Discussion RDNA4 + CDNA3 Architectures Thread

DisEnchantment · Mar 23, 2022

With the GFX940 patches in full swing since first week of March, it is looking like MI300 is not far in the distant future!
Usually AMD takes around 3Qs to get the support in LLVM and amdgpu. Lately, since RDNA2 the window they push to add support for new devices is much reduced to prevent leaks.
But looking at the flurry of code in LLVM, it is a lot of commits. Maybe because US Govt is starting to prepare the SW environment for El Capitan (Maybe to avoid slow bring up situation like Frontier for example)

See here for the GFX940 specific commits

History for llvm/lib/Target/AMDGPU - llvm/llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. - History for llvm/lib/Target/AMDGPU - llvm/llvm-project

github.com

Or Phoronix

More AMD "GFX940" Enablement Work Landing In LLVM - Phoronix

www.phoronix.com

There is a lot more if you know whom to follow in LLVM review chains (before getting merged to github), but I am not going to link AMD employees.

I am starting to think MI300 will launch around the same time like Hopper probably only a couple of months later!
Although I believe Hopper had problems not having a host CPU capable of doing PCIe 5 in the very near future therefore it might have gotten pushed back a bit until SPR and Genoa arrives later in 2022.
If PVC slips again I believe MI300 could launch before it

This is nuts, MI100/200/300 cadence is impressive.

Previous thread on CDNA2 and RDNA3 here

Question - Speculation: RDNA3 + CDNA2 Architectures Thread

Man I have been dying to make this one for a while now. First rumours for RDNA3 are here so new thread time! Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3...

forums.anandtech.com

jpiniero · Friday at 8:33 AM

gdansk said:
RTX 5060 is "$299". So N44 8GB ought to be "less".

The 9060 XT should be faster than the 5060.

GodisanAtheist · Friday at 10:34 AM

Win2012R2 said:
Because when Nvidia gets big flak for doing it, then AMD will get 10 fold.

8 GB should have died a death at least 2 years ago, minimum should be 16 GB, it costs very little over - what's there to argue about, 25-30 bucks and it's a proper card.

Nvidia is getting flak for launching a $400 8gb card (which is even more expensive street price).

8gb is once again fine at the sub $300 price point, the lower the better.

Turning down settings at $400 is a major issue. Turning down settings at $250 or $200 is less of an issue and something everyone has always done.

Intel came in strong with their 12 & 10gb cards but it looks like those are all ~$400 based on what I see on Amazon, so those aren't really a consideration sub $300 either.

Sub $300 is wide open and an 8gb card will live fine there.

gdansk · Friday at 10:41 AM

jpiniero said:
The 9060 XT should be faster than the 5060.

Not sure that moves anyone. In the low-end upscaling is more important. DLSS still has a quality and game support edge. N44 needs to be substantially cheaper to have positive reception. And since the MSRPs are fictional, why not?

jpiniero · Friday at 11:12 AM

https://videocardz.com/newz/amd-radeon-rx-9000m-mobile-rdna4-rumored-specs-radeon-rx-9080m-with-4096-cores-and-16gb-memory

AMD might be doing mobile RDNA 4 after all.

HuesToo4 · Friday at 12:04 PM

Change of plans?

Glo. · Friday at 12:09 PM

HuesToo4 said:
Change of plans?

Or they were always planned, just somebody had wrong info about it.

GodisanAtheist · Friday at 12:35 PM

jpiniero said:
https://videocardz.com/newz/amd-radeon-rx-9000m-mobile-rdna4-rumored-specs-radeon-rx-9080m-with-4096-cores-and-16gb-memory

AMD might be doing mobile RDNA 4 after all.

- Now the eternal question: Will any OEMs actually use AMD dGPUs in a laptop.

It's the eternal conundrum: If AMD releases mobile dgpus but no OEMs actually launch parts with them installed, did AMD ever actually release mobile dgpus at all?

GTracing · Friday at 1:19 PM

jpiniero said:
https://videocardz.com/newz/amd-radeon-rx-9000m-mobile-rdna4-rumored-specs-radeon-rx-9080m-with-4096-cores-and-16gb-memory

AMD might be doing mobile RDNA 4 after all.

