[SPON: Use "brokensilicon“ at CDKeyOffer to get Win 11 Pro for $23: www.cdkeyoffer.com/cko/Moore11 ] [SPON: Get 10% off Tasty Vite Ramen with code BROKENSILICON: bit.ly/3wKx6v1 ]
people overthink things. All you need is a gpu that can play the games you like for a price you can afford. PC builder just did a poll and right now what people want in a gpu is "Most raw FPS for my budget" with eighty eight percent. Dlss has four percent and raytracing has three percent. With a minimum of fourteen thousand votes. Raster is king and we'd buy cards that don't even support raytracing. If that gets the price down, why not make them?
@@TheDarksideFNothing sure but nvidia forces you buy the 4070 super TI to get 16 gigs. If you are cool with that, great. I'm not OK with 800 dollars for video games.
@@doublecrossedswine112 lol me neither. I'm just saying I don't think I'll upgrade unless something with more VRAM and meaningfully better RT than what I currently have is available
@@TheDarksideFNothingI'm guessing you play a lot of single player games where RT is somewhat ok to enable. To me RT doesn't really make a meaningful difference in the feel of the game even when it's maxed out to make it worth the performance hit. The smoother gameplay just feels so much better without it.
Transparencies are difficult to temporal upscale, since there’s only one set of motion vectors per pixel. Usually developers use the opaque geometry motion vectors and transparency doesn’t have motion vectors (it’s using motion vectors of the geometry behind the transparencies). FSR 3.1 doesn’t even support multiple sets of motion vectors. IIRC the other upscalers don’t support that either. Similarly texture animation such as water surface is hard, since technically the water surface only moves up/down, while the ”visible” wave moves (along the surface). Same problem with all implicit surface techniques (distance fields and similar). Needs special effort from devs and still all they can achieve is a compromise. I am not deeply familariar with the DLSS3 algorithm. It’s not open source like the others. But they could do AI based motion generation pass from the image to help in these cases. Kind of similar thing that HDTVs do with their pure rgb image (no motion vectors) based 120Hz image interpolation algorithms. But if you can’t get subpixel accurate accumulation, then you can’t properly reconstruct higher res image from multiple jittered low res images. It would just be blurry or blocky. Good quality motion vectors are crucial. AI based algo might be able to workaround it somehow.
RDNA3 doesn’t have separate tensor/matrix cores like CDNA3 and Nvidia do. It only has wave intrinsics (WDDM) for emitting 32 cycle optimal ALU sequence (with register routing optimizations to save power). Double ALU pipes vs RDNA2 help of course in matrix heavy workloads. It’s a big uplift over RDNA2 for AI work, but if RDNA4 is going to have CDNA3-style matrix cores, then AI perf improvement will be substantially bigger. Which likely means that AI based upscaler would not be backported.
AMD could easily make an upscaler that has better image quality however it would also produce lower frame rates which defeats the entire purpose. An upscaled that uses most of the performance gain of a lower resolution just to upscale the image is useless. May as well just run the higher resolution with better image quality.. I would dare say that AMD could make an upscaled that produces dlss like quality but you would turn on quality upscaling and LOSE performance. Unusable.
This is precisely why NVIDIA doesn't allow DLSS 3.0 (Frame Generation) to be turned on previous generations of RTX (before) 4000. It simply doesn't work as shown by hackers.
@@KarbinCryXeSS literally has 2 different modes, one using the matrix cores (and being better) and another not. If AMD is going to be using matrix cores for FSR then they wouldn’t be able to have it work on RDNA3 and older and Pascal and Older.
If Linus claims not having driver problems after x many months ago from switching to his 79XTX, AMD cards are good enough for me. I'll switch now and then depending on what used deals are coming my way - so I have no bias
Linus is currently sponsored by AMD, but yeah their drivers have been good enough for years to call them stable. They'll still sometimes drop the ball like with Fallout New Vegas crashing on start-up, but their recent fix to that was quite fast.
I've used Radeon VII for years with maybe two driver issues. Half a year ago I switched to a 4070 Ti and have already had several driver issues and frequent crashes. Individual experiences are inherently more about luck.
@@louisfriend9323the FO3/FONV thing was intentional. To drive attention to the DXVK mod - just like 4GB patch, there's no reason not to use the DXVK mod.
as a life long Nvidia and EVGA customer, I am loving my 7900XTX , and with FSR3.1 able to use xess or fsr in many games with frame gen, I am having a blast! I got it for 999 when the 4080 regular was 1200 dollars. I refused to pay over 999 for a gpu... The 4070ti only had 12 gb ram for 800, SO I LEFT NVIDIA! For 200 dollars more , I got 24 gb 7900xtx and that is how AMD won me over. I have been really enjoying it. I just hope FSR 3.2 and 3.3 etc is coming soon.
i've run Nvidia and Radeon and the driver for both are good people tend to blame amd for any stability issue in their system when it could be for exemple their overclocked RAM running xmp.
Tom, I really love these episodes with Fabio. You should have him back on after rdna4 and ps5 pro. I’d love to see you guys circle back on the discussion after we see how those products work out + any future FSR improvements.
Nice to see AncientGameplays on the channel. I found his content very helpful when transitioning from 1060 3gb to 6650xt 8gb in configuring adrenaline etc.
It's like Alex from DF pointed out it feels like FSR doesn't do anti aliasing on moving pixels so it appears really blocky. You're looking at raw pixels. Maybe they're afraid of making it too blurry or something.
Also it's just really really hard lol I imagine Nvidia is putting a lot more money into DLSS because they have a lot more money and also recognize what a win it is for them.
What you typed out makes sense as a sentence, but it doesn't actually make sense. If you're seeing pixels(ignoring moire, something is wrong with your image) Real footage off a proper camera won't be pixelated at even 1080p. Making out pixels=/being pixelated
It makes sense, when pixels become disoccluded there is no historical/temporal samples/information available. The algorithm is going to emphasize the new samples and the output appears very undersampled. It makes sense this would also occur in fast motion where the algorithm lowers the contribution of older and outdated information. You need historical samples for temporal super sampling and super sampling provides anti-aliasing so that's why these scenarios look poor. This may not occur in other upscalers if they are filtering or otherwise having this disoccluded information resolve differently. If you don't discard historical information for disoccluded pixels you will get a smearing or ghosting effect.
