Discussion RDNA4 + CDNA3 Architectures Thread

Page 33 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DisEnchantment

Golden Member
Mar 3, 2017
1,770
6,720
136





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
Or Phoronix

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

 
Last edited:

eek2121

Diamond Member
Aug 2, 2005
3,270
4,798
136
Those specs are not very likely. If we say the smaller one is 32CU, then the bigger one should be 48CU or there is no point in giving It 50% more BW.

Then there is still the question about performance. It has to have significantly higher performance than N33 and N22, or what's the point in releasing It? Just to sell It dirt cheap so someone will buy It?

I wouldn't be surprised for RDNA4 to use N3E, If even Strix Halo IOD will use It.

Raytracing even If on par with Nvidia with these specs won't be that useful. RTX 4060TI 16GB manages only 45FPS and even that only at 1080p, at 1440p It's only 27fps in Cyberpunk.
That is going to depend entirely on specs. The 4090 is a ray tracing monster, for example.
 

soresu

Diamond Member
Dec 19, 2014
3,613
2,927
136
DX9 also ahead release. I'm notice when they went to AMD, quality is massive dropped
It didn't help that HD 2xxx was a trainwreck, which was not AMD's fault given Terascale was already cooking when AMD acquired them.

Also the concentration on APU and later semi custom console contracts divided their engineering department considerably as far as GPUs go - they had too many pies in the oven relative to nVidia who concentrated on just a solid PC dGPU lineup and software, and it was the software focus that came up trumps for them in professional/compute sales.
 
Reactions: Tlh97

Frenetic Pony

Senior member
May 1, 2012
218
179
116
Ok I'll guess specs:

Best case scenario:
8900: N3E, 384bit GDDR7 28gbps, 128 CUs 96/192mb cache(?), 2.8ghz, 450w, 51% increase over 7900xtx non RT
Note: Going from around 30-40% slower than Nvidia in RT titles @4k (7900xtx v 4090) to 10% slower, all other cards similar RT performance improvement, so maybe faster than Blackwell anyway/right titles.
More likely?
8800: 4P, 256bit 32/28(?)gbps, 96 CU, 64/128mb cache, 2.9ghz, 20% increase 7900xtx
8700: 4P, 192bit, 64 CU, 48/96mb cache, 3ghz, 27% increase from 7800
8600: 4P, 128bit, 40CU, 32/64mb cache, 3.1ghz, basically a 7700
8500: 4P, 24CU, APU/SOC only?
 
Last edited:

Bigos

Member
Jun 2, 2019
175
443
136
Might help for uniform computations, i.e. calculating things that are the same across all threads in a single wavefront. These are usually branch mask and address calculations, which is why SALU was apparently INT-only before. Maybe they have found some opportunities to make some float instructions uniform as well?
 

eek2121

Diamond Member
Aug 2, 2005
3,270
4,798
136

Timorous

Golden Member
Oct 27, 2008
1,883
3,616
136
Ok I'll guess specs:

Best case scenario:
8900: N3E, 384bit GDDR7 28gbps, 128 CUs 96/192mb cache(?), 2.8ghz, 450w, 51% increase over 7900xtx non RT
Note: Going from around 30-40% slower than Nvidia in RT titles @4k (7900xtx v 4090) to 10% slower, all other cards similar RT performance improvement, so maybe faster than Blackwell anyway/right titles.
More likely?
8800: 4P, 256bit 32/28(?)gbps, 96 CU, 64/128mb cache, 2.9ghz, 20% increase 7900xtx
8700: 4P, 192bit, 64 CU, 48/96mb cache, 3ghz, 27% increase from 7800
8600: 4P, 128bit, 40CU, 32/64mb cache, 3.1ghz, basically a 7700
8500: 4P, 24CU, APU/SOC only?

Are we playing guess the product stack? I love this game even though I suck at it.

Assumptions. We are only seeing N43 and N44 and they are both monolithic. Design target is for laptops mainly but will obviously be used in dGPUs. I would expect N43 die size to be around 250mm^2. That was the die size of N23 and N10 and is about the area 2 MCD N32 so that feels like a reasonable guess to start with. N44 may simply be half like N24 is.

On N4P that would mean that a 32MB, 128bit, 60CU, 3SE part would easily fit within this kind of die area baring some substantial change in architecture. 128bit feels a bit weak though.

Another option is 2x N33 but keep the 8x PCIe interface. N4p is quite a lot denser than N6 so that also feels like it fits and would give a 64MB, 256bit 64CU 4SE design. This feels more balanced but may be a bit larger than 250mm. Probably sub 300 still though.

Nx3 has always been laptop 1st so I feel the 128bit variant is more likely and they could go 128 bit and 64MB IC with faster ram to makeup the bandwidth defecit 128 bit would have. Also possible they split the difference and go with a 192 bit bus rather than a full 256 bit and I think with fast enough ram 192 is probably sufficient.

GDDR7 would make 192 bit work quite nicely if there are 3GB chips as it would allow for 18GB of ram with 6 chips but I have no clue at all if they are being made by anybody.

N32 can clock to around 2.9Ghz and not actually break the power usage completely so that is not a bad sign. I would not be at all surprised to see RDNA 4 clocking around 3.5 or higher within a sane power envelope. That + some hardware tweaks could get you to around 20% or 30% more performance per CU. That would land it around 7900XT performance at 1440p and 4K so not terrible if it came in at around $500.