The 9070 XT compares very well against the 4080 and 5080 at lower power targets. The RX 9080m could be very competitive against the RTX 5090 mobile.

gdansk · Friday at 1:22 PM

GodisanAtheist said:
- Now the eternal question: Will any OEMs actually use AMD dGPUs in a laptop.

I didn't think they'd announce it unless they got Asus on board for at least one or two models.

adroc_thurston · Friday at 1:31 PM

GodisanAtheist said:
Will any OEMs actually use AMD dGPUs in a laptop.

no.

Jan Olšan · Friday at 11:14 PM

Ranulf said:
AMD could win some support if they put out a 9050XT at a good price.

While Nvidia will win all the money because all the BFU Joes will buy 5060 series 8GB cards.

TESKATLIPOKA · Saturday at 3:51 AM

jpiniero said:
https://videocardz.com/newz/amd-radeon-rx-9000m-mobile-rdna4-rumored-specs-radeon-rx-9080m-with-4096-cores-and-16gb-memory

AMD might be doing mobile RDNA 4 after all.

Let's also add this.

Performance should be pretty good, Vram is not great, but that was expected.
Still, availability will be almost nonexistent.

TESKATLIPOKA · Saturday at 4:07 AM

gdansk said:
Not sure that moves anyone. In the low-end upscaling is more important. DLSS still has a quality and game support edge. N44 needs to be substantially cheaper to have positive reception. And since the MSRPs are fictional, why not?

But that's only with DLSS Transformer and of course that one is more heavy on Vram usage than CNN, so with 8GB Vram you are still pretty much busted.

I don't see a good reason why It should be substantially cheaper than RTX 5060 8GB. Give 9060(50)XT 8GB a price tag of $279(-$20), that would be a good slap in the face to Nvidia for 5060Ti 8GB's $379 price tag.

Timorous · Saturday at 4:23 AM

TESKATLIPOKA said:
This 96-bit 9050XT 12GB is not a better alternative than 9060XT 128-bit 8GB.

By having only 96-bit wide controller, It will have even less BW than RX 7600(240 vs 288 GB/s) and I think this will also reduce available Infinity cache from 32MB -> 24MB, and I am not sure If ROPs are decoupled or not from the memory controller.
You will clearly lose a lot of performance while BOM won't be lower than 9060XT 8GB and then there is still the question how long will 12GB be good enough.

To me both of them are not good products, one will sacrifice performance in current games and the other one in the future games.
I would pay a few bucks more for the full 9060XT 16GB to be honest.

An 8GB 9060 at anything more than $200 is going to get absolutely savaged in the press and it won't sell.

A 12GB 9060 at $250, even with worse average performance than the 8GB card when not Vram limited, will be praised and would sell well. The difference in BOM cost, especially given the higher yield for a 96bit, 24MB L3 part even if it does use faster GDDR6 to somewhat makeup for the reduced bus width is nowhere near the difference in ASP, perceived value and critical reception.

A 96bit 12GB part with consistent 6700XT like performance at $250 is objectively better in all the important ways than a 128bit 8GB part with performance like 7700XT sometimes and unusable at others due to Vram limits at the same price. In fact a 12GB card with 6700XT perf could probably command closer to $300 and still be received well depending on the price of the 9060XT 16GB.

Edit to add. This is especially true going forward as more games offer poor IQ or require extensive extra tuning to get a playable experience.

jpiniero · Saturday at 6:59 AM

Timorous said:
An 8GB 9060 at anything more than $200 is going to get absolutely savaged in the press and it won't sell.

That's really an argument to cancel the 8 GB entirely (or release it later or make it OEM only)... not to make a 12 GB model.

12 GB BOM cost is too close. May as well just sell more 16.

TESKATLIPOKA · Saturday at 8:12 AM

Timorous said:
An 8GB 9060 at anything more than $200 is going to get absolutely savaged in the press and it won't sell.

The criticized 5060Ti 8GB is sold for $379, yet you easily say a 9060(XT) 8GB at >$200 would be "absolutely savaged" in the press and won't sell.
I simply can't agree with this.