VR. That's where we need more powerful GPUs. Modern VR headsets run around ~6000*3000pixels for their native resolutions and 90hz at minimum. And then you often still need some form of (non-temporal) anti-aliasing like MSAA on top of that to reduce shimmering and make things look sharp in the distance. Even the 4090 still struggles with a lot of titles and you have to make compromises and tweak settings (and of course really bad optimizations are a big part of that too). Apart from that? Regular gaming with ~60fps in 1080p-1440p? 1080ti is still fine :D
No one would imagine that dlss' use case was to save ram. Upscaling is creating filler frames for smoothness? Wouldn't that be.........................................................m Framegeneration?
On HU in Tim's defense he also briefly included a chart comparing the upscalers with same preset names. I guess he was covering his rear, anticipating the critique about comparing quality base on performance. And I am glad he did that from performance stance point given that almost all other content creators comparisons, did the test on matching quality names presets. So its nice to get Tims perspective.
From going to a 1070 to a 7900xtx, Ancient Gameplays helped me with going through the settings. 👍 I'd have to agree, with all these new settings available and going from 1080p to 1440p..... It can be quite difficult to set everything up to run smoothly and hopefully efficiently.
Upgraded from a 2070s to 7900 GRE red devi (my first amd card) and I'm happy with the change. Great 1440p performance, decent power usage, plays all my games on max to high on 1440p. Decided to stop sticking to nvidia for a change and amds pretty good. Still looking at what pricing models and performance will be for 50 series. Not sure what a lot of the hate towards AMD was about.
@@Dave-dh7rtthis. They don't really have a choice unless they want to throw the towel on competing above $500. Then again, AMDs favorite move is to do the opposite of what we all think is the obvious best move lol
I'm part of that community. We went through hell on 7900xtx in some scenario's for over a year and the community was able to sanity check and workaround almost every issue.
41:25 But they did it because XeSS gets less fps. They tested the upscaling settings that give roughly the same fps boost. Which is perfectly fair. If XeSS needs to run at a lower setting to achieve the same fps as DLSS or FSR, then that lower setting needs to be compared.
100%. The universal version of xess has more of a performance impact which is why it has better quality(at the same internal resolution) than FSR. AMD could easily make an upscaled that is dlss like or even better but... It would lose performance instead of gaining it
1:25:40 100% agree with Fabio through this entire section. I've never seen a game that you can't get running well with settings unless your hardware is _seriously_ lacking. Don't get me wrong, vram is a real issue and 8gb is seeing its limits, but I've been seeing it get exaggerated worse and worse over the last year and it's mostly from the AMD fans trying to use it as the sole excuse to buy AMD. That HUB video raised my eyebrows because they listed 2 games that I've personally played that their graphs were _absolutely_ exaggerated. It said R&C needed a MINIMUM 8.1gb of vram to even play at 1080p _medium_ even though it ran perfectly fine on my kids' 6gb card without any massive stutter, then claimed another needed 14gb that I've seen run perfectly fine on 12. I suspect he still kept texture resolution cranked through every test, or it's because he was using a 4090 and the game was allocating more than it actually needed when it was offered. I normally agree with HUB but this vram one seems to be a personal crusade for Steve where I'm starting to see numbers that I personally know just aren't accurate. It only takes 5 seconds to pull up a YT clip and see his numbers are flat out wrong when this game shouldn't play on a 1660 super without massive stutters at medium, let alone high. ua-cam.com/video/yAVXEGkmceE/v-deo.htmlsi=5qRXgDn0oC5uGMlo&t=92 Somehow this 6gb card isn't stuttering at settings HUB says needs 8.1 and 9.2gb. The card very clearly runs out of gas long before it runs out of vram.
The problem is that we don't have competition just duopoly cooperation. It was clear to see just after the crazy 80-class launch for $1200. What AMD did do? They could do everything to take the market share but they decided to release the 4080's competitor (7900xtx) for $1000. What should they do then? When they realized that they have only an 80-class competitor to price it correctly (~$750) and not blindly follow totally crazy Nvidia prices. AMD is not a competitor just a blind follower. If Nvidia decides to release xx70 for $1000 then AMD will release a similar card for $900 and will call it a day. Like always. This duopoly is killing this market and not only this one because without upgrading GPU people are not upgrading other parts like CPU, MB, RAM, and even monitors.
That's why my next gpu will be Intel, As much crap as people say about them only proves they are not voting with their wallets but blindly buying the duopoly option instead.
@@jasonvors1922Yeah same, if Intel prices their battlemage parts reasonably and the performance is a solid generational jump then I'll absolutely grab one of those. They (the duopoly) desperately needs the type of shakeup Ryzen and later ARM have brought to the CPU space
6800 for $360 is working great for me don't miss: MSI afterburner NVCleanstall NV control panel Windows NVIDIA telemetry fiddly, melty power connectors running out of VRAM spending $200-400 more on AI, CUDA, and sadly on this one note: RT Nice to see the ancient gameplays co-op.
Guys... you did not suck! It was a very good talk and all of those topics you talked about are interesting. Yes, the #1 requests from gamers is cheaper GPU prices. Also from Nvidia : more VRAM! From AMD : better FSR 3.x
good stuff here on frame gen. I play gta online with a 6700xt and a 5800x 3d. there are a ton of frame skips in crowded lobbies, especially with modders. turning on frame gen almost completely removes frame skips. I go from 90fps to 150-160 but the massively improved 1% lows are why I now keep it on. AMD won me over on the idea of frame gen with their execution. 💪
very proud of being a subcriber of this channel and AncientGameplays... magnificent interview as always...i am 50 years old and still playing games...AG makes best GPU drivers analisys out there....
