GPU upgrade - buy 7900xtx or 4080 Super now or wait for next gen release in 2025?

[Zach_01]

And why not? The 4080 uses less power. It generates less heat. If you opt to overclock either, you'll see 7900 XTX at the 450-500W realm, 4080's won't even allow you to go so far, even on the high-end variants. Feel free to prove me otherwise (you can't).

4080: 320W
7900XTX: 355W

That’s 35W difference or +11%
I guess this amount of extra heat is big deal today.

Yes there are AIB 7900XTX cards up to 500W but soon you realize that past 380W the return is next to nothing.

Every AIB 4080 is fixed on 320W? No OC versions exist?

My card can be set from 314W up to 467W but I keep it at 375W

The issue with the 7900XTX is that it is literally 7% faster for $400 more than the 7900XT where I live.

I think it’s more than 7% and let’s not forget that most people that buying 7900XT/XTX are not running games at 1080p. At least 1440p and UW/4K.

Here prices now are…
7800XT from 500+€
4070 from 560+€
4070S from 630+€
7900GRE from 680+€
7900XT from 700+€
4070Ti from 800+€
4070Ti S from 870+€
7900XTX from 900+€
4080S from 1100+€
4080 from 1300+€

I guess that every market region has its own unique quirks.

 

[Dr. Dro]

I'm going to point out the obvious here, Hotfixes don't count as a plus towards driver support. You are in essence patting Nvidia on the back because they fixed something they screwed up on in the first place. Are you counting the hotfix for the discord Nvidia bug as equal to a full AMD driver release? That's beyond hypocritical, you are always criticizing AMD for their bugs but here you are counting Nvidia bugs and hotfixes as a plus towards driver support.

Similar could be said of studio drivers. Studio drivers are indentical to the game ready drivers. They are mearly more tested than the game ready drivers, are on a longer update cadence, and their release typically coincides with the release of creative app updates. It could more be argued that it's a downside that Nvidia has two separate packages for studio and game drivers as users must choose between which they want to install as compared to if Nvidia to just periodically release a LTS (long term support) version of their game ready driver (in fact they should just stop calling their main driver game ready, a lot of non-gamers use this driver version in different fields). Again you are trying to present something that's actually a limination as a positive and as if Nvidia is doing some massive amount of work on top of their regular driver updates to get studio drivers out the door when it's actually the opposite. I do think they are nice to have but it's entirely misleading to pretend they are equal to the work that goes into a full driver release.

All but 1 AMD driver in the last 2 years has been WHQL and you can see that right in the link I provided. Not the I really put much stock in Microsoft's ability to test drivers thoroughly but you brought it up.

You used stalker 2 as an example of AMD not have timely driver support but in fact AMD cards already perform as expected in that title: https://www.techpowerup.com/review/stalker-2-fps-performance-benchmark/5.html

View attachment 373374


So why exactly would AMD need to release a driver for this game?

For what I gathered, you are trying to just throw raw numbers behind your argument that Nvidia has superior driver support without thinking about the reasoning behind those numbers. Having more hotfix drivers is NOT a good thing and carries other implications, having studio drivers was not the flex you thought it was, and AMD not needing a driver for stalker 2 because it already performs well is not a negative but a plus.

I think even if we ignored all the above, just looking at the number of driver releases on the AMD side you can see your original statement in regards to consistent and timely driver support was simply false. AMD's driver releases have been pretty consistent for years. Sometimes they miss a month and often when that happens you can see they release two drivers in the following month, typically one early and one later in the month. Again it's 1-2 releases per month with the average being 1. I'm not sure I'd want to increase that cadence either, quality over quantity.

You do realize that validated ISV drivers are pretty much the sole reason enterprise GPUs have existed up until now, right? I don't know how you came up with the conclusion that having these on gaming cards, extensive support, access to an entire, maintained ecosystem and not having to wait for the next release cycle for known issues that have already been fixed are negatives, but you do you

Regarding performance: I must disagree: 10 fps below the 4080 in an UE5 game that runs on software lumen is nowhere even close to the expected performance level for a 7900 XTX, unless there have been massive strides in UE5 performance in the Nvidia side since these games were released:

4080: 320W
7900XTX: 355W

That’s 35W difference or +11%
I guess this amount of extra heat is big deal today.

Yes there are AIB 7900XTX cards up to 500W but soon you realize that past 380W the return is next to nothing.

Every AIB 4080 is fixed on 320W? No OC versions exist?

While many focus solely on the usefulness of upscaling to achieve otherwise impossible performance figures from a card, usage of DLSS often allows you another road: to slash GPU power without sacrificing much in image quality, which leads to a quieter, cooler and more efficient machine. You can often have games running at ~180 W at 4K/120 on a 4080 using DLSS Quaiity with Frame Generation enabled. FSR can (and in my opinion, should) be used to the same effect - but it helps that DLSS is generally ahead in IQ comparison tests.

My personal 4080 is a ROG Strix OC and one of the craziest models out there - 320 W is base TGP limit extensible to 420 W - I've never been able to get this card to push more than 350 W even at 3 GHz on air + VRAM clocked up to 4080S speed

 

[evernessince]

You do realize that validated ISV drivers are pretty much the sole reason enterprise GPUs have existed up until now, right? I don't know how you came up with the conclusion that having these on gaming cards, extensive support, access to an entire, maintained ecosystem and not having to wait for the next release cycle for known issues that have already been fixed are negatives, but you do you

Except only Nvidia's Quadro drivers and cards are certified by independent software vendors (ISVs): https://www.nvidia.com/en-us/design-visualization/software/isv-certifications/

Did you happen to get this answer from Google AI or ChatGPT without double checking the facts? I asked them both this question and they state that Studio drivers are ISV certified but ironically they link to the quadro page I provided above. When I asked ChatGPT to take another look, it revised it's answer to my answer, in that only Quadro drivers are ISV certified as it could not find any evidence that Studio drivers are ISV verified.