Timorous said:
A 12GB 9060 at $250, even with worse average performance than the 8GB card when not Vram limited, will be praised and would sell well.

I don't see why It should be praised and sell well when you lose 1/4 of performance in the big majority of games.
It would look horrible in perf/$ charts against the 8GB full version, especially at the prices you think are fair for both of them $249 vs $199.

Timorous said:
The difference in BOM cost, especially given the higher yield for a 96bit, 24MB L3 part even if it does use faster GDDR6 to somewhat makeup for the reduced bus width is nowhere near the difference in ASP, perceived value and critical reception.

For AMD the production cost of N44 is still the same, the only advantage is that they don't need to throw away the faulty parts, but those salvaged parts can be used with both 128-bit 8GB Vram and 96-bit 12GB Vram.
9060XT will supposedly use 20Gbit modules, there is nothing faster to compensate the missing width and IC.

Timorous said:
A 96bit 12GB part with consistent 6700XT like performance at $250 is objectively better in all the important ways than a 128bit 8GB part with performance like 7700XT sometimes and unusable at others due to Vram limits at the same price. In fact a 12GB card with 6700XT perf could probably command closer to $300 and still be received well depending on the price of the 9060XT 16GB.

According to what you wrote.
9060(XT) 8GB -> 3.2GHz, 32CU, 32MB, 128-bit 20Gbit -> $200 max (performance 1440p: ~133%)
9050XT 12GB -> 2.75GHz, 28CU, 24MB, 96-bit 20Gbit -> $249-279 (performance 1440p: 100%) (I dropped TFLOPs to 75% of the full N44).
Don't you find this a bit absurd? Because I do.
Just because you have a few cases where 9060(XT) 8GB will become unplayable doesn't really make 9050XT 12GB a better card warranting a higher price, because In big majority of games you sacrifice 1/4 of performance against a fully enabled N44 8GB.

Don't take me wrong, what you want is not wrong per se, just that comparison to 8GB is absurd to me when both have significant disadvantages.
I have a bit different alternative.
AMD could release 2 models for desktop and 128-bit 8GB will be left for laptops, so there would be no need comparing full N44 8GB vs crippled N44 12GB.
9060XT 16GB -> 3.2GHz, 32CU, 32MB, 128-bit 20Gbit, 200W -> $349(121%) (performance 1440p: ~133%)
9050XT 12GB -> 2.75GHz, 28CU, 24MB, 96-bit 20Gbit, 145W -> $289(100%) (performance 1440p: 100%)
I think with this everyone would be happy.
Maybe AMD not that much, because yields are likely high, so they will have to partially deactivate functional chips, but this is also a reason why the difference is only $60, so AMD wouldn't need to ask much less for the crippled chips while manufacturers will still save money on BOM by needing less modules, weaker cooler, weaker VRM, simpler PCB, and some people will rather pay the difference to get a faster product with more Vram.
I think this is a pretty good solution.

Timorous · Saturday at 9:10 AM

TESKATLIPOKA said:
The criticized 5060Ti 8GB is sold for $379, yet you easily say a 9060(XT) 8GB at >$200 would be "absolutely savaged" in the press and won't sell.
I simply can't agree with this.

The $300 5060 is going to get destroyed in the press as well. I also see no scenario where AMD charge less than $250 for an 8GB part and the whole argument is that at that price 96bit + 12GB would be better overall.

TESKATLIPOKA said:
I don't see why It should be praised and sell well when you lose 1/4 of performance in the big majority of games.
It would look horrible in perf/$ charts against the 8GB full version, especially at the prices you think are fair for both of them $249 vs $199.

Because it will be B580 performance and that was well received until the driver overhead issue and the lack of stock became apparent.

TESKATLIPOKA said:
For AMD the production cost of N44 is still the same, the only advantage is that they don't need to throw away the faulty parts, but those salvaged parts can be used with both 128-bit 8GB Vram and 96-bit 12GB Vram.
9060XT will supposedly use 20Gbit modules, there is nothing faster to compensate the missing width and IC.

