Discussion RDNA4 + CDNA3 Architectures Thread

Page 469 - 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
1,777
6,789
136
1655034287489.png
1655034259690.png

1655034485504.png

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 :grimacing:

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

1655034362046.png

Previous thread on CDNA2 and RDNA3 here

 
Last edited:

branch_suggestion

Senior member
Aug 4, 2023
771
1,680
106
No, it's a part made for careful SR-IOV slicing into cloud gaming instances.
Gamers(tm) will get a mutilated chop with castrated membw that will be a solid chunk slower than the client halo Rubin.
That is not the Radeon way, last time they technically did not ship the full die to gamers was Vega20.
That was the only time and Navi10 was the actual gaming part anyway.
Even if they only ship the full die in limited edition overpriced parts, they will.
yeah it is.
Either they win, or they don't.
And they don't win.
Well yeah, they are avoiding NV's moats and creating new ones in hope of a paradigm shift.
NV can't properly follow cause no x86. But yes the only sure fire way is to give them the old R300 treatment and hope NV goes full Intel in bag fumbling.
Repeating the Zen strat seems to be the plan, a plan that worked cause they won at the top. They can outscale NV, but will they?
 

adroc_thurston

Diamond Member
Jul 2, 2023
6,503
9,140
106
  • Like
Reactions: marees

Joe NYC

Diamond Member
Jun 26, 2021
3,467
5,060
136
One year ago AMD announced their shift in strategy, supposedly focused on regaining market share:

AMD’s Jack Huynh

So how does this pan out? AMD market share has increased by... checks notes... -2.1%. Yep, that's a minus.

Q2’25 PC graphics add-in board shipments increased 27.0% from last quarter
View attachment 129581

Interesting that AMD, at the same time, reported more GPU revenue.

I am not really sure if JPR numbers have any validity. There was this incident where he reported big increase in Intel market share, based on, what it turned out to be some internal payment for IP between different Intel departments, while he thought this was revenue from shipped cards. He divided this phantom payment by ASPs and conjured up a big jump in Intel GPU market share.
 

adroc_thurston

Diamond Member
Jul 2, 2023
6,503
9,140
106
Interesting that AMD, at the same time, reported more GPU revenue
Because 9070XT is what sells and it has healthy ASPs.

There was this incident where he reported big increase in Intel market share, based on, what it turned out to be some internal payment for IP between different Intel departments
He tracks units, not rev.
 

branch_suggestion

Senior member
Aug 4, 2023
771
1,680
106
but also a de-facto surrender in DIY discrete graphics.
That is literally the only bright spot in dGPU for AMD.
Eventually DIY desktop and render farms will be the only dGPU markets, everything else will be Fusion(TM).
In every market with Zen, they can win, DIY dGPU is won by the biggest stick so that is not impossible either.
Meanwhile they presumably have multi-billion dollar firm orders for AT2/0, AT3/4 should also sell in billions of dollars of client parts, and Sony is the cherry on top.
RDNA5 is literally too big to fail, they can afford to ship 100k full AT0 bins to prove NV is not invincible, and AT2 is gonna be a really good N48 successor anyway.
What they are doing is clever, logical and good business, and yes easier than trying to compete with NV directly in their complete monopolies.
If not trying to beat NV in the hardest, grindiest way possible is surrendering, so be it.

If they first ship full AT0 to DIY, and with GFX14 ship the chiplet chungus then that cannot ever be called surrender.
 

adroc_thurston

Diamond Member
Jul 2, 2023
6,503
9,140
106
Eventually DIY desktop and render farms will be the only dGPU markets, everything else will be Fusion(TM).
Uh, nope.
Meanwhile they presumably have multi-billion dollar firm orders for AT2/0, AT3/4 should also sell in billions of dollars of client parts, and Sony is the cherry on top.
Duh, but that also means DT stack is hand-me-downs from elsewhere that are not built to intercept NV SKUs.
they can afford to ship 100k full AT0 bins to prove NV is not invincible
Thr thing will lose to halo Rubin hard.
It's heavily underspec'd on shaders to compete.
What they are doing is clever, logical and good business, and yes easier than trying to compete with NV directly in their complete monopolies.
They're going heads on in GPGPU which is how you compete.
Otherwise it's surrender.
 