On the FSR and DLSS issue of missing details. I find simple integer scaling to be a good alternative to AI upscaling. Setting to integer gpu scaling in NVCP works automatically if the game's resolution setting affects the output resolution, but if the game only changes the internal resolution you will also need to set desktop resolution to whatever lower resolution you want. I use 1080p integer scaled to 4K for demanding or old games with UI scaling issues and it looks really clean since it divides perfectly, which allows the screen to be filled in integer scaling mode. No artifacts, blurriness, missing details, etc. 720p for 1440p displays for the same effect.
@56:00 The problem with talking about hardware accelerated upscaling on AMD's GPUs is that they actually need to have the silicon to do it. They don't, haven't talked about it, and it goes entirely against their current strategy of using general compute resources. AMD sectioned off CDNA from RDNA for a reason and I don't think they will want to reverse that decision any time soon. So, if Sony is doing hardware accelerated upscaling, that's their strategy, their resources...
Basically, upscalers alter the creator's original intent. To this exception, we have Nvidia which claims that its DLSS rather embellishes the creator's original intent. IMO: stick with native creator intent, upscalers are gimmicky at best.
DLSS does, as it uses AI to rebuild textures when there isn't enough data from the source image. FSR on the other hand is more of a next generation anti aliasing and holds very true to the original image. DLSS is more detailed because it's adding data that isn't present in the original image, if you watch a side by side by side, FSR is much closer to the native image while DLSS is like a very minor remaster.
@@PineyJusticeRecently i watched a comparison between native path tracing and nvidia's ray reconstruction denoiser in an graphics editor. I was shocked about how much details have been lost with recent dlss still. We need way better denoisers than what the industry offers currently especially with the planed path tracing revolution.
@@aladdin8623 Upscaling will continue to get better, in another 4 years both FSR and DLSS will be be standard, totally replacing AA. The tech is one of the best things to come to tech in the last 20 years, eventually all cards will have hardware upscaling accelerators, just as we now have hardware AA accelerators. Upscaling gives hardware a generational boost in performance for very few downsides.
This is ridiculous. Changing resolution, graphics settings, or using a monitor different from the original creator would also be changing creators intent. Lol.
RDNA 4 isn't going help AMD people want see high-end gpu's. As someone that owns a 7900xtx I wanted a high-end gpu that beats the one I have but since they won't be. I'm switching to the 5000 series
AMD should make using RSR easier in titles that don't natively support up-scaling. While setting up custom resolutions works, it's an inelegant workaround that most people wouldn't be bothered with. I wonder how many people even know this is possible in the Adrenalin software. I've used 1,224p to get better performance in 1440p titles that don't have native up-scaling, or were left with only FSR 1.0.
I am always surprised how good the driver upscaling from and is. If you turn it on in the driver then open the game, all you need to do is change the ingame resolution then AMD upscales it. Super easy. You want a 50-100% slider in the game?
@@christophermullins7163 If there were a slider in every game that would be amazing, but that's not gonna happen. I have found that using RSR to up-scale from 1080p to 1440p is too much of an image quality hit at 75%. However, 85% strikes a nice balance. To get 85% a custom resolution has to be created. At the very least it should be a present within Adrenalin, rather than having to create the configuration yourself.
Guys early dlss 2 is still better than fsr 3.1 especially 2.5 because thats where dlss start introduce preset c and further image quality improvement. You are delusional to think fsr 3.1 is better than dlss 2
Generally I'd say with FSR 2.2 at 4k, at least the quality setting usually looks quite good. Balanced has been worth it if you're not quite getting the frames, and then performance was noticeably worse but decent enough. I haven't played any 3.1 games yet, but I really hope it gets back added to Avatar and Alan Wake 2. I really, REALLY hope people who know devs can request people go back and update their games for it. Alan Wake 2 in particular needs it.
Found Fabio’s channel back in 2020 after looking for under volt settings for a 5700XT. Worked perfect. Still have have the same file name for my settings on my GPUs. Ancienttune
RDNA4 is doa if they wait until q1 2025. At that point you just wait and buy Nvidia. If its a $500 7900xt performance card with less vram 6 months from now nobody cares. Wow waiting 6 months to save 100 to 200 dollars on a 7900xt with less vram.
I feel bad for anyone that bought the rx 7800xt it performs almost exactly the same as the rx 6800xt the rx 7900 gre preforms almost exactly the same as the rx 6950xt, This whole generation from amd has been a joke.
Hyped for this episode - two great content creator in one format! Fabios AMD heavy content is really helpful since AMD was and still ist the only logical option in my region when it comes to price/performance. Learned allot about undervolting AMD GPU's from this guy!
Intel iGPU drivers were ALWAYS awful. It just wasn’t as noticeable because most games wouldn’t run fast enough. But ask any emulator developer about intel support on GPU plugins. They were always missing or bugged out on a BUNCH of features. Heck, Krita and Kodi, two 2D apps with GPU acceleration, were straight up broken back in the Skylake/3000 era iGPUs.
RDNA4 is gonna be a snooze unless they can deliver XTX performance for 500$, or 7900xt perf for 400$. If it’s a 16gb 7900xt for 500$, then they can keep it. Lol
On the rain thing for AMD I assume the error is in the alpha handling, with alphas there is pretty much straight and premultiplied and coders are not compositors. Getting the interpolation wrong and likely it needs to be set per game the cutout mattes can turn a bubble into a white sphere for example. So likely the work is getting done frame wise but the texture map on the plane is getting "choked" or clipped for most poeple.
$2500 may be a bit high but 2080TI was around 30% faster than 1080TI but about 40% more expensive (MSRP). RT was not really a thing back then, so it seemed overpriced to me at first. However, thanks to RT cores and architecture it aged good and is more than 30% faster than 1080TI now (without RT of course).
