Discussion RDNA4 + CDNA3 Architectures Thread

Page 81 - 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
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:

adroc_thurston

Diamond Member
Jul 2, 2023
3,572
5,155
96
Replacement = redesign.
No. A replacement.
but where do you get your information that it was going to have 5 chips?
It came to me in the dream.
I'm looking at earlier gens, because there we know what they released
How is this relevant to RDNA4? or 5 for that matter.
Not sure what an Nvidia chip from 2004 has got to do with any of this, or why you act so arrogant when saying something so silly.
NV stands for Navi.
You can figure out the rest yourself.
 
  • Like
Reactions: Tlh97 and Mahboi

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
Well, my logic is not that I made up some supersecret research program at AMD where they develop 8 chips every gen and count up to Navi x8, but somehow have only ever released up to Navi x4 for each of the previous Navi generations.
"Super secret"
Cause internal R&D is just something that every single corpo makes public now.
Yes, they don't document or publish what their research teams are doing. If you think I'm somehow inventing this, try to go on an AMD official channel and politely ask what are their R&D teams doing and what projects are on, and you'll get a very polite "not your business, bye".

Also they were developing as many chips as it made sense. Back in Vega/RDNA 1 days, the amount of chips was incredibly small because the budget limited it. I can guarantee that Nvidia could afford multiple times the R&D targets that AMD could, and I never set foot in either company, it's just how business works. More money means more attempts.

By your non-logic, somehow, someday, an AMD engineer was walking to the coffee machine, stumbled, the XTX die and Zen 5 die he had in his pocket cracked and mixed together, and when he took it back up, he had accidentally made Strix Halo. Corpos try random stuff all the time and see how well it goes or doesn't go, and shuffle teams or reconvert them into other projects. You don't even need to look into the semiconductor industry for this: it happens literally all the time in any company. Games get started and abandoned all the time. Markets get opened by commercials and shut by management all the time. New locations for retailers get started, then shut due to bad business after a few years all the time, etc.
It's beyond obvious that at least Navi 48 was a late development, which also explains why the fastest chip has the highest number, even though in every previous Navi generation, the higher the number of the chip, the weaker it got.
No. You don't just pile on "later developments" in multi year dev cycles. You make broad plans for everything in advance, keep what works, eat what doesn't. Big corpos are ogres.
 

Aapje

Golden Member
Mar 21, 2022
1,513
2,065
106
@Mahboi

Why would AMD spend gen after gen designing chips that they were never going to use? It doesn't make any sense.

You are just using the classic conspiracy theory nonsense: "It's obviously true, because they is zero evidence. Top Secret, yo."
 

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
Why would AMD spend gen after gen designing chips that they were never going to use? It doesn't make any sense.
Because it's Research and Development.
Not Development.

They don't know what's going to work or not. They just fling money and time at the possibilities and see what works best.
This is why R&D is such a costly thing.
You are just using the classic conspiracy theory nonsense: "It's obviously true, because they is zero evidence. Top Secret, yo."
Ok I'm done with this conversation.
 

Aapje

Golden Member
Mar 21, 2022
1,513
2,065
106
It came to me in the dream.

I guess that you are just trolling now?

How is this relevant to RDNA4? or 5 for that matter.

Things that happened during previous generations typically stay roughly the same. If a clear patter that lasted for 3 gens of a pattern, that indicates that something special happened.

NV stands for Navi.
You can figure out the rest yourself.

No, I cannot, because there is no Navi 36 chip. If you evidence that you are not just openly lying here, present it.
 

Aapje

Golden Member
Mar 21, 2022
1,513
2,065
106
They don't know what's going to work or not. They just fling money and time at the possibilities and see what works best.

Except that if you were right, you'd see those extra designs getting chosen now and then. Otherwise it makes no sense.
 

Aapje

Golden Member
Mar 21, 2022
1,513
2,065
106
@adroc_thurston

You could also just provide a link to show us what you are talking about, rather than be the way you are being. But I don't think you can, because Navi is relatively recent, so how can there be an old Navi 36 chip?

Most proper gens from AMD before they went broke were 4 to 5 parts.

