GPU prices — is the worst behind us ?

noko said:
lol, depends on the game. The Last Of Us Part 2 later Nvidia drivers is giving a water bug where the animations stop. The 5700 XT is plowing through that game over 110 to 150 FPS, FSR 4.1.1 Quality looking gorgeous in 4k. With FG over 240 FPS. Since a Sony title, it works extremely well with AMD cards.
Click to expand...
I had heard about that title. It is funny how Nvidia didn't make that a priority to fix. With 2,704 current players that might not be enough players for them to bother fixing it. They could fix it if they really wanted to.
 
DaedalusHelios said:
I had heard about that title. It is funny how Nvidia didn't make that a priority to fix. With 2,704 current players that might not be enough players for them to bother fixing it. They could fix it if they really wanted to.
Click to expand...
10% of those players are having a smooth experience while 90% are having less than.
 
Lakados said:
I know people say that, but if that's the case, then no GPU maker currently has any business building gaming GPUs.
Nvidia's gaming division still made $11.40 billion in 2025, compared to AMD, which made $34.64 billion across the entire company.

So RTX GPUs alone make 1/3 of the second-place company's total revenue.
11.4 billion isn't a small number, and who in the market could absorb it if Nvidia just peaced out?

So yeah, Gaming may only be 6% of Nvidia's current revenue, which is absolutely insane.... But gaming currently accounts for roughly 7% of AMD's revenue, so by that measure AMD has no more of a business reason to continue with making gaming-related products either.

Sadly, or maybe not sadly. It means the gaming industry needs to shift to fit the product availability. Which I appreciate; maybe it means optimization becomes a thing again, and developers are forced to get the absolute most out of the available hardware. Which, ironically, will be done by AI agents, because hand optimization is a lost art and too expensive to manage at the scale of modern games.
Click to expand...

I’m not saying there is zero business case for making gaming GPUs in a normal environment for Nvidia, but there is certainly zero business case for doing it when they’re supply-locked on the significantly more profitable datacenter side. If supply isn’t an issue then yes, I agree with everything you said. The problem is Nvidia can’t get enough access to chips to satisfy the demand for datacenter computing as it is, so redirecting things like VRAM over to sell gaming GPUs for a fraction of the profit doesn’t make sense. Why sell it to gamers when some datacenter guy will give you quadruple the price for it?

Bear in mind that Nvidia’s $2000 “gaming” 5090 is going for closer to $10,000 in some cases. This isn’t coming from gamers. It means that the general compute people have moved on to the next best thing since they can’t get enough of everything else.
 

“AMD GPUs Get Price Hike in China with International Increases To Follow​

by Cpt.Jank Today, 12:48 Discuss (1 Comment)
As if GPU prices weren't high enough, thanks to the ongoing RAMpocalypse driving up BOM costs continuously for what seems like the last year, the trend is clear. According to recent reports, it now seems as though AMD GPUs are slated to get even more price hikes in the near future. As reported by VideoCardz, based on a source on Board Channels, AMD has increased GPU vendor pricing by as much as $104 in China.

Reportedly, the AMD Radeon RX 9070 XT and RX6500 XT pricing will remain unchanged, but the RX 9070 will get a price hike equivalent to $104, being the steepest price increase on the list, while the RX 9060 XT 8 GB will get the smallest price increase, at $45. It is said that these price hikes are motivated by higher memory prices and manufacturing costs, and that, while they are seemingly local to China for the time being, international board partners have also been informed of the upcoming changes.“
 
HockeyJon said:
I’m not saying there is zero business case for making gaming GPUs in a normal environment for Nvidia, but there is certainly zero business case for doing it when they’re supply-locked on the significantly more profitable datacenter side. If supply isn’t an issue then yes, I agree with everything you said. The problem is Nvidia can’t get enough access to chips to satisfy the demand for datacenter computing as it is, so redirecting things like VRAM over to sell gaming GPUs for a fraction of the profit doesn’t make sense. Why sell it to gamers when some datacenter guy will give you quadruple the price for it?