What I want from next gen is 7900 XT performance or better with 16GB VRAM at $500 or less. If they can do that it would mark the first time since I bought my 5700 XT I thought the price to performance was at a good point. RDNA2 and Ampere *should* have been good price:perf, but of course then the scalpocalypse happened which also cemented stratospheric MSRPs. I have a 6700 XT now because I cashed in on the 1:1 swap for a 5700 XT during the crypto boom, but the current gen offers so little price to performance I haven't even bothered looking for deals, I'm just not interested. But if I can get double the 6700 XT's performance with 16GB VRAM for $500 next gen I'll be all over it. I don't give a toss about ray tracing, 6 years since that hype train started and it's still a novelty add-on, maybe I might start caring in another half decade. And with appropriate raster performance why the hell would I need upscaling/frame gen?
@1:06:20 I'm not understanding what is meant by "AI" cores on RDNA 3. As far as I understand it, each CU can split the wavefront so that two FP16 calculations can be performed instead of a single FP32 calculation. That's just an accuracy thing, there's no "new" AI cores in RDNA 3 vs RDNA 2. It's the same thing going from GCN/Vega to RDNA where they broke up the FP64 functionality and improved throughput at lower accuracy...
Fabio loves to downplay the the 8GB VRAM issue. Nobody should be going anywhere near an 8GB GPU when upgrading. He contradicts himself about it by saying "settings" and memory bandwidth + memory speed/latency but every modern 8GB GPU is shipping with 128-bit memory buses or less and the cache isnt doing the job. Maybe the 7600, XT, and 4060 would all fair a lot better at 8GB if they had a much wider 256-bit memory bus, but those arent a thing anymore on lower/mid ranged cards.
What I stated it that they were fine for the price ($269 RX 7600 for example), guess I didn't explain myself correctly. Also, you nowadays have narrower bus, but more cache
Fabio just overtook Steve from gamers Nexus for silky hair in livestream. I'm hoping FSR would become standard in future just like freesync with custom vendor extensions per GPU vendors or game studios. FSR 2.1 just had better playable frame rate on handheld and older GPU. Don't know why nvidia ships lower vram and maybe it's time to add M.2 SSD slot to stream larger than memory textures, tensors and general compute just like Vega SSG.
[SPON: Use "brokensilicon“ at CDKeyOffer to get Win 11 Pro for $23: www.cdkeyoffer.com/cko/Moore11 ]
[SPON: Get 10% off Tasty Vite Ramen with code BROKENSILICON: bit.ly/3wKx6v1 ]
Tom out here looking like an 80s IBM Engineer. 🤣🤭
Reject modernity, embrace payment per LOC
I mean, he IS an engineer
Am I making this up or did he say the his dad used to work there
I think he did. Pretty sure he grew up in Rochester MN
It's a vibe 💯
Again you two together??? AWESOME!
The Pre-hype Leaker Man x Post-launch Fallout Tester
Nice to see Fabio back.
Ya he is my favorite AMD graphics channel.
He's definitely helped me with undervolting and setting up Radeon cards
Bravo guys, best content when both of you are on.
Nah
people overthink things. All you need is a gpu that can play the games you like for a price you can afford. PC builder just did a poll and right now what people want in a gpu is "Most raw FPS for my budget" with eighty eight percent. Dlss has four percent and raytracing has three percent. With a minimum of fourteen thousand votes.
Raster is king and we'd buy cards that don't even support raytracing. If that gets the price down, why not make them?
See personally I'm pretty happy with my 3070 faster performance other than I wish it had more VRAM
I will upgrade for VRAM and RT performance.
@@TheDarksideFNothing sure but nvidia forces you buy the 4070 super TI to get 16 gigs. If you are cool with that, great. I'm not OK with 800 dollars for video games.
@@doublecrossedswine112 lol me neither.
I'm just saying I don't think I'll upgrade unless something with more VRAM and meaningfully better RT than what I currently have is available
@@TheDarksideFNothingI'm guessing you play a lot of single player games where RT is somewhat ok to enable. To me RT doesn't really make a meaningful difference in the feel of the game even when it's maxed out to make it worth the performance hit. The smoother gameplay just feels so much better without it.
@@toxicavenger6172I'm guessing you see bigger number go boom lol jk
Tom also wanted to look cool with glasses 😎
Fabio's back! His last time on the show is my favorite.
Transparencies are difficult to temporal upscale, since there’s only one set of motion vectors per pixel. Usually developers use the opaque geometry motion vectors and transparency doesn’t have motion vectors (it’s using motion vectors of the geometry behind the transparencies). FSR 3.1 doesn’t even support multiple sets of motion vectors. IIRC the other upscalers don’t support that either.
Similarly texture animation such as water surface is hard, since technically the water surface only moves up/down, while the ”visible” wave moves (along the surface). Same problem with all implicit surface techniques (distance fields and similar). Needs special effort from devs and still all they can achieve is a compromise.
I am not deeply familariar with the DLSS3 algorithm. It’s not open source like the others. But they could do AI based motion generation pass from the image to help in these cases. Kind of similar thing that HDTVs do with their pure rgb image (no motion vectors) based 120Hz image interpolation algorithms. But if you can’t get subpixel accurate accumulation, then you can’t properly reconstruct higher res image from multiple jittered low res images. It would just be blurry or blocky. Good quality motion vectors are crucial. AI based algo might be able to workaround it somehow.
Always remember, Frame Gen does NOT increase Performance.
We have to stop 🛑 saying this
It increases smoothness
@@AncientGameplays Hope you start feeling better soon from being sick. 🤧
@@Kapono5150 Much better today, releasing a video soon haha
@@AncientGameplaysisn't that better performance tho lol
@@wawaweewa9159 I guess he considers performance as better fps/better smoothness and better latency all together
I can't believe it's not butter!
Wait that's not the same Fabio.
This Fabio's tagline is "I can't believe it's not buffered!"
I want Fabio to do an in depth video on the effects of upscaling and frame generation on power efficiency.
this is a fire combo of underrated PC youtubers
Ancient Gameplays is always a very entertaining and informative guest.
Just came from Fabios latest video. 2 of my favorite guys. Much appreciated to you both.