You previously compared it to the 7600XT by saying a 96bit part would have less bandwidth. Using 20Gbps Vram would reduce that deficit since the 7600XT has 18Gbps modules.

TESKATLIPOKA said:
According to what you wrote.
9060(XT) 8GB -> 3.2GHz, 32CU, 32MB, 128-bit 20Gbit -> $200 max (performance 1440p: ~133%)
9050XT 12GB -> 2.75GHz, 28CU, 24MB, 96-bit 20Gbit -> $249-279 (performance 1440p: 100%) (I dropped TFLOPs to 75% of the full N44).
Don't you find this a bit absurd? Because I do.
Just because you have a few cases where 9060(XT) 8GB will become unplayable doesn't really make 9050XT 12GB a better card warranting a higher price, because In big majority of games you sacrifice 1/4 of performance against a fully enabled N44 8GB.

A few cases? There are dozens and there will be many more by the end of the year. Also the reality is that there is zero chance that an 8GB part would be sold for $200 even though that is the only price point that would make sense given the shortcomings of 8GB or VRAM.

If somehow AMD did decide to make a $200 8GB 9060 then sure thing don't bother with a 12GB part. The point is though that the 8GB part if released is going to be $250 to $300 and at that price point being obsolete in the latest AAA titles and having insufficient Vram to use features like RT is not acceptable. That is something a 12GB variant would not suffer from (at 1080p) and despite not having the same peak performance as an 8GB part would it won't have the troughs either and it will be a more consistent experience for the user.

TESKATLIPOKA said:
Don't take me wrong, what you want is not wrong per se, just that comparison to 8GB is absurd to me when both have significant disadvantages.
I have a bit different alternative.
AMD could release 2 models for desktop and 128-bit 8GB will be left for laptops, so there would be no need comparing full N44 8GB vs crippled N44 12GB.
9060XT 16GB -> 3.2GHz, 32CU, 32MB, 128-bit 20Gbit, 200W -> $349(121%) (performance 1440p: ~133%)
9050XT 12GB -> 2.75GHz, 28CU, 24MB, 96-bit 20Gbit, 145W -> $289(100%) (performance 1440p: 100%)
I think with this everyone would be happy.
Maybe AMD not that much, because yields are likely high, so they will have to partially deactivate functional chips, but this is also a reason why the difference is only $60, so AMD wouldn't need to ask much less for the crippled chips while manufacturers will still save money on BOM by needing less modules, weaker cooler, weaker VRM, simpler PCB, and some people will rather pay the difference to get a faster product with more Vram.
I think this is a pretty good solution.

That is what I am proposing....

The 8GB part is DOA unless it costs $200, which it won't. With that in mind the better alternative is 12GB 96bit at the $280ish price point the 8GB part is actually going to be priced at (assuming it is not cancelled)

Edit. And the best alternative would be a 16GB 9060 with slower ram, 28CUs, and lower clocks in that $280-$300 bracket.

jpiniero · Saturday at 9:12 AM

I am actually thinking that they might (should) rebrand N33 and sell it at it's original MSRP. Offload any cut down needed N44 to mobile (if they can)

Timorous · Saturday at 9:31 AM

jpiniero said:
I am actually thinking that they might (should) rebrand N33 and sell it at it's original MSRP. Offload any cut down needed N44 to mobile (if they can)

the 7600 would be a good sub $200 card to replace the 6600. 8GB at that price point would be totally acceptable and N33 is a touch cheaper to make than N23 was so there is little reason why AMD can't sell 7600's for the same price they have been selling 6600's for a while.

GTracing · Saturday at 12:02 PM

Timorous said:
A few cases? There are dozens and there will be many more by the end of the year.

I'm really curious what these dozens of games are. I'd challenge anyone to list one dozen games out right now where the 7600 XT 16GB is 33% faster than 7600 8GB.

And before someone replies with 4k ray tracing results where the XT gets 20fps and the non-XT gets 10fps, I'm talking about settings that people will actually use. That means 1440p or 1080p, and no ray tracing unless the 16GB card actually gets 60fps with it on, and no stupid settings like Doom Eternal's "Ultra Nightmare" that increase the VRAM buffer with no increase in visual fidelity.