Bear in mind that Nvidia’s $2000 “gaming” 5090 is going for closer to $10,000 in some cases. This isn’t coming from gamers. It means that the general compute people have moved on to the next best thing since they can’t get enough of everything else.
Click to expand...
Everything is supply-locked right now; it's not getting immediately better any time soon.
I hope that Nvidia's new NVHBM module makes things better; it moves a physically large part of the GPU die out of the GPU die and onto the memory module itself.
That means cheaper GPU dies, and because of the process type used for memory modules, it does not significantly change the cost of the memory module, but it does result in 30% more bandwidth with a lower power draw.

Pair that with the newer AI-based compression algorithms, and we could reasonably see VRAM requirements decrease by as much as 80%.

Smaller dies with significantly less memory would make them affordable again. It sounds like a step backwards, but it just may be the only feasible path forward.

Because the supply crunch is only going to get worse.
 
Since both Nvidia and AMD have gone total "darkside" with regards to the consumer, I don't think we'll get meaningful stats from either with regards to consumer sales.

Another rare endangered AMD user. Meet and greets (with limited photo ops) available at reasonable prices.
 
cjcox said:
Since both Nvidia and AMD have gone total "darkside" with regards to the consumer, I don't think we'll get meaningful stats from either with regards to consumer sales.

Another rare endangered AMD user. Meet and greets (with limited photo ops) available at reasonable prices.
Click to expand...
Sales data is required for publicly traded companies via SEC requirements. Are you referring to monthly instead of annual and quarterly fiscal updates?
Regulation FD(Full Disclosure) and FINRA Rule 2210 or Reg BI apply to retail sales. We get the data either way on publicly traded companies. If they were a private company they could hide that data but creditors, IRS, and discovery in sue cases can "open up the books". For large private companies discovery in sue cases is typically filtered through a "special master" appointed by the court that will limit the scope of information.
 
Last edited:
DaedalusHelios said:
Sales data is required for publicly traded companies via SEC requirements. Are you referring to monthly instead of annual and quarterly fiscal updates?
Regulation FD(Full Disclosure) and FINRA Rule 2210 or Reg BI apply to retail sales. We get the data either way on publicly traded companies. If they were a private company they could hide that data but creditors, IRS, and discovery in sue cases can "open up the books". For large private companies discovery in sue cases is typically filtered through a "special master" appointed by the court that will limit the scope of information.
Click to expand...
Sales data doesn't have to "meaningful". It can be grouped however they want to shape how they want to be perceived. And no, you can't just "get the details, just because". Go ahead and try though.
 
cjcox said:
Sales data doesn't have to "meaningful". It can be grouped however they want to shape how they want to be perceived. And no, you can't just "get the details, just because". Go ahead and try though.
Click to expand...
Yes, businesses can be optimistic with projections and group segments how they choose to see things. You said, "Go ahead and try though."

AMD Q2 ending August 27th, 2026
What we were talking about:
Client and Gaming segment: Combined revenue was $3.8 billion, up 6% year-over-year. Within this segment, Client business revenue was $3.1 billion (up 23% YoY), while Gaming revenue was $779 million (down 31% YoY).
What other data would you like to see? That is enough to know from an investment side if looking at that market segment. How should they disclose quarterly data that would meet your standards? You came off as really bitter, and if it is authentic and not just angst, you should elaborate your point. Smugness is not an argument.
 
Last edited:
~8% of the market wants an AMD GPU, the numbers are no secret, but I am sure the excuses will come fast:
NVIDIA.png


And it has been like that for more than a year.
 
littlegreenmenfromhome said:
~8% of the market wants an AMD GPU, the numbers are no secret, but I am sure the excuses will come fast:
View attachment 830270

And it has been like that for more than a year.
Click to expand...
Nvidia not only has better marketing, but the DLSS integration in so many titles, plus the AI components in future games make AMD unattractive to most. We need real competition in that market segment to get better products and performance per dollar. VRAM shortages plus low competition is pricing people out of gaming. Even consoles are becoming less attainable due to the price increases. AMD giving up on making high end cards was the writing on the wall. There is no real successor to the 7900XTX unless you are counting ray tracing only with games on the 9070 XT. I am stuck with RTX 5080 cards on my primary rig due to the multiple ounces of gold price range of the RTX 5090. It is not justifiable to purchase the 5090 for gaming in my opinion.
 