RDNA3 doesn’t have separate tensor/matrix cores like CDNA3 and Nvidia do. It only has wave intrinsics (WDDM) for emitting 32 cycle optimal ALU sequence (with register routing optimizations to save power). Double ALU pipes vs RDNA2 help of course in matrix heavy workloads. It’s a big uplift over RDNA2 for AI work, but if RDNA4 is going to have CDNA3-style matrix cores, then AI perf improvement will be substantially bigger. Which likely means that AI based upscaler would not be backported.
Why not? XeSS performs OK on RDNA3 and even RDNA2. Why couldn't "FSR ML" do the same?
AMD could easily make an upscaler that has better image quality however it would also produce lower frame rates which defeats the entire purpose. An upscaled that uses most of the performance gain of a lower resolution just to upscale the image is useless. May as well just run the higher resolution with better image quality.. I would dare say that AMD could make an upscaled that produces dlss like quality but you would turn on quality upscaling and LOSE performance. Unusable.
This is precisely why NVIDIA doesn't allow DLSS 3.0 (Frame Generation) to be turned on previous generations of RTX (before) 4000.
It simply doesn't work as shown by hackers.
@@KarbinCryXeSS literally has 2 different modes, one using the matrix cores (and being better) and another not. If AMD is going to be using matrix cores for FSR then they wouldn’t be able to have it work on RDNA3 and older and Pascal and Older.
@@dex6316 I know XeSS has an XMX and DP4 path. Why couldn't "FSR ML" do the same though?
Another week, another Broken Silicon EPISODE🙌🔥
Glad to see Fabio on Moore's Law is Dead for a second time !!
fabio is awesome, congrats on the collaboration 😊
HEY Fabio! GPU Tech Jesus! I had to click since I saw Ancient Gameplay thumbnail. Been watching this channel for some time.
If Linus claims not having driver problems after x many months ago from switching to his 79XTX, AMD cards are good enough for me.
I'll switch now and then depending on what used deals are coming my way - so I have no bias
Linus is currently sponsored by AMD, but yeah their drivers have been good enough for years to call them stable. They'll still sometimes drop the ball like with Fallout New Vegas crashing on start-up, but their recent fix to that was quite fast.
I've used Radeon VII for years with maybe two driver issues.
Half a year ago I switched to a 4070 Ti and have already had several driver issues and frequent crashes.
Individual experiences are inherently more about luck.
@@louisfriend9323the FO3/FONV thing was intentional. To drive attention to the DXVK mod - just like 4GB patch, there's no reason not to use the DXVK mod.
as a life long Nvidia and EVGA customer, I am loving my 7900XTX , and with FSR3.1 able to use xess or fsr in many games with frame gen, I am having a blast! I got it for 999 when the 4080 regular was 1200 dollars. I refused to pay over 999 for a gpu... The 4070ti only had 12 gb ram for 800, SO I LEFT NVIDIA! For 200 dollars more , I got 24 gb 7900xtx and that is how AMD won me over. I have been really enjoying it. I just hope FSR 3.2 and 3.3 etc is coming soon.
i've run Nvidia and Radeon and the driver for both are good people tend to blame amd for any stability issue in their system when it could be for exemple their overclocked RAM running xmp.
LOVE THIS COMBO! Thanks for bringing us no BS hardware news❤
Excellent interview, with a lovely chilled vibe. Thank you both.
What a 💎 of an episode. Thank you both, keep it up!
Fabio, you can’t use frame generation if you’re running out of vram!
It doesn't help for sure lol
47:09 Thank you finally someone bringing up the PoE2 Illumination, when I saw the tech explanatiion I was just wow, amazing.
Good to see Fabio. Great guest.
I was waiting for this !!
Always a pleasure to have Fabio over. Great to see you, man!
Tom, I really love these episodes with Fabio. You should have him back on after rdna4 and ps5 pro. I’d love to see you guys circle back on the discussion after we see how those products work out + any future FSR improvements.
Ever since someone told me about this channel I have basically watched every single video here. Never change Tom 👌👌
Yeah man I stumbled across it one day like 6 months before alder lake came out and I don't think I've missed an episode since
@@TheDarksideFNothingI stumbled across a video with steam deck in the title of a video now I'm watching everything 😁
This was an amazing discussion to listen to. Great job and thank you to the both of you!
I want a 7800xt with nvidia ray tracing.... at $400.... is that so much to ask?
Depends on who you're asking? If it's GPU manufacturers, yes.
So , basically a next gen 8700xt ( hopefully )
Nice to see AncientGameplays on the channel. I found his content very helpful when transitioning from 1060 3gb to 6650xt 8gb in configuring adrenaline etc.
Love Fabio. So good to see another link up with a well respected YTer
Future FSR stands for Fidelity Sony Resolution 😆
It's like Alex from DF pointed out it feels like FSR doesn't do anti aliasing on moving pixels so it appears really blocky. You're looking at raw pixels. Maybe they're afraid of making it too blurry or something.
Also it's just really really hard lol
I imagine Nvidia is putting a lot more money into DLSS because they have a lot more money and also recognize what a win it is for them.
Hi, I am native. 😥
What you typed out makes sense as a sentence, but it doesn't actually make sense.
If you're seeing pixels(ignoring moire, something is wrong with your image)
Real footage off a proper camera won't be pixelated at even 1080p.
Making out pixels=/being pixelated
It makes sense, when pixels become disoccluded there is no historical/temporal samples/information available. The algorithm is going to emphasize the new samples and the output appears very undersampled.
It makes sense this would also occur in fast motion where the algorithm lowers the contribution of older and outdated information.
You need historical samples for temporal super sampling and super sampling provides anti-aliasing so that's why these scenarios look poor.
This may not occur in other upscalers if they are filtering or otherwise having this disoccluded information resolve differently.
If you don't discard historical information for disoccluded pixels you will get a smearing or ghosting effect.
I have noticed in general DLSS is more blurry than FSR but on the other end FSR is how you say a bit blocky and raw
VR. That's where we need more powerful GPUs. Modern VR headsets run around ~6000*3000pixels for their native resolutions and 90hz at minimum. And then you often still need some form of (non-temporal) anti-aliasing like MSAA on top of that to reduce shimmering and make things look sharp in the distance. Even the 4090 still struggles with a lot of titles and you have to make compromises and tweak settings (and of course really bad optimizations are a big part of that too).