Rule number 1 of using AI, always double check it's output.

Regarding performance: I must disagree: 10 fps below the 4080 in an UE5 game that runs on software lumen is nowhere even close to the expected performance level for a 7900 XTX, unless there have been massive strides in UE5 performance in the Nvidia side since these games were released:

First, I never stated only in UE5 games or only those with software lumen. I just want to specify that as it appears you may be trying to change some goals posts to use in your next comment as a gatcha.

I'm going to give you a list of recent titles that were benched by TPU that demonstrate that the 7900 XTX's performance in Stalker 2 is within the normal range:

Star Wars Outlaws Performance Benchmark Review - 35 GPUs Tested

Star Wars Outlaws is a new title in the iconic Star Wars universe, which lets you play as member of the criminal underworld. The game uses the Snowdrop engine, and delivers great graphics. In our performance review, we'll look at the game's graphics quality, VRAM consumption, and how it runs...

www.techpowerup.com

Silent Hill 2 Remake Performance Benchmark Review - 35 GPUs Tested

Silent Hill 2 Remake lets you relive the iconic classic with modern graphics thanks to Unreal Engine 5. The PC version offers enhanced visuals, featuring support for cutting-edge technologies like DLSS, FSR and ray tracing. In our performance review, we'll look at the game's graphics quality...

www.techpowerup.com

Dragon Age: The Veilguard Performance Benchmark Review

Dragon Age: The Veilguard finally brings us back to the fantastic Dragon Age Universe. Built on Frostbite, the graphics are good, and there's support for RT and upscalers, too. In our performance review, we'll look at the game's graphics quality, VRAM consumption, and how it runs across a range...

www.techpowerup.com

God of War Ragnarök Performance Benchmark Review - 35 GPUs Tested

The long-awaited God of War Ragnarök is now available on PC, bringing the latest chapter of the epic action-adventure series to a whole new audience. The PC version offers enhanced visuals, featuring support for cutting-edge technologies like DLSS, FSR, and Frame Generation. In our performance...

www.techpowerup.com

Black Myth Wukong In-Game Performance Benchmark Review - 35 GPUs Tested

Black Myth Wukong has over two million gamers playing the game. This souls-like action RPG is a huge hit, especially in China. Powered by Unreal Engine 5, the graphics look great, but hardware requirements are definitely not on the light side, we had to use upscaling. In our performance review...

www.techpowerup.com

The 7900 XTX tends to run at or around 4080 performance. It'll only beat it in raster heavy games, as in the example you provided. Thus as I stated before, it's performing within expected margins in stalker 2. You could make the argument that it's a bit weak but it's not even close to an outlier. I could easily make the argument that your example is more an outlier, at least when we are only considering AAA releases over the current year.

Considering that you are trying to prove that AMD is being lazy by not releasing a driver for Stalker 2, you need to provide significant evidence that Stalker 2 is well outside normal performance margins. Something the data simply doesn't support.

 

[Dr. Dro]

Except only Nvidia's Quadro drivers and cards are certified by independent software vendors (ISVs): https://www.nvidia.com/en-us/design-visualization/software/isv-certifications/

Did you happen to get this answer from Google AI or ChatGPT without double checking the facts? I asked them both this question and they state that Studio drivers are ISV certified but ironically they link to the quadro page I provided above. When I asked ChatGPT to take another look, it revised it's answer to my answer, in that only Quadro drivers are ISV certified as it could not find any evidence that Studio drivers are ISV verified.

Rule number 1 of using AI, always double check it's output.




First, I never stated only in UE5 games or only those with software lumen. I just want to specify that as it appears you may be trying to change some goals posts to use in your next comment as a gatcha.

I'm going to give you a list of recent titles that were benched by TPU that demonstrate that the 7900 XTX's performance in Stalker 2 is within the normal range:

Star Wars Outlaws Performance Benchmark Review - 35 GPUs Tested

Star Wars Outlaws is a new title in the iconic Star Wars universe, which lets you play as member of the criminal underworld. The game uses the Snowdrop engine, and delivers great graphics. In our performance review, we'll look at the game's graphics quality, VRAM consumption, and how it runs...

www.techpowerup.com

Silent Hill 2 Remake Performance Benchmark Review - 35 GPUs Tested

Silent Hill 2 Remake lets you relive the iconic classic with modern graphics thanks to Unreal Engine 5. The PC version offers enhanced visuals, featuring support for cutting-edge technologies like DLSS, FSR and ray tracing. In our performance review, we'll look at the game's graphics quality...

www.techpowerup.com

Dragon Age: The Veilguard Performance Benchmark Review

Dragon Age: The Veilguard finally brings us back to the fantastic Dragon Age Universe. Built on Frostbite, the graphics are good, and there's support for RT and upscalers, too. In our performance review, we'll look at the game's graphics quality, VRAM consumption, and how it runs across a range...

www.techpowerup.com

God of War Ragnarök Performance Benchmark Review - 35 GPUs Tested

The long-awaited God of War Ragnarök is now available on PC, bringing the latest chapter of the epic action-adventure series to a whole new audience. The PC version offers enhanced visuals, featuring support for cutting-edge technologies like DLSS, FSR, and Frame Generation. In our performance...

www.techpowerup.com

Black Myth Wukong In-Game Performance Benchmark Review - 35 GPUs Tested

Black Myth Wukong has over two million gamers playing the game. This souls-like action RPG is a huge hit, especially in China. Powered by Unreal Engine 5, the graphics look great, but hardware requirements are definitely not on the light side, we had to use upscaling. In our performance review...

www.techpowerup.com

View attachment 373427

View attachment 373423
View attachment 373422
View attachment 373424
View attachment 373425


The 7900 XTX tends to run at or around 4080 performance. It'll only beat it in raster heavy games, as in the example you provided. Thus as I stated before, it's performing within expected margins in stalker 2. You could make the argument that it's a bit weak but it's not even close to an outlier. I could easily make the argument that your example is more an outlier, at least when we are only considering AAA releases over the current year.