Last edited:
DaedalusHelios said:
Nvidia not only has better marketing, but the DLSS integration in so many titles, plus the AI components in future games make AMD unattractive to most. We need real competition in that market segment to get better products and performance per dollar. VRAM shortages plus low competition is pricing people out of gaming. Even consoles are becoming less attainable due to the price increases. AMD giving up on making high end cards was the writing on the wall. There is no real successor to the 7900XTX unless you are counting ray tracing only with games on the 9070 XT. I am stuck with RTX 5080 cards on my primary rig due to the multiple ounces of gold price range of the RTX 5090. It is not justifiable to purchase the 5090 for gaming in my opinion.
Click to expand...
Why would a devleoper use a lot of effort implementing stuff for 8% of users?
It is easy to see why they want to do it for NVIDIA, when they have 90% of sales and an ever inscreasing marketshare.

And AMD only got them self to thank for their shrinking marketshare, they missed the train on DXR, so late infact that that people (wrongly) call raytracing for RTX.
They also missed the train on upscaling, so DLSS has become a de facto standard.
And with DLSS 5 we have seen that NVIDIA keep out-feature and out performing AMD.
I mean modders even hack DLSS5 to run on AMD cards, meaning AMD's is living more and more on the grace of NVIDIA.
Which also mean you can safely ignore all the "nay sayers" and "Raytracing doesn't matter" users, because the data shows that their are horrible wrong.

8% of sales over the last 1½ year also show the disconnect between forums and buying consumers.
When you read forums you get the impression that half the users are on AMD (and Linux too) when infact their a the small minority, crying VERY loud.
So keep that in mind when reading posts here, it is more and more the last die hard AMD fans buying AMD GPU's and nobody else.

But the price increases have nothing to do with NIVIDIA's 90% market sales to do.
It is simply that Enterprise (and thus A.I.) have far deeper pockets and a lot more money to spend.
Infact NVIDIA's marketshare in Enterprise, datacenter and A.I are even higher than 90%.

The moment Intel launches a CPU that mathces AMD, AMD is in deep trouble.
Their GPU revenue is shrinking.
Consoles are LOW revenue products.
Only thing keeping AMD alive right now is their CPU's.
 
littlegreenmenfromhome said:
Why would a devleoper use a lot of effort implementing stuff for 8% of users?
It is easy to see why they want to do it for NVIDIA, when they have 90% of sales and an ever inscreasing marketshare.
Click to expand...
You thought I was saying that developers should do that? I wasn't and I don't see anyone making that argument.
 
Marees said:
  1. Sony PlayStation
  2. Microsoft DirectX 13 play anywhere
basically any console first game. for example Starfield
Click to expand...
But that doesn't translate into the PC space as we can see on the data.
There is very little "console" effect as the console are kinda weak machines, Microsoft havn't released any FRS 4 on consoels either so that point is moot and nither is Sony's PSSR nor QSSR availbe on PC that that argument is false to.
Add to that that most console games use vert lite RT because they lack the horsepower to run any but lite stuff and you arguments falls flat on it face.

Any console title is getting more/better features on the PC and still AMD's features are trailing behind.
Your whole argument sounds like wishful thinking and not reality.
 
littlegreenmenfromhome said:
But that doesn't translate into the PC space as we can see on the data.
There is very little "console" effect as the console are kinda weak machines, Microsoft havn't released any FRS 4 on consoels either so that point is moot and nither is Sony's PSSR nor QSSR availbe on PC that that argument is false to.
Add to that that most console games use vert lite RT because they lack the horsepower to run any but lite stuff and you arguments falls flat on it face.