  • Like
Reactions: exquisitechar

SolidQ

Golden Member
Jul 13, 2023
1,483
2,423
106
It's gonna have more stuff than AMD will ship for RDNA5.
maybe it's can be another blackwell. We just need final benchmarks


Unlikely, the channel sell-thru for n44 looks tiny.
Amazon Spain
d65caf80bae995fee9416c07ece5962e.png

Amazon De
03091d13c5237de3809d314c2082f946.png

Amazon US
b0b6ffb4c9825d109f8772dd9b20ef22.png

and bonus Mindfactory(last week)
GzbAlPcWoAAN4ld

It's fine
 
  • Like
Reactions: Tlh97 and Joe NYC

branch_suggestion

Senior member
Aug 4, 2023
771
1,680
106
Uh, nope.
The share dGPU will have in mobile and non-DIY desktop will shrink steadily, though tbh the latter market is whatever really.
Duh, but that also means DT stack is hand-me-downs from elsewhere that are not built to intercept NV SKUs.
If the IP is good, it is good, AT2 should compete fine even though NV can bracket against it.
The thing will lose to halo Rubin hard.
It's heavily underspec'd on shaders to compete.
It seems to be the same number of shaders as GB202 (w/VOPD), sure GR102 can have like a fair bit more but they also need to sort out their lack of scaling as of late, both in core PPC being virtually unchanged since Ampere, poor scaling at the high end and finally their driver overhead/display core nonsense.
AT0 GMD is already looking a tad over 600mm^2 so NV doesn't have that much more area to use.
They're going heads on in GPGPU which is how you compete.
Otherwise it's surrender.
Well yeah GPGPU is a infinite money hack even with a small share, easy to throw the kitchen sink at.
I truly think it will be tighter at the top end than N21 vs GA102.
 

adroc_thurston

Diamond Member
Jul 2, 2023
6,503
9,140
106
The share dGPU will have in mobile and non-DIY desktop will shrink steadily, though tbh the latter market is whatever really.
Mobile dGFX will last since config flexibility is nice.
If the IP is good, it is good, AT2 should compete fine even though NV can bracket against it
IP's gonna be good either way, it's just that config options are now stapled to other markets with zero pop flexibility or any ability to muster a competitive response.
Something like N36 physically can't happen inside RDNA5.
AT0 GMD is already looking a tad over 600mm^2
Unlikely, it's N3p and they hacked off a ton of SRAM.
Well yeah GPGPU is a infinite money hack even with a small share, easy to throw the kitchen sink at.
No, it's just down to dr. Lisa Su liking big server iron and not liking the gamers™.
I truly think it will be tighter at the top end than N21 vs GA102
Oh no it's not gonna be even remotely close on the top-end.
Gonna be n31 vs ad102 electric boogaloo.
 

branch_suggestion

Senior member
Aug 4, 2023
771
1,680
106
IP's gonna be good either way, it's just that config options are now stapled to other markets with zero pop flexibility or any ability to muster a competitive response.
Something like N36 physically can't happen inside RDNA5.
We shall see how RDNA5 goes in the wild, if all is well enough they may go back to gamer optimised parts and trying to counter NV directly.
Unlikely, it's N3p and they hacked off a ton of SRAM.
The giant bump in L2 offsets the area loss of MALL pretty heavily, so you must be talking about the shader core.
No, it's just down to dr. Lisa Su liking big server iron and not liking the gamers™.
No doubt there, since the start everything must feed Forrest's hunger for more iron. The roadmap never ends.
Oh no it's not gonna be even remotely close on the top-end.
Gonna be n31 vs ad102 electric boogaloo.
Still better than now, or worse, it depends.
If 8SE/96WGP equivalent is that far off the top then the rest of the stack is doomed no matter what.
Like you really expecting Rubin to be another Ampere>Ada leap? With a much less significant node shrink?
 
  • Like
Reactions: Tlh97