Considering that you are trying to prove that AMD is being lazy by not releasing a driver for Stalker 2, you need to provide significant evidence that Stalker 2 is well outside normal performance margins. Something the data simply doesn't support.

I was specifically bringing this apparent performance regression in UE5 up (Wukong and now Stalker 2 are really not measuring up to the earlier UE5 games), but you seem to have found an even worse pattern, it seems that the performance on these newer games has fallen off a cliff? The XTX was never much of a competitor to the 4090 but it surely didn't look that bad next to the 4080 either... how is that a justification for any of the issues brought up? It's actually losing to the 3090 Ti in that many games? That's... terrible, even worse than I thought.

You accuse me of using ChatGPT (something I do not do and I will never do), but just something I thought you'd like to know? The Quadro branding was retired with Turing, half a decade ago. And I'd like to know what does AMD offer as an alternative to Omniverse, RTX Studio, Canvas, or even their LLM solution, all of which are fully available at no extra cost to GeForce customers today. In fact, let me take a few steps back, where's ROCm for consumer cards?

And I didn't say AMD was lazy. said that their product does not measure up to the competition. I know for a fact they aren't lazy. But there comes a time I want results instead of promises, talk of resetting the development roadmap and coming up with a reunified architecture isn't even a secret anymore. Look, writing a GPU driver is a complex thing. It's not easy. However, it's high time you people start demanding AMD to do what multi-billion-dollar companies their size ought to do; relying on eternal goodwill of a few boomers who have fond memories of their vintage ATi's on tech forums and the one-off Linux user is not a sound business strategy - as their latest earnings report clearly shows.

 

[_roman_]

Love to hear your thoughts.

I think it's a very bad point to buy now a graphic card.

For myself the graphic cards are already too long on the market. I suspect a new generation. I personally believe that buying a graphic card at release is the best when it is a high end card, like a Nvidia 4080. (first few week after release - when the price has calmed down)

Advise: Buy if you need a card / Buy if you see a decent offer / Wait if you can wait (which most likely means another 18 months)

Just a bit off topic - as i saw linux above me:
Regarding Linux and AMD. The amd graphic cards suck. Amd does not force me to recompile the kernel module - nvidia does. But same as nvidia - no proper support for advanced features - like undervolting, frequency control or fan curve control. I checked it several times in the past with current linux-firmware and current kernel sources

 

[Dr. Dro]

I think it's a very bad point to buy now a graphic card.

For myself the graphic cards are already too long on the market. I suspect a new generation. I personally believe that buying a graphic card at release is the best when it is a high end card, like a Nvidia 4080. (first few week after release - when the price has calmed down)

Advise: Buy if you need a card / Buy if you see a decent offer / Wait if you can wait (which most likely means another 18 months)

Just a bit off topic - as i saw linux above me:
Regarding Linux and AMD. The amd graphic cards suck. Amd does not force me to recompile the kernel module - nvidia does. But same as nvidia - no proper support for advanced features - like undervolting, frequency control or fan curve control. I checked it several times in the past with current linux-firmware and current kernel sources

I believe you have to set up the coolbits thing for the hardware controls to work

NVIDIA/Tips and tricks - ArchWiki

Green with Envy for GUI control similar to Afterburner on Windows

Roberto Leinardi / GreenWithEnvy · GitLab

System utility designed to provide information, control the fans and overclock your NVIDIA card

gitlab.com

It's been a while I last tried it. That said, I never got any of my NV cards working well with Linux, especially if they're the current generation, while I did have my previous AMD cards working well there.

 

[evernessince]

I was specifically bringing this apparent performance regression in UE5 up (Wukong and now Stalker 2 are really not measuring up to the earlier UE5 games), but you seem to have found an even worse pattern, it seems that the performance on these newer games has fallen off a cliff? The XTX was never much of a competitor to the 4090 but it surely didn't look that bad next to the 4080 either... how is that a justification for any of the issues brought up? It's actually losing to the 3090 Ti in that many games? That's... terrible, even worse than I thought.

The cost of RT effects has increased and this disproportionately impacts AMD. This is why you see the performance gap between the chart you provided, which is of a more raster heavy game, and the charts I provided.

Raster heavy game? 7900 XTX tends to lead, sometimes by a decent margin.
Software Lumen / Light RT? 7900 XTX tends to be around or slightly behind 4080
Heavy RT? 7900 XTX falls to around 4070 Ti performance.

And yes it's bad. I mentioned this market trend in a review awhile back but it stands to further weaken an already weak AMD GPU division.

You accuse me of using ChatGPT (something I do not do and I will never do), but just something I thought you'd like to know?

That was not an accusal but a genuine question given the odd coincidence.

The Quadro branding was retired with Turing, half a decade ago. And I'd like to know what does AMD offer as an alternative to Omniverse, RTX Studio, Canvas, or even their LLM solution, all of which are fully available at no extra cost to GeForce customers today. In fact, let me take a few steps back, where's ROCm for consumer cards?

This has zero relevance to driver release consistency, which is the subject matter of contention. I already stated that I agree Nvidia has the superior feature set so it's pointless to bring up.

And I didn't say AMD was lazy. said that their product does not measure up to the competition.

Yes and of which we already agreed on except for two points. You stated driver consistency was poor and I demonstrated otherwise.

I know for a fact they aren't lazy. But there comes a time I want results instead of promises, talk of resetting the development roadmap and coming up with a reunified architecture isn't even a secret anymore. Look, writing a GPU driver is a complex thing. It's not easy. However, it's high time you people start demanding AMD to do what multi-billion-dollar companies their size ought to do; relying on eternal goodwill of a few boomers who have fond memories of their vintage ATi's on tech forums and the one-off Linux user is not a sound business strategy - as their latest earnings report clearly shows.

