiddle with the CPU settings? And I am a little bit wary about using ryzen master. I like to do BIOS stuff in BIOS and not via windows. Any disadvantage of applying your recommended settings just in the BIOS? And what about low-power useage, like just idling the system, will there be any problems when the cpu doesn‘t get enough juice?

1. Most of those techtubers made same comments about how AMD marketed these chips. 2. Those "lazy" 😅 techtubers who run a ridiculous amount of benchmarks and engage with AMD to try and figure out what is going on... Could not recommend the CPU compared to the similar version of 7 series "AS IS" with the CPUs pricing. 3. Because I watched a another lower sub techtuber. I was confident in buying 9700x for an over priced 350 euro. Using buildzoids (old) memmory timings and 105w fix with some slight fije tuning. You can get near enough to 9800x3d. Which at the time of purchase is half the price. 4. Faster ram is not the fix that is being claimed here which leads to a small fps increase. At the price for higher rate memory its insane. Just get the well priced 32gb 6000mhz CL30 ram and use the buildzoids timing which you can find with a quick google.

It is a good question, and I ran into an interesting problem with my AMD 7800 XT. When running Stable Diffusion models, having 16GB of VRAM is not enough to get the special pony and SDXL models running. However, if I use my older Nvidia 3060 Ti, it runs them all fine - with just 8GB of VRAM. Apparently Nvidia has something where they can share system memory with video card memory when it runs out, which is huge for that situation. Thumbs up.

Nice to see you blowing up, glad to be able to say that i was pretty much there from the beginning! Great video, I pray i can snipe a 5080 when it releases. 🙏

So, would it be better to just get 8000, future proof wise? Seems hard though, not too many manufacturers are even pushing it right now at this time.

What is the tdp of this configuration?

What I know for sure is this, despite the backlash and hate towards nvidia, and legitimate anger against their marketing decision of promoting 5070 as an equal of 4090. The nvidia 50 fake frame series will sell like donuts near a police station. Do not wait to much, it will be a shortage and 5090 will most likely be out of stock for months. I already know about 3 of my friends which are buying the Asus Astral 5090 in the first day they hit the shelf. For those who buy nvidia nothing matters, they will buy the cards and enjoy the games they like, so no matter how much the techtube couch experts will backlash, these cards will be out of stock. Happy gaming and thanks Matt here who is trying to offer the best content without bias and with tons of empirical evidence. All the frames generated by these cards are the result of hardware processing, like it or not, this is the reality.

So 4090 vs 5080? I think I'm leaning towards 4090. I'm making a new PC for myself this coming February. There's a shop in my country that sells MSI Suprim X 4090 for 2k usd.

I would still call those generated frames as fake. As explained, a frame is generated between frame 1 and frame 0(frame 0, being the previous rendered frame. But my issue with frame gen is that, regardless if the game is online or offline, game engines still only recognized and include real frames on it's calculation. Allow me to give you an extreme kind of example. Let's they that you are playing an fps online shooter game. While roaming, you saw your opponent running fast from point A to point D. The frame zero showed he opponent's body/hit box at point A, while frame 1(next frame), is already at point D. Like if a character uses a skill to be extra fast. But that game has frame gen, and it generated two frames between point A and D, created an image of your opponent being in point B and C. Again, this is an extreme example. But if my chance you shot and hit your opponent, but while your opponent is at point B and C, the game engine, won't recognized it as a hit. That's why no one is using frame gen in fps games. imagine using the x4, where in there are 3 fake frames in between real frames. Until the game engine includes those fake frames in their computations, I would only consider frame gen as fake and cosmetics.

Is a B650 on with the 9800x3d?

16gb minimum. just try stalker 2 at 1440p with those GPUs and you will see why i say 16gb. Rosstok area in that game will take you over 12gb.

Hi, again I have a question. I read about Chipset driver "Version 7.01.08.129" is it woth it to update to this Version with the 9700x? Or did I understand this right, that its only food for the x3d Versions? I´m asking because I dont want to loose all my settings in the bios.

one of the ever best gaming tech youtubers out threre! Keep it up Matt! Your content is amazing and so worthy!

Matt we need an update for that video now or in a few month! I'm very curious how much the updates will squeeze our of the 9700X =))

Let's be real, a 2500€ RTX 5090 with 32 GB VRAM is enough for anything upto GTA 5 at 16k resolution. You don't always need to buy the shiniest new toy on the market.

The most Vram i have seen my GPU use in games is around 13GB, i think 16GB is the current sweetspot.

Big unknown if most 9700x will undervolt and run higher speed ram kits with all the different motherboard revisions available. Maybe AMD has improved the odds since AM4.

You use ultra when testing cpus because there are settings that impact the cpu? Higher population density will affect the cou petformance so obviously its worth turning it up. Are you saying the rtx 4090 was limiting the cpus at 1080p ultra? Cuz if not then it doesnt matter that they tested it in ultra and is probably better to do so.

Sumary of internet talks about graphic cards in 2025: fake frames, fake frames fake frames fake frames fake frames fake frames fake frames fake frames fake frames fake frames fake frames fake frames. This is how much the understanding of users and 'reputable reviewers' is about computing, ray work equations, frame generator algorithms, and tech in general.