• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

Top AMD RDNA4 Part Could Offer RX 7900 XTX Performance at Half its Price and Lower Power

They haven't abandoned it they simply can't get multi GCD's working correctly if the rumours are to be believed. So this gives them room to get the following generation working correctly rather than wasting resources fire fighting. It boils down to Radeon group not having the resources it needed and being forced to make hard choices but thankfully with AI it appears that's likely to change but if that helps or hinders GPU development who knows.
Don't know if they will implement it or not in their consumer cards, but their Instinct MI300X is already having 2xGCD and it's working and is being treated by software as a single GPU. I know that there is a difference between that arch and RDNA, but still seems they are making good progress on that.

If they could make a GPU with 2GCD, each performing as 7900XTX, with MCDs on them, improve RT perf then I would sell my 7900XTX and buy it instantly.

EDIT
Did a read up on that Instinct card:


It uses XCDs and has eight of them and they are all exposed as a single GPU.
Question is does it convert to consumer market and a GPU used for gaming, we'll see I guess.
 
Last edited:
It uses XCDs and has eight of them and they are all exposed as a single GPU.
Question is does it convert to consumer market and a GPU used for gaming, we'll see I guess.
IIRC the reason it's difficult to have multiple GCDs in consumer GPUs for graphics applications is that those are exceedingly latency-sensitive to avoid poor minimum FPS and to get even frame pacing. Adding a chip-to-chip interconnect inevitable adds latency to shader operations that span multiple GCDs.

Making compute GPGPUs with multiple compute dies is easy because the criteria are different - finish the single massive task as fast as possible - no need to worry about how evenly it speeds through that task, so higher inter-die latency is acceptable if the additional die(s) add enough performance to overcome the latency downsides. Presumably the latency increase of moving to MCD is a one-off cost, so two dies is the worst-case scenario and scaling up the number of dies beyond two adds all of the compute benefits with no further latency penalties.
 
7900xtx level of performance for half the price is decent but it would have been nice to have an option for something faster from AMD camp nonetheless.
 
If AMD can turn this into going really hard after the $150-$400 price range, they could have a winner on their hands. Sure, 2% will be upset that they aren't going after the ultra high-end. But looking at Steam, the top 20 cards are made up of a lot of GF 10 series, 2060, 3050, 3060. They make up 31.11% of the market. 20% of them can't do DLSS 3.0 and they basically all aren't worth running RT on. A lot of these are likely vendor supplied (Dell, HP, Lenovo) but if this can help AMD penetrate back into these markets as well.

If we can get 7900XTX RT performance as is at $400, on man that a fantastic RT card. Performance is all relative to price. If the 7900XTX was priced at $1,500 it would be an absolute terrible card. If it was $200, there is no point in literally buying anything else. Sure, there might be some higher margins in something like a 4090, but when it makes up less than 1% of the market. If you can secure volume, volume will be more profitable. And I am sure it'll help a lot to have more cards out there then less cards.
 
It's also not like you can escape RT as we're already seeing games where RT is default or where raster has broken lightning because it was developed with RT in mind. Hell, even one of the latest AMD sponsored games has mandatory RT.
In this year of 2024 I am playing the first game I have ever played with ray tracing. Metro Exodus Enhanced Edition is the only game in my steam library at this time that even has ray tracing as an option. I have no intent to play any other game with ray tracing any time soon. It is actually rather easy to escape RT if you aren't playing the latest AAA games. By the time I get around to playing most RT games, they will almost certainly run amazing on my modern hardware.
 
If AMD can turn this into going really hard after the $150-$400 price range, they could have a winner on their hands. Sure, 2% will be upset that they aren't going after the ultra high-end. But looking at Steam, the top 20 cards are made up of a lot of GF 10 series, 2060, 3050, 3060. They make up 31.11% of the market. 20% of them can't do DLSS 3.0 and they basically all aren't worth running RT on. A lot of these are likely vendor supplied (Dell, HP, Lenovo) but if this can help AMD penetrate back into these markets as well.

If we can get 7900XTX RT performance as is at $400, on man that a fantastic RT card. Performance is all relative to price. If the 7900XTX was priced at $1,500 it would be an absolute terrible card. If it was $200, there is no point in literally buying anything else. Sure, there might be some higher margins in something like a 4090, but when it makes up less than 1% of the market. If you can secure volume, volume will be more profitable. And I am sure it'll help a lot to have more cards out there then less cards.
That also helps build brand loyalty, reputation, and happy customers with one gpu are likely to come back to buy more.
Look at nvidia customers, most never have an issue, so they keep buying nvidia.
(I said most customers as, for whatever reason, TPU is somehow filled with all the people who are constantly having issues on their GPUs, no matter the hardware vendor)
 
I believe this is AMD's logic;
Why would I bother and put so much effort just to let (most) users buy Nvidia cards for lower prices?
İt is bitter but true unfortunately. People who never used AMD cards and RT would always complain about bad AMD drivers and low rt performance. And most people just believe that.
 