This is nice boilerplate marketing speak but nothing of value is stated. Mind you, you are calling them lazy here. That's the obvious implication when you say they aren't doing enough or do nothing but talk and promises. It's a round about way to say it without having to address anything that's actually being discussed. It's like when anyone says "no offense but", my eyes roll over when the platitudes come out.

My question is, how does this relate to your statement that AMD has poor driver consistency and your argument that AMD needs to release a driver for stalker 2?

 

[eidairaman1]

The cost of RT effects has increased and this disproportionately impacts AMD. This is why you see the performance gap between the chart you provided, which is of a more raster heavy game, and the charts I provided.

Raster heavy game? 7900 XTX tends to lead, sometimes by a decent margin.
Software Lumen / Light RT? 7900 XTX tends to be around or slightly behind 4080
Heavy RT? 7900 XTX falls to around 4070 Ti performance.

And yes it's bad. I mentioned this market trend in a review awhile back but it stands to further weaken an already weak AMD GPU division.



That was not an accusal but a genuine question given the odd coincidence.



This has zero relevance to driver release consistency, which is the subject matter of contention. I already stated that I agree Nvidia has the superior feature set so it's pointless to bring up.



Yes and of which we already agreed on except for two points. You stated driver consistency was poor and I demonstrated otherwise.



This is nice boilerplate marketing speak but nothing of value is stated. Mind you, you are calling them lazy here. That's the obvious implication when you say they aren't doing enough or do nothing but talk and promises. It's a round about way to say it without having to address anything that's actually being discussed. It's like when anyone says "no offense but", my eyes roll over when the platitudes come out.

My question is, how does this relate to your statement that AMD has poor driver consistency and your argument that AMD needs to release a driver for stalker 2?

This should prove a point, yes it's GCN but when unnecessary proprietary cruft is removed the truth comes out.

And having to use AI to enhance graphics is laziness, and a pc should be able to go offgrid/air gapped and have the same graphics capability as if it was on the net. AI is bandwitch hogging.

 

[Redwoodz]

I wouldn't count on any price drop on 4090's. In fact you can expect shortages on all NV top tier cards for the next couple of years atleast. Also I wouldn't expect much uplift in price/performance for the 5090. Gaming is not what these cards are made for anymore.

 

[Dr. Dro]

This should prove a point, yes it's GCN but when unnecessary proprietary cruft is removed the truth comes out.

And having to use AI to enhance graphics is laziness, and a pc should be able to go offgrid/air gapped and have the same graphics capability as if it was on the net. AI is bandwitch hogging.

Internet access is not required to use AI upscalers and frame generators such as DLSS, FSR 2/3 or XeSS, you may run these completely offline with no loss of functionality.

NVIDIA and Intel GPUs have something called matrix multiplication units, and it is in this portion of the hardware that the "AI calculations" to make these upscalers possible take place. AMD is the only vendor which does not have them; which is the true reason why FSR is "vendor-agnostic", it's just shader-level code and thus would run on anything that can run the graphics API in question. Matrix multiplication functionality is not new or even remotely exclusive technology, dedicated tensor processing was introduced with Volta architecture in 2017 and Intel introduced this with ARC-branded graphics (Alchemist/A770).

To AMD's credit, while Navi 31 remains the only current high-end hardware which does not have dedicated tensor processing, they did introduce some very useful extensions for processing this type of workload with RDNA 3, although these still take the GPU's shader resources to execute. This is why something like FSR 3 is viable with high performance on this hardware. Thankfully, since AMD open sourced their implementation, it's quite easy for you to see how it works:

FidelityFX Super Resolution 3.1.4 (FSR3) - Upscaler

gpuopen.com

I wouldn't count on any price drop on 4090's. In fact you can expect shortages on all NV top tier cards for the next couple of years atleast. Also I wouldn't expect much uplift in price/performance for the 5090. Gaming is not what these cards are made for anymore.

I'm fairly sure it's gonna be 70-100% faster than the 4090, so 2-2.5x faster than the 4080 or 7900 XTX. Price/performance or not, 4K gamers are gonna buy them as fast as they can make those. Can't say I'm gonna resist. If they don't meet this target, then the 5090 might very well be a pointless product from the get go unless you mean to buy them for LLMs or something.

My question is, how does this relate to your statement that AMD has poor driver consistency and your argument that AMD needs to release a driver for stalker 2?

It's simple, man. AMD tends to implement only the strictly required API features. Nothing novel or off the beaten path. Even with the advent of the PAL-based UMDs, it seems they're endlessly stuck in a rut. New features take forever to be implemented, and when they are, they often don't work. Then they spend the lifetime of a generation fixing things, often in games that are at that point years old. Just 24.10.1 was targeting fixes for crashes and stability issues on games like DayZ and Doom Eternal, with open problems on Space Marine 2, known VR crashes, etc.

https://www.amd.com/en/resources/support-articles/release-notes/RN-RAD-WIN-24-10-1.html

Rome wasn't built in a day, but if they were laying only a brick each day, the Coliseum wouldn't be finished to this day...

 

[Kapone33]

Internet access is not required to use AI upscalers and frame generators such as DLSS, FSR 2/3 or XeSS, you may run these completely offline with no loss of functionality.

NVIDIA and Intel GPUs have something called matrix multiplication units, and it is in this portion of the hardware that the "AI calculations" to make these upscalers possible take place. AMD is the only vendor which does not have them; which is the true reason why FSR is "vendor-agnostic", it's just shader-level code and thus would run on anything that can run the graphics API in question. Matrix multiplication functionality is not new or even remotely exclusive technology, dedicated tensor processing was introduced with Volta architecture in 2017 and Intel introduced this with ARC-branded graphics (Alchemist/A770).