Apart from that? Regular gaming with ~60fps in 1080p-1440p? 1080ti is still fine :D
DLSS uses memory as well, it's not at all a big VRAM saving feature
It's not supposed to save ram. It's supposed to add frames. All the "upscaling" stuff is just to create extra filler frames so things "feel smoother".
No one would imagine that dlss' use case was to save ram. Upscaling is creating filler frames for smoothness? Wouldn't that be.........................................................m Framegeneration?
On HU in Tim's defense he also briefly included a chart comparing the upscalers with same preset names. I guess he was covering his rear, anticipating the critique about comparing quality base on performance. And I am glad he did that from performance stance point given that almost all other content creators comparisons, did the test on matching quality names presets. So its nice to get Tims perspective.
Who cares, Tim breathlessly boosted wccftech's bizarre DLSS conspiracy theory
@@shanent5793 Yup it was really bizarre and Tim lost a lot of respect from me.
This was a great discussion, thoroughly enjoyed listening to it.
Loved the pod dude
Nice 2 of my top 3 computer hardware guys in one show. So much appreciated. Rock On!
From going to a 1070 to a 7900xtx, Ancient Gameplays helped me with going through the settings. 👍
I'd have to agree, with all these new settings available and going from 1080p to 1440p..... It can be quite difficult to set everything up to run smoothly and hopefully efficiently.
@ancientgameplays is one of my favorite channels, so it was great seeing you together! Great discussion. 👏
Upgraded from a 2070s to 7900 GRE red devi (my first amd card) and I'm happy with the change. Great 1440p performance, decent power usage, plays all my games on max to high on 1440p. Decided to stop sticking to nvidia for a change and amds pretty good. Still looking at what pricing models and performance will be for 50 series. Not sure what a lot of the hate towards AMD was about.
33:18 RDR2 has had 2.2 since march this year
My bad, didn't actually play the game for quite a while!
@@AncientGameplays Is understandable with amount of testing you do over all the different games
1080ti is going to carry me stoicly for another 2 years until rdna 5 9900xtx or better
I doubt we are getting rdna5 in 2 years if we are still 6 months away from RDNA4
@@WayStedYouwouldnt surprise me, if they cancelled high end RDNA4 to get RDNA5 out quicker
@@Dave-dh7rtthis. They don't really have a choice unless they want to throw the towel on competing above $500. Then again, AMDs favorite move is to do the opposite of what we all think is the obvious best move lol
Bury it in a lead box when it finally takes its last rest. We need a ceremony at the very least.
@@WayStedYou last time AMD didn't release a high end card we got their next gen after one year. (5700 XT in 2019, RDNA 2 in 2020)
Wuuut unexpected collab. Nice, Tom!
I'm part of that community. We went through hell on 7900xtx in some scenario's for over a year and the community was able to sanity check and workaround almost every issue.
41:25 But they did it because XeSS gets less fps. They tested the upscaling settings that give roughly the same fps boost. Which is perfectly fair. If XeSS needs to run at a lower setting to achieve the same fps as DLSS or FSR, then that lower setting needs to be compared.
100%. The universal version of xess has more of a performance impact which is why it has better quality(at the same internal resolution) than FSR. AMD could easily make an upscaled that is dlss like or even better but... It would lose performance instead of gaining it
Great to see Fabio back.. one fantatsic guy
Its Fabio.... Nice to see him back in broken silicon
1:25:40 100% agree with Fabio through this entire section. I've never seen a game that you can't get running well with settings unless your hardware is _seriously_ lacking. Don't get me wrong, vram is a real issue and 8gb is seeing its limits, but I've been seeing it get exaggerated worse and worse over the last year and it's mostly from the AMD fans trying to use it as the sole excuse to buy AMD.
That HUB video raised my eyebrows because they listed 2 games that I've personally played that their graphs were _absolutely_ exaggerated. It said R&C needed a MINIMUM 8.1gb of vram to even play at 1080p _medium_ even though it ran perfectly fine on my kids' 6gb card without any massive stutter, then claimed another needed 14gb that I've seen run perfectly fine on 12. I suspect he still kept texture resolution cranked through every test, or it's because he was using a 4090 and the game was allocating more than it actually needed when it was offered. I normally agree with HUB but this vram one seems to be a personal crusade for Steve where I'm starting to see numbers that I personally know just aren't accurate.
It only takes 5 seconds to pull up a YT clip and see his numbers are flat out wrong when this game shouldn't play on a 1660 super without massive stutters at medium, let alone high.
ua-cam.com/video/yAVXEGkmceE/v-deo.htmlsi=5qRXgDn0oC5uGMlo&t=92 Somehow this 6gb card isn't stuttering at settings HUB says needs 8.1 and 9.2gb. The card very clearly runs out of gas long before it runs out of vram.
i was happy to see anciengameplays get feature it is a nice channel who put lot of time testing.
Nice find for a good fsr vs dlss discussion
PSSR sounds like some kind of soviet socialist republic
Polish, Prussian, or Persian SSR
@@shanent5793Nailed it with Persian, if any country is gonna start PSSR, it's Iran.
AMD is Fabio's domain
The problem is that we don't have competition just duopoly cooperation. It was clear to see just after the crazy 80-class launch for $1200.
What AMD did do? They could do everything to take the market share but they decided to release the 4080's competitor (7900xtx) for $1000.
What should they do then? When they realized that they have only an 80-class competitor to price it correctly (~$750) and not blindly follow totally crazy Nvidia prices.
AMD is not a competitor just a blind follower. If Nvidia decides to release xx70 for $1000 then AMD will release a similar card for $900 and will call it a day. Like always.
This duopoly is killing this market and not only this one because without upgrading GPU people are not upgrading other parts like CPU, MB, RAM, and even monitors.