The key thing is when do RDNA5 get released.

If it's like:
RDNA 4 - mid 2024
RDNA 5 - Early 2025

Then it's do not really matter, RNDA 4 would be just a kind of architectural refresh.

We are here to talk and have fun so let's have another theory that i have no proof at all that is true or not

We all know that PS5/XBOX series will have a refresh. Would it be possible that RDNA 4 is the maximum they can push the architecture without going too far and break think for the refreshed APU of both console.

It's possible that RDNA 4 is just a step gap used for APU in console and laptop/desktop and RDNA 5 will come later with the full range. It's not because RDNA 4 doesn't have a top end card that AMD will never have one anymore. RNDA1 didn't had top end card either.

And to be honest, in the current market, the midrange is what AMD need to win. In my opinion, when you are spending 800+$ you better just get the product that have all the feature and perf.


if you shop low/mid range, well, the best price/perf is probably what you are after as you want to get the most bang for the bucks.
 
I believe this is AMD's logic;
Why would I bother and put so much effort just to let (most) users buy Nvidia cards for lower prices?
İt is bitter but true unfortunately. People who never used AMD cards and RT would always complain about bad AMD drivers and low rt performance. And most people just believe that.
True. Just like how those people praise Nvidia because they have the best top-end card, but 99% of them just buy the xx50-xx70 cards, because that's what "affordable". The category where AMD is actually competitive.
 
well If Ill be waiting for 9950x3d in 2025 anyway I might as well be waiting for next gen RDNA so perhaps a RDNA 4 cheap 2024 edition is the right choise for this year since people will want a fast highend card with there CPUs in 2025
 
RX 7900XTX performance for 500€ sounds like a really nice upgrade path from my 6900XT, especially if there's an additional featureset upgrade. And if AMD bets on value proposition for a performance that e.g. already surpasses the PS5 Pro, AMD could be working on boosting their marketshare over profits.


However, it doesn't look like AMD is in that mindset at all.
 
Alas here we go. People making stories about the narrative is crazy. Is there any statement from an AMD employee that they are not looking for the most powerful GPU to get over 50% more performance than the 7900XTX? How quickly the people in this thread forget that the 7900 series cards are faster than the 3090. I guess that does not matter in a world where the narrative is more trusted than the truth. This reminds me of the stories that purported that Ryzen was focused for the mainstream market and not for the high end.
 
If they can develop a chip at the performance of a 7900XTX, at half the price and power, theres a 10000000% chance they scale that up to hit similar 300-320w (8900XTX?) to fill that generational slot. To believe otherwise is idiotic at this point. AMDs entire goal between Zen and new RDNA parts has been cost effective & scaling designs as opposed to monolithic ones.
 
Not happy about this. The RX7900XTX is a bit lacking in 4K and apparently I won't have any options anytime soon to remedy that problem...:banghead:
you do have options drop settings or using upscaling.

But to be honest if one is looking to push 4K the 4090 is a better option but it will cost you.

It is finally going to have proper tensor cores, not the half assed matrix cores in the 7xxx series.
If this change is true and they don't reduce the VRAM buffer I would consider it for a side grade. But I will probably just sit on RDNA 3 and wait for RDNA 5.
 
For keeping the competitiveness alive.
AMD is several years behind with their RDNA architecture. They need something new and something soon in order to stay relevant.



That's not exactly right. AMD offers worse performance per money as is. Look at the benchmarks.
RTX 4090 is 71% faster than RX 7900 XTX for approximately 95% more money (1850 euros vs 950 euros).
Given Nvidia's far superior brand recognition, AMD is the big underdog and loser here.
Second tier manufacturer.

View attachment 331981
Why don't you add a pure raster chart in there also instead of just ray tracing only. Since the majority of the games on the market are still raster only? should be showing both sides and not just cherry-picking no?

1706555538769.png
 
But it does mean AMD does not go forward for another 2 or so years. Is that good in your book? AMD will be 3 generations behind nGreedia, which puts immense pressure on AMD to claw back that vast perf gap. If AMD aren't careful, Intel will catch up and maybe overtake them, then AMD/Radeon is finished. I would be very worried if I were Sony/Microsoft...