To AMD's credit, while Navi 31 remains the only current high-end hardware which does not have dedicated tensor processing, they did introduce some very useful extensions for processing this type of workload with RDNA 3, although these still take the GPU's shader resources to execute. This is why something like FSR 3 is viable with high performance on this hardware. Thankfully, since AMD open sourced their implementation, it's quite easy for you to see how it works:

FidelityFX Super Resolution 3.1.4 (FSR3) - Upscaler

gpuopen.com

I'm fairly sure it's gonna be 70-100% faster than the 4090, so 2-2.5x faster than the 4080 or 7900 XTX. Price/performance or not, 4K gamers are gonna buy them as fast as they can make those. Can't say I'm gonna resist. If they don't meet this target, then the 5090 might very well be a pointless product from the get go unless you mean to buy them for LLMs or something.



It's simple, man. AMD tends to implement only the strictly required API features. Nothing novel or off the beaten path. Even with the advent of the PAL-based UMDs, it seems they're endlessly stuck in a rut. New features take forever to be implemented, and when they are, they often don't work. Then they spend the lifetime of a generation fixing things, often in games that are at that point years old. Just 24.10.1 was targeting fixes for crashes and stability issues on games like DayZ and Doom Eternal, with open problems on Space Marine 2, known VR crashes, etc.

https://www.amd.com/en/resources/support-articles/release-notes/RN-RAD-WIN-24-10-1.html

Rome wasn't built in a day, but if they were laying only a brick each day, the Coliseum wouldn't be finished to this day...

Do you ever stop? Every time someone proves you wrong you move the Goalposts with some other narrative. What does the Nvidia data say I wonder?

Fixed Gaming Bugs

• DSR/DLDSR custom resolutions may not appear in certain games [4839770]
• [Call of Duty MWIII] filename change preventing users from using GFE Freestyle Filters [4927183]

Fixed General Bugs

• [Bluestacks/Corsair iCUE] May display higher than normal CPU usage [4895184][4893446]
• When "Shader Cache size" is set to "disabled" cache files may still be created [4895217]

 

[Dr. Dro]

Do you ever stop? Every time someone proves you wrong you move the Goalposts with some other narrative. What does the Nvidia data say I wonder?

Fixed Gaming Bugs

• DSR/DLDSR custom resolutions may not appear in certain games [4839770]
• [Call of Duty MWIII] filename change preventing users from using GFE Freestyle Filters [4927183]

Fixed General Bugs

• [Bluestacks/Corsair iCUE] May display higher than normal CPU usage [4895184][4893446]
• When "Shader Cache size" is set to "disabled" cache files may still be created [4895217]

I don't stop, because while I am offending your feelings, I am not lying, nor did I ever say that Nvidia has no bugs. it's just that they tend to be on the insignificant side, more often than not.

And before you rush to your knightly defense of AMD, read what you post (and the replies to my own post, mind you, I wasn't "proven wrong", if anything, there's a point or two of disagreement there with mostly it being cordial back and forth), perhaps. There was never a bug in question regarding MW3 (this is not the 2011 game they are talking about - it's this year's reboot) - Activision developers changed a file's name in their game which then required them to release an update so the software continued to work. Perfectly reasonable and such changes are to be fully expected in an always-online, GaaS title. Or do I need to remind you of the recent CS2 VAC incident?

And wow, what a grave showstopper of a problem that DSR resolutions hadn't shown in a couple of games, because clearly every piece of software out there works the same way and requests resolution lists in the exact same way... also another great whopper of a major bug, shader cache files being generated but not used with the cache off (even though, in most cases, you shouldn't ever turn it off). Wow! High severity issues.

 

[evernessince]

This should prove a point, yes it's GCN but when unnecessary proprietary cruft is removed the truth comes out.

And having to use AI to enhance graphics is laziness, and a pc should be able to go offgrid/air gapped and have the same graphics capability as if it was on the net. AI is bandwitch hogging.

Unfortunately proprietary junk has become the norm now and Nvidia dictates the implementation of so much in games. It not only makes it harder for AMD to compete but it's severely locks out any other competitors as well (Intel and a potential ARM competitor for example). There's just way too much software lock-in nowadays. AMD could come back in the CPU market because those CPUs could utilize all the features of existing software. How in the world does anyone do that in the GPU market though? It's nearly impossible from the decades of proprietary features Nvidia has baked into the ecosystem.

Internet access is not required to use AI upscalers and frame generators such as DLSS, FSR 2/3 or XeSS, you may run these completely offline with no loss of functionality.

Correct. The only instance you'd need internet for AI is something like Microsoft's AI or the one Skyrim mod that lets you talk to NPCs. I really really hope we don't start seeing games that do require internet for games.

To AMD's credit, while Navi 31 remains the only current high-end hardware which does not have dedicated tensor processing, they did introduce some very useful extensions for processing this type of workload with RDNA 3

You are probably thinking of WMMA that increases matrix multiplication 2x (at most) over the prior gen. Not as good as dedicated units, hence the 7900 XTX gets about half the performance of the 4090 is most AI tasks.

It's simple, man. AMD tends to implement only the strictly required API features. Nothing novel or off the beaten path. Even with the advent of the PAL-based UMDs, it seems they're endlessly stuck in a rut. New features take forever to be implemented, and when they are, they often don't work. Then they spend the lifetime of a generation fixing things, often in games that are at that point years old. Just 24.10.1 was targeting fixes for crashes and stability issues on games like DayZ and Doom Eternal, with open problems on Space Marine 2, known VR crashes, etc.

Strictly speaking in regards to the last 8-10 years I agree with you. If we are talking old, ATI / AMD both AMD and Nvidia used to take turns releasing exciting new features.

And yeah, AMD has been stuck in a rut. Their GPU division doesen't make a lot of sales or money which in turn makes devs unlikely to spend much resources on optimizing for them and makes it unlikely that AMD will invest a significant amount of money closing that gap when the returns aren't there. I mean just look at Intel, they made huge strides in the driver department but it hasn't paid off for them sales wise. Intel cards can't use any of the proprietary software features in any game because they've haven't been in the dGPU market until recently. There is litterally decades of inertia behind Nvidia in terms of both mindshare and software.