Timorous · Saturday at 12:29 PM

GTracing said:
I'm really curious what these dozens of games are. I'd challenge anyone to list one dozen games out right now where the 7600 XT 16GB is 33% faster than 7600 8GB.

And before someone replies with 4k ray tracing results where the XT gets 20fps and the non-XT gets 10fps, I'm talking about settings that people will actually use. That means 1440p or 1080p, and no ray tracing unless the 16GB card actually gets 60fps with it on, and no stupid settings like Doom Eternal's "Ultra Nightmare" that increase the VRAM buffer with no increase in visual fidelity.

And while this is not comparing a 7600XT to a 7600 this 5060Ti 8GB Vs 16GB comparison shows 15 games where 8GB struggles with poor frame times or with low quality textures despite the chosen settings. Something that would manifest on the 7600XT Vs 7600 at settings where the 7600XT is playable.

https://www.techspot.com/review/2980-nvidia-geforce-rtx-5060-ti-8gb/

Be in denial if you want but the facts are clear to see.

GTracing · Saturday at 12:49 PM

Timorous said:
And while this is not comparing a 7600XT to a 7600 this 5060Ti 8GB Vs 16GB comparison shows 15 games where 8GB struggles with poor frame times or with low quality textures despite the chosen settings. Something that would manifest on the 7600XT Vs 7600 at settings where the 7600XT is playable.

https://www.techspot.com/review/2980-nvidia-geforce-rtx-5060-ti-8gb/

Be in denial if you want but the facts are clear to see.

The vast majority of people who buy a $300 GPU for gaming are not using a 4k monitor, which is why I asked for 1440p results. I'll give you Indiana Jones, that's one. As for the rest of the games you sent, the 16GB card is less than 33% faster. Still waiting on the other 11.

Timorous · Saturday at 12:55 PM

GTracing said:
The vast majority of people who buy a $300 GPU for gaming are not using a 4k monitor, which is why I asked for 1440p results. I'll give you Indiana Jones, that's one. As for the rest of the games you sent, the 16GB card is less than 33% faster. Still waiting on the other 11.

A denialist, okay pso.

Edit. I really must add but how on earth can you state the 7600XT is less than 33% faster. The smallest delta is Spider Man 2 which is around 35%. Every other example is greater than that, some cases are > 100%.

Doom Eternal is 63.9 FPS Vs 17.7 at 1440p
F1 24 RT is 47.3 Vs 18.1 at 1080p
TLOU is 62.7 Vs 41.7 at 1080p
Spider Man 2 is 69.9 Vs 51.9 at 1080p
F1 24 is 59.4 Vs 23.9 at 4k (some people plug their PC into a TV)

Like how can I take anything you say seriously when it is so divorced from reality.

This is also not even looking at 1% lows where I bet the deltas are even larger as we see with 5060Ti 8GB Vs 16GB comparisons.

GTracing · Saturday at 1:05 PM

Timorous said:
A denialist, okay pso.

I'm on board with the idea that 8GB cards are inadequate for many games. I do not agree that cutting down 128bit GPUs to 96bits is a good way to solve it.

Timorous · Saturday at 1:18 PM

GTracing said:
I'm on board with the idea that 8GB cards are inadequate for many games. I do not agree that cutting down 128bit GPUs to 96bits is a good way to solve it.

As a stop gap at a $250 price point it is imo an acceptable compromise. Ideally though NV would release 5060 and 5060Ti as 12GB parts using 3GB chips and AMD would stick with 16GB down the whole stack to save cutting the bus.

Don't confuse me advocating for an acceptable compromise that is superior to an 8GB part at $250 or more as thinking I consider it a viable long term solution.

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Lifer

Diamond Member

Diamond Member

Lifer

Member

Diamond Member

Diamond Member

Senior member

Diamond Member

Diamond Member

Senior member

Platinum Member

Platinum Member

Golden Member

Lifer

Platinum Member

Golden Member

Lifer

Golden Member

Senior member

Golden Member

Senior member

Golden Member

Senior member

Golden Member