But I was referring to the generations after that...
 

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
We're not even certain that XBOX will put out a new console.
Last he was seen, Phil Spencer looked like he was speaking under duress. Like management was forcing him to accept giving up on XBOX's future by holding his family hostage or something.
 

SolidQ

Senior member
Jul 13, 2023
540
632
96

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
I hadn't realised how tiny it is. 240mm². A 6600 xt was 237mm².
We're talking about an entry-midrange chip tied to an oversized 256 bit bus. For comparison the 4070 Ti is 294mm².
Even if it's the most unambitious, tiny thing ever, it is going to be a real kicker when it reaches above 7900 XT performance in raster and XTX/near 4070 Ti in RT.

I'm kind of giving up on my "maybe it'll be bigger RT improvement than that" copium, I don't think a chip this smol is going to reach into 4080 Super RT perf. If it is even just around 4070 Ti, even below it, it'll be impressive.
The last question is how well it'll handle PT/very high throughput RT. Does it just fall apart like RDNA 3, or does it hold up well when heavily queried, thanks to that BVH walker?

Also, considering that the 6600 xt was on 170W and it is on 215W, I'm really expecting some good clocks. Not just 3Ghz. 3.2, 3.3 maybe.
 
  • Like
Reactions: Tlh97

Timorous

Golden Member
Oct 27, 2008
1,748
3,240
136
I hadn't realised how tiny it is. 240mm². A 6600 xt was 237mm².
We're talking about an entry-midrange chip tied to an oversized 256 bit bus. For comparison the 4070 Ti is 294mm².
Even if it's the most unambitious, tiny thing ever, it is going to be a real kicker when it reaches above 7900 XT performance in raster and XTX/near 4070 Ti in RT.

I'm kind of giving up on my "maybe it'll be bigger RT improvement than that" copium, I don't think a chip this smol is going to reach into 4080 Super RT perf. If it is even just around 4070 Ti, even below it, it'll be impressive.
The last question is how well it'll handle PT/very high throughput RT. Does it just fall apart like RDNA 3, or does it hold up well when heavily queried, thanks to that BVH walker?

Also, considering that the 6600 xt was on 170W and it is on 215W, I'm really expecting some good clocks. Not just 3Ghz. 3.2, 3.3 maybe.

N48 looks like N10 all over again in terms of relative positioning and maybe price (inflation adjusted).
 

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
N48 looks like N10 all over again in terms of relative positioning and maybe price (inflation adjusted).
"Its price at launch was 399 US Dollars."(TPU)
I'd love it, but there's absolutely no way that AMD will spit on those juicy margins when they'll effectively be selling a 215W 7900 xt+ with an XTX's raytracing. Or a 4070 Ti's rather.
It'll be $600, at least. And for all the madness I don't think inflation has been 50% yet since 2019.
 

Timorous

Golden Member
Oct 27, 2008
1,748
3,240
136
"Its price at launch was 399 US Dollars."(TPU)
I'd love it, but there's absolutely no way that AMD will spit on those juicy margins when they'll effectively be selling a 215W 7900 xt+ with an XTX's raytracing. Or a 4070 Ti's rather.
It'll be $600, at least. And for all the madness I don't think inflation has been 50% yet since 2019.

The original announced price was $450 or $500 for the anniversary addition. Taking the $450 that would be about $550 now and with it looking slightly more competitive than the 5700XT was then yea I can see $600, I said as much in my guess table.
 

soresu

Diamond Member
Dec 19, 2014
3,229
2,515
136
N48 looks like N10 all over again in terms of relative positioning and maybe price (inflation adjusted).
If it's anywhere near the rumoured 7900 XT target for raster gfx then it's a bit above where RX 480 was in positioning, and if the rumoured revamp for BVH units is true it could be way beyond 7900 XTX for ray tracing.

I guess only time will tell, and I'm far less inclined to trust AMD's word on it after the way over optimistic projections in the RDNA3 PR slides.
 
