Discussion RDNA4 + CDNA3 Architectures Thread

Page 152 - 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
1,747
6,598
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:

marees

Senior member
Apr 28, 2024
396
455
96
New wgp + high clock, should be between XT and XTX
Mlid also always stays between XT and XTX, in worst scenario like XT
Let us see

4/60% more shaders = 1.06x
22% more clocks = 1.06x × 1.22x == 1.29x
5% to 10% more ipc = 1.29x * 1.1x (max) = 1.4x

So yeah, 2% to 5% more than 7900xt possible assuming clock & ipc targets are met (within the power budget & memory speed constraints)

Realistically, I expect it to just fall short of the 7900xt (due to power & memory speed constraints)
 
Reactions: Tlh97

SolidQ

Senior member
Jul 13, 2023
504
578
96
5700XT vs 6700XT. Despite 6700XT have less bandwidth and bus, but same CU(Zero ipc between RDNA1 and RDNA2)
 

Attachments

  • IMG_20240827_170552.jpg
    31.1 KB · Views: 26

marees

Senior member
Apr 28, 2024
396
455
96
5700XT vs 6700XT. Despite 6700XT have less bandwidth and bus, but same CU(Zero ipc between RDNA1 and RDNA2)
The hype was real for RDNA 2 (navi 22). It was branded as nv killer targetting performance between 2080 super & 2080 ti. It delivered

This time there is no incentive for AMD to beat 7900xt with navi 48 as both cards can continue to be produced in parallel

Edit:
Also infinity cache
 

ToTTenTranz

Member
Feb 4, 2021
152
268
106
That seems optimistic. I expect raster to be between 7900gre & 7900xt based on available information

The top Navi 48 with 64CUs at 3.1GHz should do 50.8 TFLOPs, whereas the 7900XT with 84CUs at 2.4GHz does 51.7 TFLOPs.
If there are no IPC improvements then these should be about equal, but otherwise the top Navi 48 should be a bit better.



I do wonder if Navi 48's 16GB VRAM will eventually become a bottleneck during its lifetime, though.
 
Reactions: Tlh97 and marees

marees

Senior member
Apr 28, 2024
396
455
96
Yet another "our rumours vs MLID rumours".
When will RDNA 4 come out in the end? Answer in a month. If he turns out correct, I'll be vindicated that it made no sense for pretty much all RDNA 4 drivers to be already in Linux and yet the non-selling RDNA 3 to linger until 2025.

Also his perf predictions are embellished since he's an AMD proponent, but it's very much in the ballpark of our predictions and my earlier napkin math: Roughly between 7900 XT and XTX Raster (he says 4080, which I don't buy), about 4070 Ti Raytracing (and again he says 4070 Ti Super which I also don't buy).
MLID doubling down

Against him:
Powercolor & other board partners
Kyle Bennet (hardocp)
Wccftech
Videocardz
Toms
Techpowerup
etc. etc.

 

maddie

Diamond Member
Jul 18, 2010
4,879
4,951
136
The hype was real for RDNA 2 (navi 22). It was branded as nv killer targetting performance between 2080 super & 2080 ti. It delivered

This time there is no incentive for AMD to beat 7900xt with navi 48 as both cards can continue to be produced in parallel

Edit:
Also infinity cache
Why would AMD want to continue producing the 7900 & 7800 models? They aren't the cheapest designs to make.
 

marees

Senior member
Apr 28, 2024
396
455
96
Why would AMD want to continue producing the 7900 & 7800 models? They aren't the cheapest designs to make.
I am guessing as a stop gap until the RDNA 5 equivalent of 6700xt makes an appearance

Btw the packaging costs of n31 should be reasonable compared to CoWoS
 

maddie

Diamond Member
Jul 18, 2010
4,879
4,951
136
I am guessing as a stop gap until the RDNA 5 equivalent of 6700xt makes an appearance

Btw the packaging costs of n31 should be reasonable compared to CoWoS
It's not just the packaging, it's also the sheer area of silicon needed. Also, why mention CoWoS in this context?

Supposedly, N48 = 230-240 mm^2 (N4) versus N31 = 300 mm^2 (N5) + 225 mm^2 (N6) + advanced packaging.

N48 is going to be much cheaper to produce.
 

marees

Senior member
Apr 28, 2024
396
455
96
N48 is going to be much cheaper to produce.
True. N48 is a replacement for N43

N42 & N41 both are scrapped. Without any replacements.

So AMD has to keep producing N31 to meet the demand of N42. Sunk costs / investments should be able to keep the production line going for N31
 

Tuna-Fish

Golden Member
Mar 4, 2011
1,474
1,966
136
True. N48 is a replacement for N43

N42 & N41 both are scrapped. Without any replacements.

Based on the memory interface, N48 is a replacement for N32. N43 is a replacement for N33. N31 doesn't seem to have a direct replacement.

Unless N48 really misses expectations on RT, I suspect N31 will not stay in production because there will be a lot of benchmarks where N48 beats it, and this makes it harder to sell. Also, it's much, much more expensive to make, so unless AMD can get a much better price for it, they'd rather sell you a N48.
 

Mahboi

Golden Member
Apr 4, 2024
1,013
1,839
96
In a way, I have some good hopes for this gen.
I don't expect much, but AMD always shone precisely at realistic, decent cost midrange. This is perfect. It's the right amount of RAM, it's nothing fancy but nothing needlessly expensive.
Crossing fingers to see them pop by Oct-Nov at most and to see them actually shine in the $500-600 range with great overall perf & power draw.
 

marees

Senior member
Apr 28, 2024
396
455
96
Isn't N48 replace n31?
N48 is below 7900xtx in raster
Maybe equal to 7900xt in raster

Clearly there is a place for 7900xtx (price depends on 5070/5080)
Maybe even 7900xt since it has more vram (but with a significant discount to $650/$600)
7900 gre sells for $530 today & can continue for the same price (or $500)
 

branch_suggestion

Senior member
Aug 4, 2023
390
865
96
N21>N31>N/A
N22>N32>N48
N23>N33>N44

The goal for a replacement is to have similar or lower BoM to the previous part with a significant PPA bump.
If ordered by perf, it should go.

N31>N48>N21>N32>N44>N22>N33>N23

Yes, I do think N44 will just beat N22 so long as VRAM isn't an issue.
N32 and N48 end up in the same relative spot, roughly between the cutdown and full die of the previous gen flagship.
I do firmly believe that N32 was built to beat N21 by 5-10%, RDNA3 just had to be flawed.
 
Reactions: marees and Tlh97

leoneazzurro

Golden Member
Jul 26, 2016
1,051
1,711
136
N21>N31>N/A
N22>N32>N48
N23>N33>N44

The goal for a replacement is to have similar or lower BoM to the previous part with a significant PPA bump.
If ordered by perf, it should go.

N31>N48>N21>N32>N44>N22>N33>N23

Yes, I do think N44 will just beat N22 so long as VRAM isn't an issue.
N32 and N48 end up in the same relative spot, roughly between the cutdown and full die of the previous gen flagship.
I do firmly believe that N32 was built to beat N21 by 5-10%, RDNA3 just had to be flawed.
That's very probable in raster, for RT it will be quite probably N48>N31if the rumors about the vastly improved RT output are correct.
 
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/    |