 

[Dr. Dro]

Unfortunately proprietary junk has become the norm now and Nvidia dictates the implementation of so much in games. It not only makes it harder for AMD to compete but it's severely locks out any other competitors as well (Intel and a potential ARM competitor for example). There's just way too much software lock-in nowadays. AMD could come back in the CPU market because those CPUs could utilize all the features of existing software. How in the world does anyone do that in the GPU market though? It's nearly impossible from the decades of proprietary features Nvidia has baked into the ecosystem.

Correct. The only instance you'd need internet for AI is something like Microsoft's AI or the one Skyrim mod that lets you talk to NPCs. I really really hope we don't start seeing games that do require internet for games.

You are probably thinking of WMMA that increases matrix multiplication 2x (at most) over the prior gen. Not as good as dedicated units, hence the 7900 XTX gets about half the performance of the 4090 is most AI tasks.

Strictly speaking in regards to the last 8-10 years I agree with you. If we are talking old, ATI / AMD both AMD and Nvidia used to take turns releasing exciting new features.

And yeah, AMD has been stuck in a rut. Their GPU division doesen't make a lot of sales or money which in turn makes devs unlikely to spend much resources on optimizing for them and makes it unlikely that AMD will invest a significant amount of money closing that gap when the returns aren't there. I mean just look at Intel, they made huge strides in the driver department but it hasn't paid off for them sales wise. Intel cards can't use any of the proprietary software features in any game because they've haven't been in the dGPU market until recently. There is litterally decades of inertia behind Nvidia in terms of both mindshare and software.

And there you have it, we are in full agreement!

 

[Kapone33]

I don't stop, because while I am offending your feelings, I am not lying, nor did I ever say that Nvidia has no bugs. it's just that they tend to be on the insignificant side, more often than not.

And before you rush to your knightly defense of AMD, read what you post (and the replies to my own post, mind you, I wasn't "proven wrong", if anything, there's a point or two of disagreement there with mostly it being cordial back and forth), perhaps. There was never a bug in question regarding MW3 (this is not the 2011 game they are talking about - it's this year's reboot) - Activision developers changed a file's name in their game which then required them to release an update so the software continued to work. Perfectly reasonable and such changes are to be fully expected in an always-online, GaaS title. Or do I need to remind you of the recent CS2 VAC incident?

And wow, what a grave showstopper of a problem that DSR resolutions hadn't shown in a couple of games, because clearly every piece of software out there works the same way and requests resolution lists in the exact same way... also another great whopper of a major bug, shader cache files being generated but not used with the cache off (even though, in most cases, you shouldn't ever turn it off). Wow! High severity issues.

Knightly. Or you shouting out your Bias. You act like nothing AMD does is appreciated. Are you not the one that has said that your experience with Linux is not that good for you? Meanwhile AMD being an advocate of Open source means those handhelds are super smooth. It is obvious that even though they are the same GPU cores the narrative handles APUs much differently. Someone posts a video about RX580 vs 1060 and you respond with Upscaling tech that the 1060 does not even support. It is almost insane how the narrative tries to make AMD cards useless even though Gaming benchmarks on this very site belie that very narrative. I don't blame you though. I have been investigating for myself and have posted evidence of how the narrative is applied and you can find it in chat comments of some big Youtubers like Robeytech, PC World, MSI Gaming and Hardware Unboxed. You will go on though as this narrative will have people pay 5k where I live for the 5090. Don't worry you can expect tariffs to add another 25% to that too. I am of the firm opinion that unless money is no issue that spending over 1k on a GPU is daft but the kicker is that already today the 7900XTX is in some scenarios 25% the cost of the 4090 brand new. Is it worth that much to have tech that the next generation of card is going to have some feature to make your 5k brick obsolete? Does the narrative even mention the 3090?

Look at how you got triggered because I posted Nviidia bugs from their own site.

 

[_roman_]

Are you not the one that has said that your experience with Linux is not that good for you? Meanwhile AMD being an advocate of Open source means those handhelds are super smooth.

No! Please do not write nonsense.

why?

AMD open source support is not worth mentioning. There is no gui for creating fan curves, undervolting, underclocking. High clock memory bug - and loud fans with the 6800 card - both in windows 11 pro and gnu gentoo linux. This topic is endless why amd has bad software in my current windows 11 pro or my gnu gentoo linux.

Go ahead and tell us how long you are using the linux kernel with gnu userspace? How many distros you used?
I still have my first installer Slackware 96. Just for explanation that Stands for 1996.

Before you write, I expect from you your current screenshot of your current gnu linux setup in the screenshot thread in this forum. + validation that its you like time + date + your forum name. (when you check my posts you will see several i3wm.org screenshot of my current setup + also one in the mentioned screenshot thread)

This is always with my current gnu gentoo linux, which dates back to 2006. (my forums.gentoo.org account which is basically the installation date)
I used nvidia graphic card for several months again in 2023 - a msi 960 gtx 4gb - second hand - to determinet the current driver status. AFter i had sold 2 or 3 years before my gaming laptop asus g75vw.
As mentioned sold ASUS G75VW laptop I had several, also duplicate, AMD processors: 5800X -> 3700X downgrade -> another fresh purchased 5800X -> second hand 3100x (as i was selling hte mainboard and 5800x) mainboard faulty intel wifi - 5800x never did boost -> current ryzen 7600X (i think i did not bougt better because four different 5800x processors on three different mainboard never boosted I expected. 7600x is cheap garbage - so less expected functionality or less promised functionality from amd does not really annoy me that much - i do compile a lot - i see if marketing is bullshit or not)
gpu: asrock challenger 6600XT 8GB D -> Nvidia 960GTX -> nothing - ryzen 7600X graphics -> Msi Radeon 6800 Z Trio -> current Powercolor 7800XT hellhoud (a good one in regards of beeing very silent)
On my socket am4 days i had 3 different mainboards to try + several different cpu cooler solutions.