Any console title is getting more/better features on the PC and still AMD's features are trailing behind.
Your whole argument sounds like wishful thinking and not reality.
Click to expand...
AMD is not a software company (unlike Nvidia — which has more software engineers than hardware engineers). you can't expect new software features from AMD. it will be more of a Sony initiative or a DX13 feature implemented by AMD

If you want something extra in terms of features you will have to pay the Nvidia tax for those features
 
Marees said:
AMD is not a software company (unlike Nvidia — which has more software engineers than hardware engineers). you can't expect new software features from AMD. it will be more of a Sony initiative or a DX13 feature implemented by AMD

If you want something extra in terms of features you will have to pay the Nvidia tax for those features
Click to expand...
It is not really a "tax" when you get higher perfomance with better image quality, quite the opposite it is AMD being the inferior product and thus forced to sell cheaper...and still only get an 8% market sales.
NVIDIA performance > AMD performance
DLSS > FSR
Ray Reconstruction > Ray Regeneration
NVIDIA FG > AMD FG
DLSS 5 > [AMD has Nothing]
RTX Remix > [AMD has Nothing]
CUDA > ROCm
NVIDIA Broadcast AI> [AMD has Nothing]
OptiX > OpenCL/HIP
NVENC > AMD Encoder
NVIDIA Reflex > AMD anti-lag
NVIDIA Studio Ecosystem > [AMD has Nothing]

I mean, my 2022 RTX 4090 beats anything AMD has still in 2026.

There is a reason for the 8% of sales: AMD is the budget brand with less performance and less features.

There is no NVIDIA TAX, only an AMD discount due to lack for performance/features.
 
littlegreenmenfromhome said:
It is not really a "tax" when you get higher perfomance with better image quality, quite the opposite it is AMD being the inferior product and thus forced to sell cheaper...and still only get an 8% market sales.
NVIDIA performance > AMD performance
DLSS > FSR
Ray Reconstruction > Ray Regeneration
NVIDIA FG > AMD FG
DLSS 5 > [AMD has Nothing]
RTX Remix > [AMD has Nothing]
CUDA > ROCm
NVIDIA Broadcast AI> [AMD has Nothing]
OptiX > OpenCL/HIP
NVENC > AMD Encoder
NVIDIA Reflex > AMD anti-lag
NVIDIA Studio Ecosystem > [AMD has Nothing]

I mean, my 2022 RTX 4090 beats anything AMD has still in 2026.

There is a reason for the 8% of sales: AMD is the budget brand with less performance and less features.

There is no NVIDIA TAX, only an AMD discount due to lack for performance/features.
Click to expand...
Yep before all the rtx stuff amd was viable because being a bit slower didn't matter as much to most gamers. But most pc gamers are probably also geeks that like playing with the new thing and you can't say it isn't somewhat exciting to try new dlss models or other new rtx things.

The flip side to that is now you basically have to go Nvidia and because of that your old 4090 is Nvidia biggest competitor so the next gen add some new cool ai rtx stack that you can't use unless you upgrade. Also they put melting plugs on them now..... We really are in a crappy situation with the whole state of the industry
 
littlegreenmenfromhome said:
It is not really a "tax" when you get higher perfomance with better image quality, quite the opposite it is AMD being the inferior product and thus forced to sell cheaper...and still only get an 8% market sales.
NVIDIA performance > AMD performance
DLSS > FSR
Ray Reconstruction > Ray Regeneration
NVIDIA FG > AMD FG
DLSS 5 > [AMD has Nothing]
RTX Remix > [AMD has Nothing]
CUDA > ROCm
NVIDIA Broadcast AI> [AMD has Nothing]
OptiX > OpenCL/HIP
NVENC > AMD Encoder
NVIDIA Reflex > AMD anti-lag
NVIDIA Studio Ecosystem > [AMD has Nothing]

I mean, my 2022 RTX 4090 beats anything AMD has still in 2026.

There is a reason for the 8% of sales: AMD is the budget brand with less performance and less features.

There is no NVIDIA TAX, only an AMD discount due to lack for performance/features.
Click to expand...
This right here.

If AMD were just slightly slower like they used to be and charged the appropriate price for that, the landscape would be much like it was a few years ago. Today, the gap between them has probably never been wider.
 
1_rick said:
I haven't paid that close attention recently but I thought the 9000 series did a lot better in ray tracing and stuff.
Click to expand...
Don't get me wrong. AMD has improved its ray tracing since the 7000 series. However, it's still well behind NVIDIA. I think AMD's about 20% slower than "equivalent" cards at it instead of being half as fast like it was in the previous generation.
 