I would assume that unless nGreedia is going to separate their AI and Consumer GPU designs, then we should expect at least another 40-60% perf from Blackwell. But at what cost, well, that's only limited by Jensen's greed and arrogance.
Microsoft and Sony are "so worried" that they already plan to use RDNA in future consoles.
Microsoft console business has different problems not linked to AMD, but linked to their GamePass model.
 
Microsoft and Sony are "so worried" that they already plan to use RDNA in future consoles.
Microsoft console business has different problems not linked to AMD, but linked to their GamePass model.
After Microsoft got burned by Nvidia I don't think they will ever go back. And Sony just seems to prefer AMD.
 
After Microsoft got burned by Nvidia I don't think they will ever go back. And Sony just seems to prefer AMD.
AMD culture is the reason. You should watch Level One's visit to AMD. It reminded me of my first years working for a tech Company. Everyone talks about technology like it is NFL football and there is a relaxed atmosphere. Those holistic qualities are why I will always buy AMD. That is vs the back drop that Nvidia creates a frankenstien 4090 to try to avoid sanctions from the Country that allowed them to amass their position.
 
I take anything that MLID says at face value, but I believe it's a good strategy for AMD. Make a refined midranger, work on the feature set, the stability, architectural nooks and crannies and then prepare for a true high end next-gen product.

So AMD's best card will be about the performance of a low-end RTX 5070 or 5060. AMD really are incompetent. Good thing they were gifted an amazing CPU architecture that is easy to iterate on, otherwise they would be bankrupt by now.

If this is true, nGreedia are going to completely screw us all over.

And no, "nGreedia" won't screw any of us over. If anything their current tiering system is working rather well for them. AMD hasn't been able to compete at the high end for a very good number of years now. With Navi 21 they got close on performance (as long as you didn't know or care about what "raytracing" or "clipping your games" meant), but other than that? AMD are straight losers in every round since at least the Hawaii vs. Big Kepler bout (where the 290X turned out to be far better than the original Titan, especially in the long run), and they haven't been ahead in technology since Evergreen, the first DX11 GPU. The RTX 4090 just completely murdered them this generation.

Nothing's gonna change.

AMD culture is the reason. You should watch Level One's visit to AMD. It reminded me of my first years working for a tech Company. Everyone talks about technology like it is NFL football and there is a relaxed atmosphere. Those holistic qualities are why I will always buy AMD. That is vs the back drop that Nvidia creates a frankenstien 4090 to try to avoid sanctions from the Country that allowed them to amass their position.

What a naïve take. That's just an external image being sold. They're a multi-billion-dollar corporation with the drive and demands of such corporations. And it's the suits that decide everything.
 
What a naïve take. That's just an external image being sold. They're a multi-billion-dollar corporation with the drive and demands of such corporations. And it's the suits that decide everything.
The King of Saudi Arabia owns 75% of AMD public stock. They are not catering to the Greed cartel like Nvidia.
 
Nvidia's RTX 4090 with disabled AD102 is 25% faster in ordinary raster and whooping 60-65% faster in ray-traced gaming. AMD failed miserably with the chiplets approach, what do they think?
It is rather you who miserably failed to acknowledge engineering effort to move consumer GPUs from monolithic to chiplet-based.
Nvidia has not even tried to release chiplet-based consumer GPUs, which is more complex process.
For the first generation of chiplet-based GPUs, RDNA3 is quite ok for the beginning. Far away from great architecture, but good enough.
True, it is slower, but it is much cheaper both for manufacturers and consumers, and it will be improved in next iterations.

4090 is only 25% faster in 4K raster, but it is more than 100% more expensive than 7900XTX. 4090 is currenly arguably the worst possible value GPU for gaming. Silly gaming product for halo users for $2,000. Whoever wants to spend two grand on this card in order to game, please enjoy it, by no means, but don't try to tell us that there is 'value' in paying twice as much as the next card only to get 25% more performance in 4K. Nonsense.
4090 is definitely more suitable and valuable for creators, engineers, AI enthusiasts and graphics designers.
 
That would be an incredible upgrade proposition.
Not really for current 7900 XT(X) owners but for others maybe yeah.

Would be great if AV1 performance is improved as well.
As someone who just got a discount second hand 7900 XTX, how bad is it? Was curious on that front while I wait for it to ship.

(and yes I know I'm insane)
 
Back
Top