Edit: Mainboards - a very nice topic. AMD Mainboards wiht B550 or X670. AMD - I blame amd for it - novuton does not release specs for the chip. You can barely read out any mainboards. or barely any laptops in gnu linux. the topic is endless about linux + amd + bad support. And that is just looking at my hardware. There is so much hardware available on the market.

edit: compiler support - next big topic - the optimisations suck for amd processors especially the ryzen 5800x and the ryzen 7600x at the point of time i owned those - amd does nothing - really nothing -

 

[Ruru]

-Is there enough benefit with pcie 5.0 GPU's upcoming?

I highly doubt. With 4090, you practically can't tell that are you running it @ 3.0 x8 or @ 4.0 x16.

NVIDIA GeForce RTX 4090 PCI-Express Scaling

The new NVIDIA GeForce RTX 4090 is a graphics card powerhouse, but what happens when you run it on a PCI-Express 4.0 x8 bus? In our mini-review we've also tested various PCI-Express 3.0, 2.0 and 1.1 configs to get a feel for how FPS scales with bandwidth.

www.techpowerup.com

 

[TSiAhmat]

I'm fairly sure it's gonna be 70-100% faster than the 4090, so 2-2.5x faster than the 4080 or 7900 XTX. Price/performance or not, 4K gamers are gonna buy them as fast as they can make those. Can't say I'm gonna resist. If they don't meet this target, then the 5090 might very well be a pointless product from the get go unless you mean to buy them for LLMs or something.

Wait... 70 - 100 % faster? Isn't it the same node and ""only"" 150 more W? (33% increase in W).

Therefore a increase in frames per watt (because of the architecture) of around 27 % ~ 50 %

(I am sure I am missing/wrong about something here...)

Also that GPU will probably cost around 2700 $ ~ 3000 $ if it lives up to your expectations...

 

[AnomalouS]

I wouldn't count on any price drop on 4090's. In fact you can expect shortages on all NV top tier cards for the next couple of years atleast. Also I wouldn't expect much uplift in price/performance for the 5090. Gaming is not what these cards are made for anymore.

I am with you on those thoughts regarding pricing, I am right on the fence for a 4080S or 5080, 90's are just out of control and I won't spend that kind of money.

 

[eidairaman1]

Unfortunately proprietary junk has become the norm now and Nvidia dictates the implementation of so much in games. It not only makes it harder for AMD to compete but it's severely locks out any other competitors as well (Intel and a potential ARM competitor for example). There's just way too much software lock-in nowadays. AMD could come back in the CPU market because those CPUs could utilize all the features of existing software. How in the world does anyone do that in the GPU market though? It's nearly impossible from the decades of proprietary features Nvidia has baked into the ecosystem.



Correct. The only instance you'd need internet for AI is something like Microsoft's AI or the one Skyrim mod that lets you talk to NPCs. I really really hope we don't start seeing games that do require internet for games.



You are probably thinking of WMMA that increases matrix multiplication 2x (at most) over the prior gen. Not as good as dedicated units, hence the 7900 XTX gets about half the performance of the 4090 is most AI tasks.



Strictly speaking in regards to the last 8-10 years I agree with you. If we are talking old, ATI / AMD both AMD and Nvidia used to take turns releasing exciting new features.

And yeah, AMD has been stuck in a rut. Their GPU division doesen't make a lot of sales or money which in turn makes devs unlikely to spend much resources on optimizing for them and makes it unlikely that AMD will invest a significant amount of money closing that gap when the returns aren't there. I mean just look at Intel, they made huge strides in the driver department but it hasn't paid off for them sales wise. Intel cards can't use any of the proprietary software features in any game because they've haven't been in the dGPU market until recently. There is litterally decades of inertia behind Nvidia in terms of both mindshare and software.

Intel was in dgpu a long time ago with the i740 and it was a flop then larrabee was a failure to launch and it seems the igpu still haunts them. But anyways i got off the nv plantation a long time ago especially after they had no vista support for NF2.

 

[Dr. Dro]

Wait... 70 - 100 % faster? Isn't it the same node and ""only"" 150 more W? (33% increase in W).

Therefore a increase in frames per watt (because of the architecture) of around 27 % ~ 50 %

(I am sure I am missing/wrong about something here...)

Also that GPU will probably cost around 2700 $ ~ 3000 $ if it lives up to your expectations...

The GB202 is said to be exceptionally large. We'll have to wait and see but I am expecting those figures, remember AD102 in the 4090 is cut down (128 of 144 SMs with 72 of 96 MB of L2 enabled), it uses the slowest variant of GDDR6X used in Ada generation cards while GB202 is rumored to be 512-bit GDDR7...

It's all speculating of course but that has been consistently the target, with all improvements cumulated and some of the rumors aligned, I certainly see it happening. What you are absolutely right is that it probably will be very expensive.

We are going to be knowing as soon as January for sure, though. W1zz is already preparing his benchmark suite for it (saw this thread asking for what games we want benched?) so expect news Soon(TM)

 

[TSiAhmat]

The GB202 is said to be exceptionally large. We'll have to wait and see but I am expecting those figures, remember AD102 in the 4090 is cut down (128 of 144 SMs with 72 of 96 MB of L2 enabled), it uses the slowest variant of GDDR6X used in Ada generation cards while GB202 is rumored to be 512-bit GDDR7...

It's all speculating of course but that has been consistently the target, with all improvements cumulated and some of the rumors aligned, I certainly see it happening. What you are absolutely right is that it probably will be very expensive.

We are going to be knowing as soon as January for sure, though. W1zz is already preparing his benchmark suite for it (saw this thread asking for what games we want benched?) so expect news Soon(TM)

that is ludicrous... my whole prebuilt PC was cheaper in the chip crisis era than that card at launch (if it really is 2700 $ or higher)

i know that thing will curb stomp my 3080 10GB, but... . . . . .