1_rick said:
I haven't paid that close attention recently but I thought the 9000 series did a lot better in ray tracing and stuff.
Click to expand...
yes really good, they are in regular ray tracing Nvidia equivalent now in the marketing product stack if not transistor count,
https://www.techpowerup.com/review/sapphire-radeon-rx-9070-gre-pulse/34.html

with the 9070 xt performing close to the 5070ti, 9060xt close to the 5060ti, pathtracing is more where we still see some gap, they have shader execution reordering, powerful tensor core to denoise the very noisy work fast, independant hardware for BVH, big L2 cache get useful here (AMD patents and some leak indicate they will go toward that Nvidia cache strategy for RDNA 5, big unified L2 instead of L1->L2->L3) .
 
Many path tracing implementations are optimized for Nvidia’s ecosystem, partly due to sponsorship and technical support, making it difficult for AMD to achieve comparable visual fidelity and performance.
 
1_rick said:
I haven't paid that close attention recently but I thought the 9000 series did a lot better in ray tracing and stuff.
Click to expand...
More or less competitive in their class as Luke mentions. Looks like someone would have to spend more for worse performance from Nvidia with the current pricing.
 
LukeTbk said:
yes really good, they are in regular ray tracing Nvidia equivalent now in the marketing product stack if not transistor count,
https://www.techpowerup.com/review/sapphire-radeon-rx-9070-gre-pulse/34.html

with the 9070 xt performing close to the 5070ti, 9060xt close to the 5060ti, pathtracing is more where we still see some gap, they have shader execution reordering, powerful tensor core to denoise the very noisy work fast, independant hardware for BVH, big L2 cache get useful here (AMD patents and some leak indicate they will go toward that Nvidia cache strategy for RDNA 5, big unified L2 instead of L1->L2->L3) .
Click to expand...
Eh, I was off about 10%. AMD is still solidly behind NVIDIA when it comes to Ray Tracing and Path Tracing. That said, at the current insane pricing, the price favors AMD in that market segment. A quick search shows a $300-$400 difference to jump up to the 5070 Ti.
 
Prince Valiant said:
More or less competitive in their class as Luke mentions. Looks like someone would have to spend more for worse performance from Nvidia with the current pricing.
Click to expand...
Well, the other thing is if you don't care about DLSS et al, you're probably better off with AMD right now. My local Micro Center looks to have 40+ 9070XTs for around $800, but the 5070 Ti starts at 1150. (I used a quick search for the closest comparable nvidia card to the 9070xt in raster) and that's what it suggested. There's a couple of 5070s as cheap as $850 but most are $870 or more.

Given I don't like upscaling in particular, if I had to buy a card today it'd be the Radeon. Although honestly I'd probably get a 9070, not the XT.
 
For lowest prices on eBay (with legit sellers):

RX 9070 16GB = $649.99 (although one was on sale on Amazon Prime Day for $629.99)
RTX 5070 12GB = $749.99
RX 9070 XT 16GB = $769.99
RTX 5070 Ti 16GB = $1,119.95

In pure raster, and overclocked at stock voltages, the RX 9070 XT 16GB is on par with a RTX 5070 Ti 16GB. It really depends on the RTX 5070 Ti's overclocking ability and the game. The RX 9070 16GB isn't far behind. The RTX 5070 12GB trails noticeably, but is still capable at 4K 60FPS in certain games.

For all these cards, pathtracing is going to really lower performance. They same with raytracing. It's best to select one or two raytracing elements if you're really wanting it with pure raster performance. However, if you want all the bells and whistles, you'll have to rely on DLSS or FSR (which I personally don't prefer).

There's a few things to disable for the NVIDIA app, and quite a bit more on the AMD Adrenalin software, but both are okay once you know what you're doing.

The RX 9070 16GB is a no-brainer, and people keep overlooking it. The only graphics card aside from the RTX 5090 and RTX 4090 that'll handle raytracing with 60FPS performance at native 2K is the RTX 5080. Forget it if you're going to try it with the four cards I listed. Pathtracing, forget about it at native 2K, it'll be close to unplayable for most games.