That's why my next gpu will be Intel, As much crap as people say about them only proves they are not voting with their wallets but blindly buying the duopoly option instead.
When all companies are based in US, there wont be a real competition.
@@jasonvors1922Yeah same, if Intel prices their battlemage parts reasonably and the performance is a solid generational jump then I'll absolutely grab one of those. They (the duopoly) desperately needs the type of shakeup Ryzen and later ARM have brought to the CPU space
We all know that both of the CEO company are cousins, so... yeah what do you expect
Didn't ever expect Fabio on this channel, while he's not exactly an "expert" on cpu and gpu tech, he's a great guest to have
I consider Fabio to be an expert on AMD GPU drivers!
He's more knowledgeable on certain things then your average tech UA-camr. Especially Radeon software.
At this point I wouldn’t say he’s not an expert, you’re bugging.
You mean like all those "experts" who take checks and free hardware from hardware makers?
@@TheMusicHeals.kjhjhhgExactly! I believe Fabio's opinions are more unbiased than those experts.
People who think 4090 are overpowered... havent played in VR lol. Try cyberpunk in VR. A 4090 isnt enough.
6800 for $360 is working great for me
don't miss:
MSI afterburner
NVCleanstall
NV control panel
Windows NVIDIA telemetry
fiddly, melty power connectors
running out of VRAM
spending $200-400 more on AI, CUDA, and sadly on this one note: RT
Nice to see the ancient gameplays co-op.
This is such a classic fanboy post. I love amd too.
Nice to see my man Tech Godfather sharing his wisdom here
Guys... you did not suck! It was a very good talk and all of those topics you talked about are interesting. Yes, the #1 requests from gamers is cheaper GPU prices. Also from Nvidia : more VRAM! From AMD : better FSR 3.x
good stuff here on frame gen. I play gta online with a 6700xt and a 5800x 3d. there are a ton of frame skips in crowded lobbies, especially with modders. turning on frame gen almost completely removes frame skips. I go from 90fps to 150-160 but the massively improved 1% lows are why I now keep it on. AMD won me over on the idea of frame gen with their execution. 💪
very proud of being a subcriber of this channel and AncientGameplays... magnificent interview as always...i am 50 years old and still playing games...AG makes best GPU drivers analisys out there....
Both of you are my favorite :)
One of the best guest I've heard on this channel, Fabio is interesting to listen to.
For the thumbnail alone i thought Jared Leto would be the guest 😂
On the FSR and DLSS issue of missing details. I find simple integer scaling to be a good alternative to AI upscaling. Setting to integer gpu scaling in NVCP works automatically if the game's resolution setting affects the output resolution, but if the game only changes the internal resolution you will also need to set desktop resolution to whatever lower resolution you want. I use 1080p integer scaled to 4K for demanding or old games with UI scaling issues and it looks really clean since it divides perfectly, which allows the screen to be filled in integer scaling mode. No artifacts, blurriness, missing details, etc. 720p for 1440p displays for the same effect.
2 of my favorite guys 🎉🎉
A Radeon Chill update for monitors over 300 Hz/FPS would be nice.
@56:00 The problem with talking about hardware accelerated upscaling on AMD's GPUs is that they actually need to have the silicon to do it.
They don't, haven't talked about it, and it goes entirely against their current strategy of using general compute resources.
AMD sectioned off CDNA from RDNA for a reason and I don't think they will want to reverse that decision any time soon.
So, if Sony is doing hardware accelerated upscaling, that's their strategy, their resources...
Basically, upscalers alter the creator's original intent. To this exception, we have Nvidia which claims that its DLSS rather embellishes the creator's original intent. IMO: stick with native creator intent, upscalers are gimmicky at best.
DLSS does, as it uses AI to rebuild textures when there isn't enough data from the source image. FSR on the other hand is more of a next generation anti aliasing and holds very true to the original image. DLSS is more detailed because it's adding data that isn't present in the original image, if you watch a side by side by side, FSR is much closer to the native image while DLSS is like a very minor remaster.
@@PineyJusticeRecently i watched a comparison between native path tracing and nvidia's ray reconstruction denoiser in an graphics editor. I was shocked about how much details have been lost with recent dlss still. We need way better denoisers than what the industry offers currently especially with the planed path tracing revolution.
@@aladdin8623 Upscaling will continue to get better, in another 4 years both FSR and DLSS will be be standard, totally replacing AA. The tech is one of the best things to come to tech in the last 20 years, eventually all cards will have hardware upscaling accelerators, just as we now have hardware AA accelerators. Upscaling gives hardware a generational boost in performance for very few downsides.
Upscalers are not gimmicky. That is a fact.
This is ridiculous. Changing resolution, graphics settings, or using a monitor different from the original creator would also be changing creators intent. Lol.
Ah Fabio again...awesome!!!
Great show today guys
RDNA 4 isn't going help AMD people want see high-end gpu's. As someone that owns a 7900xtx I wanted a high-end gpu that beats the one I have but since they won't be. I'm switching to the 5000 series
Atta boy Fabio.
You guys had good chemistry this time around
I learnt a few things, well done.
What can push the next gen hardware? High refresh rate, max quality Primax Crytstal running beautiful games!
A man who sells quality ramen is a man I trust. No sarcasm.
AMD should make using RSR easier in titles that don't natively support up-scaling. While setting up custom resolutions works, it's an inelegant workaround that most people wouldn't be bothered with. I wonder how many people even know this is possible in the Adrenalin software. I've used 1,224p to get better performance in 1440p titles that don't have native up-scaling, or were left with only FSR 1.0.
I am always surprised how good the driver upscaling from and is. If you turn it on in the driver then open the game, all you need to do is change the ingame resolution then AMD upscales it. Super easy. You want a 50-100% slider in the game?
@@christophermullins7163 If there were a slider in every game that would be amazing, but that's not gonna happen. I have found that using RSR to up-scale from 1080p to 1440p is too much of an image quality hit at 75%. However, 85% strikes a nice balance. To get 85% a custom resolution has to be created. At the very least it should be a present within Adrenalin, rather than having to create the configuration yourself.