Well at least these GPU hold their values like iPhones (which is bad for those who don't have a lot of disposable income)


Anyways, maybe you can get a good deal on a used 4080S/4090 Card when Balckwell comes out?
(Ebay GPUs in my area are more expensive than new, its ridiculous looked it up again and i was wrong, 4090 on ebay, for an example, are 100 ~ 150 € less expensive)

(Edit: Also thanks for explaining what I missed/wrong about)

 

[Kapone33]

No! Please do not write nonsense.

why?

AMD open source support is not worth mentioning. There is no gui for creating fan curves, undervolting, underclocking. High clock memory bug - and loud fans with the 6800 card - both in windows 11 pro and gnu gentoo linux. This topic is endless why amd has bad software in my current windows 11 pro or my gnu gentoo linux.

Go ahead and tell us how long you are using the linux kernel with gnu userspace? How many distros you used?
I still have my first installer Slackware 96. Just for explanation that Stands for 1996.

Before you write, I expect from you your current screenshot of your current gnu linux setup in the screenshot thread in this forum. + validation that its you like time + date + your forum name. (when you check my posts you will see several i3wm.org screenshot of my current setup + also one in the mentioned screenshot thread)

This is always with my current gnu gentoo linux, which dates back to 2006. (my forums.gentoo.org account which is basically the installation date)
I used nvidia graphic card for several months again in 2023 - a msi 960 gtx 4gb - second hand - to determinet the current driver status. AFter i had sold 2 or 3 years before my gaming laptop asus g75vw.
As mentioned sold ASUS G75VW laptop I had several, also duplicate, AMD processors: 5800X -> 3700X downgrade -> another fresh purchased 5800X -> second hand 3100x (as i was selling hte mainboard and 5800x) mainboard faulty intel wifi - 5800x never did boost -> current ryzen 7600X (i think i did not bougt better because four different 5800x processors on three different mainboard never boosted I expected. 7600x is cheap garbage - so less expected functionality or less promised functionality from amd does not really annoy me that much - i do compile a lot - i see if marketing is bullshit or not)
gpu: asrock challenger 6600XT 8GB D -> Nvidia 960GTX -> nothing - ryzen 7600X graphics -> Msi Radeon 6800 Z Trio -> current Powercolor 7800XT hellhoud (a good one in regards of beeing very silent)
On my socket am4 days i had 3 different mainboards to try + several different cpu cooler solutions.

Edit: Mainboards - a very nice topic. AMD Mainboards wiht B550 or X670. AMD - I blame amd for it - novuton does not release specs for the chip. You can barely read out any mainboards. or barely any laptops in gnu linux. the topic is endless about linux + amd + bad support. And that is just looking at my hardware. There is so much hardware available on the market.

edit: compiler support - next big topic - the optimisations suck for amd processors especially the ryzen 5800x and the ryzen 7600x at the point of time i owned those - amd does nothing - really nothing -

Do you know what AMD software is? I can see you have also been triggered. I could write a laundry list of Intel problems but all I have to say is none of those CPUs lost performance as the silicone melted from too much voltage but that would be juvenile.

What does the Steam Deck run on? Is it not a Linux based distro? Is that OS not designed to run on AMD hardware?

Well I use my AM4 chips for CPU mining and they are doing just fine.

 

[Dr. Dro]

that is ludicrous... my whole prebuilt PC was cheaper in the chip crisis era than that card at launch (if it really is 2700 $ or higher)

i know that thing will curb stomp my 3080 10GB, but... . . . . .

Well at least these GPU hold their values like iPhones (which is bad for those who don't have a lot of disposable income)


Anyways, maybe you can get a good deal on a used 4080S/4090 Card when Balckwell comes out?
(Ebay GPUs in my area are more expensive than new, its ridiculous)

I'm personally holding out hope it'll cost more or less the same as the 4090 does today on average (that means, up to $1800-2000 for the basic models), I just hope it doesn't exceed that value, I can't afford anything above that.

Do you know what AMD software is? I can see you have also been triggered. I could write a laundry list of Intel problems but all I have to say is none of those CPUs lost performance as the silicone melted from too much voltage but that would be juvenile.

What does the Steam Deck run on? Is it not a Linux based distro? Is that OS not designed to run on AMD hardware?

Well I use my AM4 chips for CPU mining and they are doing just fine.

View attachment 373674

You haven't even read his post... Jesus, man. You try to say your defense is justified and not knightly, and then you pull that one. Slow down. We get it, you have an inflamed, passionate love for AMD, but fighting everyone you see speak negatively of their products on forums isn't gonna keep them afloat, demanding change and improvement will. If not, they deserve to close up shop.

If no one in that multi-billion-dollar international conglomerate ever stopped to sit down and look at what makes its competition successful and at least try to follow in the same vein, perhaps they actually deserve to.

 

[Kapone33]

I'm personally holding out hope it'll cost more or less the same as the 4090 does today on average (that means, up to $1800-2000 for the basic models), I just hope it doesn't exceed that value, I can't afford anything above that.



You haven't even read his post... Jesus, man. You try to say your defense is justified and not knightly, and then you pull that one. Slow down. We get it, you have an inflamed, passionate love for AMD, but fighting everyone you see speak negatively of their products on forums isn't gonna keep them afloat, demanding change and improvement will. If not, they deserve to close up shop.

If no one in that multi-billion-dollar international conglomerate ever stopped to sit down and look at what makes its competition successful and at least try to follow in the same vein, perhaps they actually deserve to.

"AMD open source support is not worth mentioning. There is no gui for creating fan curves, undervolting, underclocking. High clock memory bug - and loud fans with the 6800 card - both in windows 11 pro and gnu gentoo linux. This topic is endless why amd has bad software in my current windows 11 pro or my gnu gentoo linux."