Discussion RDNA 5 / UDNA (CDNA Next) speculation

Page 30 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

gdansk

Diamond Member
Feb 8, 2011
4,330
7,258
136
AT0 might also be used for game streaming services. Maybe also for Xbox game streaming?
That's a scenario I can see it actually being made. If a large customer expressed some interest in such a chip. Then we might get some hand me downs.
 

basix

Member
Oct 4, 2024
163
324
96
Well, a large customer might be Microsoft ;)

But afaik no game streaming service is really killing it. Stadia was cancelled. Nvidias solution is not too bad but I do not know how many people use it.
 

marees

Golden Member
Apr 28, 2024
1,312
1,877
96
AT0 might also be used for game streaming services. Maybe also for Xbox game streaming?

Pack 6x AT0 together with a 96C Zen 6 EPYC CPU and you get pretty dense and fast game streaming servers (12 instances with 6 CPU cores and a halved AT0 per server). If gaming does not fly or is underutilized, you can still reuse them for some cloud Compute/ML/AI services.
I can think of 2 kinds of cloud hardware

1 identical to AT2

Another based on AT0 but virtualized into 4x (or 3x?) AT3 for a 1080p/1440p cloud streaming hardware
 
  • Like
Reactions: basix

marees

Golden Member
Apr 28, 2024
1,312
1,877
96
Well, a large customer might be Microsoft ;)

But afaik no game streaming service is really killing it. Stadia was cancelled. Nvidias solution is not too bad but I do not know how many people use it.
Nvidia's solution is very good because it has a pay as you go model. But it is region limited

Xcloud is worldwide & tie ups with Samsung TV but big issue is annual subscription model. That hurts for us. Satya would be eyeing precisely that as Azure is his baby.
 

soresu

Diamond Member
Dec 19, 2014
3,929
3,362
136
The chart MLID showed must be quite old at this point I presume? What are the odds that AMD decided to play it safe again and go for AT1 and AT3 aka the logical successors for Navi 48 and 44?
Depends how far along the process was when they got their first sales figures from 9070 XT, which were pretty promising.

Given this If there was still enough time they might extend their ambitions for the RDNA5 generation.

Albeit the sheer length of time after RDNA3 combined with the low stock and less than stellar perceived improvement of RTX 5000 series may well have significantly colored those sales numbers.
 
  • Like
Reactions: Tlh97

adroc_thurston

Diamond Member
Jul 2, 2023
6,161
8,687
106
Depends how far along the process was when they got their first sales figures from 9070 XT, which were pretty promising.

Given this If there was still enough time they might extend their ambitions for the RDNA5 generation.

Albeit the sheer length of time after RDNA3 combined with the low stock and less than stellar perceived improvement of RTX 5000 series may well have significantly colored those sales numbers.
Sales have absolutely nothing to do with decisions on this level (halo parts will be moving 3.5 units in any case).
 

dangerman1337

Senior member
Sep 16, 2010
357
23
81
I think AT0 exists because RDNA 5 looks to be a very good architecture and Blackwell was just disappointing. I mean next-gen is the opportunity for AMD to really show off on the GPU side with Nvidia becoming complacent with Blackwell (even AI performance is lackluster which is why the top PCIe Pro card has bee pushed to 600W Vs 300W of before basically) and Rubin/RTX 60 doesn't seem more than a tweaked Blackwell architecturally on a smaller node.

The fact @Kepler_L2 has retweeted makes me think the Document is more recent than last year's roadmaps which showed them just targeting similar tiers to RDNA 4, maybe a tad higher. Though the interesting part is the potential having it with a smaller companion die (probably on 6nm mostly) that repurposes the Graphics Die into a dGPU Consumer card, Datacenter, Console etc. Keeps it lean and gives AMD the ability to redirect Graphics dies if one segment goes badly to another (if Xbox next-gen flops, no worries just-redirect those dies to AT2 using Radeon SKUs).
Not quite so trivial as that.

Not unless they want to abandon the advantages consoles have typically had over the PC platform for a given spec.

By using a more limited hw spec game devs can optimise performance in pre compiled shaders and assets tailored to fit.

Even a mid cycle refresh is adding significantly more required effort to supporting each platform.

Abandoning the current regime would probably end up with even less devs wanting to go with MS without being bought into it.
If there is no strix halo laptop / tablet then there is no use case for a strix halo like medusa halo

AMD might as well replace the 192 bit 72 CU AT2 in magnus with a 128 bit 48 CU AT3 & call it a day 🤔🤔
MY view is that we may end up with multiple Xbox SKUs with AT3, 2 & 1. If AT3 is manufactured on N3C (does that have the same density as N3P? if so and the difference is slightly lower performance for a few thousand cheaper per wafer price kinda a no brainer) I can see a 1TB nVME $500 base Xbox Home console with 12GB of VRAM + 16 or so GB of LPDDR5X as the "entry level" box, AT2 with a 2TB nVME be the $800 Series X successor effectively and $1200+ AT1 with 4TB nVME acting as the stand in for a pro.

WIth the costs of nodes below N3 looking expensive... I get the feeling MS will keep the costs offering by the end of 2027 three home console tiers that each is paired with the Magnus CPU featuring IOD that can design wise be swapped (albeit with different mobo + chassis).
 