Thank you for the podcast
Thanks for listening :)
Guys early dlss 2 is still better than fsr 3.1 especially 2.5 because thats where dlss start introduce preset c and further image quality improvement. You are delusional to think fsr 3.1 is better than dlss 2
Generally I'd say with FSR 2.2 at 4k, at least the quality setting usually looks quite good. Balanced has been worth it if you're not quite getting the frames, and then performance was noticeably worse but decent enough.
I haven't played any 3.1 games yet, but I really hope it gets back added to Avatar and Alan Wake 2. I really, REALLY hope people who know devs can request people go back and update their games for it. Alan Wake 2 in particular needs it.
Found Fabio’s channel back in 2020 after looking for under volt settings for a 5700XT. Worked perfect. Still have have the same file name for my settings on my GPUs. Ancienttune
RDNA4 is doa if they wait until q1 2025. At that point you just wait and buy Nvidia. If its a $500 7900xt performance card with less vram 6 months from now nobody cares. Wow waiting 6 months to save 100 to 200 dollars on a 7900xt with less vram.
I feel bad for anyone that bought the rx 7800xt it performs almost exactly the same as the rx 6800xt the rx 7900 gre preforms almost exactly the same as the rx 6950xt, This whole generation from amd has been a joke.
Fr fr lol, unless mvidia is selling 10 and 12 gb for 500
😂 kinda but the 5070 is gonna have 12gb 😂😂😂 in 2025
@@nicane-9966 except you don't know that
@@TV-xv1le 192 bits according to kopite so 12 or 24, do you think theyll put 24 on a 70? Delusional if you think so.
Just grabbed a 4090 for $1299, I think it was worth it. That’s the price it should have been anyway 😂
Hyped for this episode - two great content creator in one format!
Fabios AMD heavy content is really helpful since AMD was and still ist the only logical option in my region when it comes to price/performance. Learned allot about undervolting AMD GPU's from this guy!
Amd starting to do what 3dfx was doing in the 90s we was able to make drivers for quake 1,2,3 and Unreal good old days.
Intel iGPU drivers were ALWAYS awful. It just wasn’t as noticeable because most games wouldn’t run fast enough.
But ask any emulator developer about intel support on GPU plugins. They were always missing or bugged out on a BUNCH of features. Heck, Krita and Kodi, two 2D apps with GPU acceleration, were straight up broken back in the Skylake/3000 era iGPUs.
RDNA4 is gonna be a snooze unless they can deliver XTX performance for 500$, or 7900xt perf for 400$. If it’s a 16gb 7900xt for 500$, then they can keep it. Lol
On the rain thing for AMD I assume the error is in the alpha handling, with alphas there is pretty much straight and premultiplied and coders are not compositors. Getting the interpolation wrong and likely it needs to be set per game the cutout mattes can turn a bubble into a white sphere for example. So likely the work is getting done frame wise but the texture map on the plane is getting "choked" or clipped for most poeple.
Tom, this year is really good with content - I watch almost all. Asked Daniel Owen if he's joining but no response. He probably missed my comment.
I wouldn't mind Daniel on the show.
$2500 for a next gen card that’s only 30% faster than a 4090. Man has low expectations.
$2500 may be a bit high but 2080TI was around 30% faster than 1080TI but about 40% more expensive (MSRP). RT was not really a thing back then, so it seemed overpriced to me at first. However, thanks to RT cores and architecture it aged good and is more than 30% faster than 1080TI now (without RT of course).
Let's gooo. Fabio is great
What I want from next gen is 7900 XT performance or better with 16GB VRAM at $500 or less. If they can do that it would mark the first time since I bought my 5700 XT I thought the price to performance was at a good point. RDNA2 and Ampere *should* have been good price:perf, but of course then the scalpocalypse happened which also cemented stratospheric MSRPs.
I have a 6700 XT now because I cashed in on the 1:1 swap for a 5700 XT during the crypto boom, but the current gen offers so little price to performance I haven't even bothered looking for deals, I'm just not interested. But if I can get double the 6700 XT's performance with 16GB VRAM for $500 next gen I'll be all over it. I don't give a toss about ray tracing, 6 years since that hype train started and it's still a novelty add-on, maybe I might start caring in another half decade. And with appropriate raster performance why the hell would I need upscaling/frame gen?
@1:06:20 I'm not understanding what is meant by "AI" cores on RDNA 3.
As far as I understand it, each CU can split the wavefront so that two FP16 calculations can be performed instead of a single FP32 calculation.
That's just an accuracy thing, there's no "new" AI cores in RDNA 3 vs RDNA 2.
It's the same thing going from GCN/Vega to RDNA where they broke up the FP64 functionality and improved throughput at lower accuracy...
3 forward 1 back? Fsr 3.1? I guess it wasnt just an iteration
Hunt Showdown 2.0 is going to be amazing with a 5090 lol
Fabio loves to downplay the the 8GB VRAM issue. Nobody should be going anywhere near an 8GB GPU when upgrading.
He contradicts himself about it by saying "settings" and memory bandwidth + memory speed/latency but every modern 8GB GPU is shipping with 128-bit memory buses or less and the cache isnt doing the job.
Maybe the 7600, XT, and 4060 would all fair a lot better at 8GB if they had a much wider 256-bit memory bus, but those arent a thing anymore on lower/mid ranged cards.
What I stated it that they were fine for the price ($269 RX 7600 for example), guess I didn't explain myself correctly.
Also, you nowadays have narrower bus, but more cache
@@AncientGameplaysthe bus width actually affects how much ram that can even be added to the card doesn't it?
Fabio just overtook Steve from gamers Nexus for silky hair in livestream.
I'm hoping FSR would become standard in future just like freesync with custom vendor extensions per GPU vendors or game studios.
FSR 2.1 just had better playable frame rate on handheld and older GPU.
Don't know why nvidia ships lower vram and maybe it's time to add M.2 SSD slot to stream larger than memory textures, tensors and general compute just like Vega SSG.