IMO, raytracing and pathtracing is a gimmick, and I much prefer an OLED or mini-LED monitor over either of them, and especially over any DLSS or FSR.
 
littlegreenmenfromhome said:
It is not really a "tax" when you get higher perfomance with better image quality, quite the opposite it is AMD being the inferior product and thus forced to sell cheaper...and still only get an 8% market sales.
NVIDIA performance > AMD performance
DLSS > FSR
Ray Reconstruction > Ray Regeneration
NVIDIA FG > AMD FG
DLSS 5 > [AMD has Nothing]
RTX Remix > [AMD has Nothing]
CUDA > ROCm
NVIDIA Broadcast AI> [AMD has Nothing]
OptiX > OpenCL/HIP
NVENC > AMD Encoder
NVIDIA Reflex > AMD anti-lag
NVIDIA Studio Ecosystem > [AMD has Nothing]

I mean, my 2022 RTX 4090 beats anything AMD has still in 2026.

There is a reason for the 8% of sales: AMD is the budget brand with less performance and less features.

There is no NVIDIA TAX, only an AMD discount due to lack for performance/features.
Click to expand...

You missed or i guess it would fall under cuda that AMD has ZERO way of doing any 3d rendering with Octane/Redshift/Arnold/Karma/Renderman or Vray. Like NONE of the major players support AMD. NONE OF THEM. The only thing a amd card is good for is games. It literally can't do anything else. The world runs on Cuda.
 
With my 5090 sold (over 3x what I paid for the 5080+5070 ti combined), much happier gaming wise. It's been a scorcher of a fall here (100 degrees) and the 5090 is just too hot. Funny enough my Corsair H100 AIO no longer goes crazy when gaming (something unique to the 5090's heat output, even the 4090 didn't do this). Still have to swap the ancient Antec 1100 to the H7 flow sitting in the box and swap the H100 for the CM Elite 360 some day...one reason I kept it so long is the side fans on the Antec 1100 blow directly onto the jinxed 16 pin GPU connector.
 
Epyon said:
You missed or i guess it would fall under cuda that AMD has ZERO way of doing any 3d rendering with Octane/Redshift/Arnold/Karma/Renderman or Vray. Like NONE of the major players support AMD. NONE OF THEM. The only thing a amd card is good for is games. It literally can't do anything else. The world runs on Cuda.
Click to expand...
Is that still true in 2026 ?

I think vray has an HIP version now:
https://documentation.chaos.com/space/VMAX/113587140/V-Ray+GPU+Setup
https://www.pugetsystems.com/labs/a...g-performance-is-radeon-a-viable-alternative/

Redshift has well, they are still so far behind that the support must be limited work too.
 
undertaker2k8 said:
With my 5090 sold (over 3x what I paid for the 5080+5070 ti combined), much happier gaming wise. It's been a scorcher of a fall here (100 degrees) and the 5090 is just too hot. Funny enough my Corsair H100 AIO no longer goes crazy when gaming (something unique to the 5090's heat output, even the 4090 didn't do this). Still have to swap the ancient Antec 1100 to the H7 flow sitting in the box and swap the H100 for the CM Elite 360 some day...one reason I kept it so long is the side fans on the Antec 1100 blow directly onto the jinxed 16 pin GPU connector.
Click to expand...
The difference between a 5090 and everything else is pretty substantial. I'll take the performance it offers over anything else. (Granted, I paid MSRP for my card.)
 
Dan_D said:
The difference between a 5090 and everything else is pretty substantial. I'll take the performance it offers over anything else. (Granted, I paid MSRP for my card.)
Click to expand...
No question the performance is phenomenal but it's also a room heater. If you're making money from using it, it's still worthwhile but I run most of my AI workflows on our enterprise accounts and Bedrock. Too bad I only captured the upside on one card, sold a tuf 5090 and a FE here basically at cost last year and actually returned a Zotac to MC (MSRP one)...I never learn...Halo GPUs are unobtanium forever.
 
You must log in or register to reply here.
Back
Top