  • Like
Reactions: Tlh97 and Snoota

Vikv1918

Member
Mar 12, 2025
27
62
51
MY view is that we may end up with multiple Xbox SKUs with AT3, 2 & 1. If AT3 is manufactured on N3C (does that have the same density as N3P? if so and the difference is slightly lower performance for a few thousand cheaper per wafer price kinda a no brainer) I can see a 1TB nVME $500 base Xbox Home console with 12GB of VRAM + 16 or so GB of LPDDR5X as the "entry level" box, AT2 with a 2TB nVME be the $800 Series X successor effectively and $1200+ AT1 with 4TB nVME acting as the stand in for a pro.
Unless these consoles can play your PC library or have exclusives, they're DOA. And even if they could play PC games, it wont sell more than 30 million lifetime. I think Xbox is quiet quitting the console business, which is why they are using off-the shelf parts for the next one, to minimize the investment required and make porting easier.

I think this would be a good thing for PC Radeon. MS winding down console, and Sony's mediocre sales means Radeon needs to start being serious about dGPUs and laptops/OEMs, in order to keep up those billions of revenue that was basically for granted in the past.
 
  • Like
Reactions: marees

blckgrffn

Diamond Member
May 1, 2003
9,684
4,331
136
www.teamjuchems.com
They (PS4 vs PS5) are pretty close, according to Google:

Overall Sales (Worldwide)
  • As of May 2025 (55 months post-launch), the PS4 had sold 79.45 million units, while the PS5 had sold 76.31 million units. This means the PS4 is currently ahead of the PS5 in worldwide sales by 3.15 million units at that point in their respective lifecycles.
  • Recent Performance (June 2025)
    • In June 2025, which marks the 57th month since the PS5's launch, the PS4's sales gap against the PS5 widened compared to the aligned launch period by 82,955 units globally.
    • In the last 12 months (leading up to June 2025), the PS5 outsold the PS4 by 39,237 units globally.
"AI responses may include mistakes."
 

soresu

Diamond Member
Dec 19, 2014
3,929
3,362
136
They (PS4 vs PS5) are pretty close, according to Google:

Overall Sales (Worldwide)
  • As of May 2025 (55 months post-launch), the PS4 had sold 79.45 million units, while the PS5 had sold 76.31 million units. This means the PS4 is currently ahead of the PS5 in worldwide sales by 3.15 million units at that point in their respective lifecycles.
  • Recent Performance (June 2025)
    • In June 2025, which marks the 57th month since the PS5's launch, the PS4's sales gap against the PS5 widened compared to the aligned launch period by 82,955 units globally.
    • In the last 12 months (leading up to June 2025), the PS5 outsold the PS4 by 39,237 units globally.
"AI responses may include mistakes."
Ah right.

Not quite so impressive then given increased population worldwide since the PS4.
 

blckgrffn

Diamond Member
May 1, 2003
9,684
4,331
136
www.teamjuchems.com
Ah right.

Not quite so impressive then given increased population worldwide since the PS4.
I guess that is a point of perspective. As you said, the PS5 had to deal with many garbage factors in its lifetime, so it seems like its doing pretty well.

Also, I am not sure if this includes the Pro numbers, and that might tip the scales since (from what I understand) the PS5 Pro has been much more brisk compared to the PS4 Pro.

1753985918120.png

"Recent" Eurogamer article, I don't think the initial Pro momentum stuck and now they don't call it out, I guess.
 
  • Like
Reactions: Tlh97 and Mopetar

Claudiovict

Junior Member
Jul 21, 2025
2
0
6
I think AT0 exists because RDNA 5 looks to be a very good architecture and Blackwell was just disappointing. I mean next-gen is the opportunity for AMD to really show off on the GPU side with Nvidia becoming complacent with Blackwell (even AI performance is lackluster which is why the top PCIe Pro card has bee pushed to 600W Vs 300W of before basically) and Rubin/RTX 60 doesn't seem more than a tweaked Blackwell architecturally on a smaller node.
It is possible to AMD get in pair with NVidia's Path Tracing performance on RDNA5?
 

SolidQ

Golden Member
Jul 13, 2023
1,464
2,387
106
From AMD presentation 2023
ovJCj65Qu7V3BCmE.png
 

ToTTenTranz

Senior member
Feb 4, 2021
488
888
136
From AMD presentation 2023
ovJCj65Qu7V3BCmE.png


This isn't great, but the handheld looks quite a bit worse:


1754050410803.png


This doesn't look like a PS6 handheld. For that they'd need at least CPU parity, which the 4 core can't achieve.
And unless those Zen6c cores have a massive IPC increase over Zen5 and/or are clocking at 5GHz (doubtful for a handheld), it doesn't look like it can realistically emulate the PS5 CPU either.

Also, LPDDR5X in early 2028 doesn't sound super convincing.

If these specs come true, this looks more like a handheld with its own SDK that won't really play any PS6 titles and will only play the handful PS5 titles whose developers bothered to make a "low-TDP" version.
And if that's the case, the chances for Sony to release this and then just leave it to die after launch (like they did with PSVR2 and Vita) are pretty big.
 

marees

Golden Member
Apr 28, 2024
1,312
1,877
96
This isn't great, but the handheld looks quite a bit worse:


View attachment 128033


This doesn't look like a PS6 handheld. For that they'd need at least CPU parity, which the 4 core can't achieve.
And unless those Zen6c cores have a massive IPC increase over Zen5 and/or are clocking at 5GHz (doubtful for a handheld), it doesn't look like it can realistically emulate the PS5 CPU either.

Also, LPDDR5X in early 2028 doesn't sound super convincing.

If these specs come true, this looks more like a handheld with its own SDK that won't really play any PS6 titles and will only play the handful PS5 titles whose developers bothered to make a "low-TDP" version.
And if that's the case, the chances for Sony to release this and then just leave it to die after launch (like they did with PSVR2 and Vita) are pretty big.
It is more like a PS5 handheld

Could even release before the PS6, I reckon