  • Like
Reactions: Tlh97 and Mahboi

Timorous

Golden Member
Oct 27, 2008
1,748
3,240
136
If it's anywhere near the rumoured 7900 XT target for raster gfx then it's a bit above where RX 480 was in positioning, and if the rumoured revamp for BVH units is true it could be way beyond 7900 XTX for ray tracing.

I guess only time will tell, and I'm far less inclined to trust AMD's word on it after the way over optimistic projections in the RDNA3 PR slides.

Those RDNA 3 PR slides were such a disgrace IMO. AMD had spent years building trust with their marketing decks to just dump it down the drain by massively overpromising with RDNA 3.
 

Tuna-Fish

Golden Member
Mar 4, 2011
1,486
2,023
136
2028 is still pretty far away, idk how they can really call that full speed

High-end chips generally take more than 4 years to design and manufacture, starting from scratch. If they intend to have a new console with bleeding-edge tech out in 2028, the project had to start last year at the latest.

I hadn't realised how tiny it is. 240mm². A 6600 xt was 237mm². We're talking about an entry-midrange chip tied to an oversized 256 bit bus.

N7 -> N4P.

I think the "oversized bus" makes sense because it uses past-gen GDDR for cost-efficiency reasons, you need the higher width to match the increased compute resources if the memory isn't getting any faster.

Looks really a lot like Polaris to me.
 
Last edited:

Mahboi

Golden Member
Apr 4, 2024
1,033
1,897
96
N7 -> N4P.

I think the "oversized bus" makes sense because it uses past-gen GDDR for cost-efficiency reasons, you need the higher width to match the increased computer resources if the memory isn't getting any faster.

Looks really a lot like Polaris to me.
I have this friend who has been wanting to change his GPU forever.
He's still running an rx 480. Even ran RDR2 at a whopping 18 fps average on it.
Clearly, this GPU was made for him.
 
  • Like
Reactions: Tlh97 and Tuna-Fish

Aapje

Golden Member
Mar 21, 2022
1,513
2,065
106
The delay IMO would've been the best course.

A delay was not possible since they thought that they could fix it in the drivers, and they were already in volume production. So a respin without releasing the buggy RDNA3 would have meant tossing out all those chips at great cost.
 
  • Like
Reactions: Tlh97

branch_suggestion

Senior member
Aug 4, 2023
392
875
96
Overall RDNA3 was still mildly successful overall, they made up market share and increased ASP's.
Issue was dGPU volumes this gen were way down on previous, many reasons for that of course.
AMD made the right call, keep moving forward and make do with what you have instead of getting bound up over the present.
I also believe that neutering RDNA4's lineup for TTM and to accelerate RDNA5 are going to pay dividends, dGPU isn't going to see a proper sales bump until something big happens.
 

ToTTenTranz

Member
Feb 4, 2021
163
286
106
Those RDNA 3 PR slides were such a disgrace IMO. AMD had spent years building trust with their marketing decks to just dump it down the drain by massively overpromising with RDNA 3.
Yes, this particular slide was pretty jarring to be honest, particularly the part where they claim "Architected to exceed 3GHz - Industry 1st" on chips that never really managed to exceed 2.7GHz on gaming workloads.


ovVKK9wGfQBnUco9VRTLyb.jpg


And that 1.8x RT performance @2.5GHz... Perhaps they were counting on a much larger boost from the double-pumped ALUs? It's hard to say.



The respin option would also have been painful in several ways. First off, people that bought RDNA 3 would be quite angry at RDNA 3+, because it would have spelled out how much of a dud they bought. Betrayed customers aren't returning customers.
I disagree. There are no bad products, only bad prices. And bad names.

They could have released the initial N31 as 7900 and 7900XT, and then the N31+ as 7950XT and 7950XTX, for example. Or they could have released N31+ as 8900XT and 8900XTX (similar to RX480 -> RX580, which didn't seem to bother anyone at the time).


I also believe that neutering RDNA4's lineup for TTM and to accelerate RDNA5 are going to pay dividends, dGPU isn't going to see a proper sales bump until something big happens.
I'm hopeful that RDNA5 is sped up, but it's a bold assumption. Even more if RDNA5 is dependent on a new foundry process that will compete with all the AI demand.