Won't try and guess product names but I could see the following.

N43 full - $500 - Around 7900XT perf - 16-18GB of VRAM
N43 cut - $400 - Around 7800XT perf - 16-18GB of VRAM but maybe slower
N44 full - $300 - Around 7700XT perf - 12GB of VRAM
N44 cut - $250 - Around 6700XT perf - 12GB of VRAM

An alternative N44 cut would be $200 - 6700 perf - 8GB VRAM but I expect on what will probably be around a 150mm die the yields will be excellent so a cut this hard would be overkill.

As for RT performance, maybe a tier up so full N43 matching the 7900XTX maybe. I expect a bigger gain with RDNA 5 since that seems to be what is going into the next Xbox at least and I suspect that will be a focus for MS and probably Sony as well for PS6.

Will be fun to come back to this when RDNA 4 is announced / launched to see how miles off I am.
 
Reactions: Tlh97

branch_suggestion

Senior member
Aug 4, 2023
610
1,324
96
I did some speculative math scaling from N31 measurements with allowances for uArch area changes and the slight shrink from N4P and found that a 3SE/9SA/72CU - 48MB IC/192-bit design can fit in at around 250mm^2, that is with 15mm^2 allowance for the data fabrics. If that is too compute heavy then 3SE/6SA/60CU would be a comfortable fit. One potential issue is the size of GDDR7 PHY's vs GDDR6, as these designs would be wasted without the capacity and BW bump.
 

Mopetar

Diamond Member
Jan 31, 2011
8,282
7,236
136
Which is capable of only 25FPS in Phantom Liberty at 4K(native) with RT on.
Doesn't look monstrous enough to me.

It'll be another 3 generations before there's solid mainstream performance. GPUs aren't really built for it and anything can cripple their performance by using more than a token amount of ray tracing.

Alternatively the top-end GPU can handle a ~20 year old game that's fully implemented using ray tracing, but just barely. Maybe if ray tracing takes off enough someone will build a card that's designed for ray tracing first and has fall back options for taster games.

I wonder how close we are to the point where adding new shader engines won’t help because games simply don’t utilize them all?

I think it's mostly just a matter of how the game engines are designed. Just like no one used to make games that could use more than 2 or 4 cores because almost no one had a system with more than that amount. Now plenty of games use 8 comfortably, but most don't scale past 16 because there aren't many people with more than that.

Few developers will make the effort to design an engine that will scale to use more than twice as many shaders as are commonly available around the release of the game. Even if they could do it, by the time most people would be able to actually take advantage of that the game will be 5+ years old.

If there's a move to 8K, which I suspect will happen eventually, that'll do a lot to eat up any excess performance.
 

Ajay

Lifer
Jan 8, 2001
16,094
8,111
136
If there's a move to 8K, which I suspect will happen eventually, that'll do a lot to eat up any excess performance.
Unfortunately. 8K is a fools errand on small(ish) monitors. It's really only useful for xtra large TVs (85" and above). But - the spec wars continue undaunted. Same with 500Hz refresh rate gaming monitors. Nobody's visual cortex processes images that fast.
 
Reactions: Tlh97 and Joe NYC

Mopetar

Diamond Member
Jan 31, 2011
8,282
7,236
136
Unfortunately. 8K is a fools errand on small(ish) monitors. It's really only useful for xtra large TVs (85" and above). But - the spec wars continue undaunted. Same with 500Hz refresh rate gaming monitors. Nobody's visual cortex processes images that fast.

Consoles will want to get there as soon as they can (and now with all the upscaling and fake frames it's sort of with reach even though it's a total farce) and I can see it being used to sell high-end GPUs for the kind of people who would put it with an expensive TV just to say they game at 8K.

I think the competitive gaming isn't so much about being able to process that fast, but being able to have the game run at a higher frame rate so your shot goes off one frame before the other guy. Humans respond to audio faster than visual stimuli anyways, but that might be tied to frame rate as well.
 

dr1337

Senior member
May 25, 2020
445
723
136
Unfortunately. 8K is a fools errand on small(ish) monitors. It's really only useful for xtra large TVs (85" and above). But - the spec wars continue undaunted. Same with 500Hz refresh rate gaming monitors. Nobody's visual cortex processes images that fast.
I can still resolve the pixels on my 15" 1080p laptop when its as far away as my main monitor. And after spending the last 3 years at 120hz has made 60hz feel choppy to me whenever a game dips; I don't think 8k even at that size would be enough to accurately recreate an image of something the ways my eye see it in reality.

We live in the future don't we? Why would you not want your device you hold close to your eyes to have a better resolution than printed media?

I think A LOT of people have been fooled into thinking their brain or reality isn't nearly as sharp as it actually is. Especially when it comes to things like text, which doesn't need a ton of compute to drive high res and each pixel really makes a difference.
 

Ajay

Lifer
Jan 8, 2001
16,094
8,111
136
I think A LOT of people have been fooled into thinking their brain or reality isn't nearly as sharp as it actually is. Especially when it comes to things like text, which doesn't need a ton of compute to drive high res and each pixel really makes a difference.
Well, reality is infinitely sharp (essentially). I suppose I'd prefer my monitor to be 4K, when it came to just reading text, coding or doing professional graphics editing. But even at 60Hz - 8K, that is just a ton of data which is pointless on smallish monitors at close range. Our brains, when say hunting, just ignore allot of unimportant stuff so we can focus on killing our next meal.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |