Discussion RDNA4 + CDNA3 Architectures Thread

DisEnchantment · Mar 23, 2022

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

History for llvm/lib/Target/AMDGPU - llvm/llvm-project

The LLVM Project is a collection of modular and reusable compiler and toolchain technologies. - History for llvm/lib/Target/AMDGPU - llvm/llvm-project

github.com

Or Phoronix

More AMD "GFX940" Enablement Work Landing In LLVM - Phoronix

www.phoronix.com

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

Question - Speculation: RDNA3 + CDNA2 Architectures Thread

Man I have been dying to make this one for a while now. First rumours for RDNA3 are here so new thread time! Just going to start off with this one for now: kopite7kimi on Twitter: "@VideoCardz Ah, I mean a simple mcm design with 10240 cores is not enough. Because the lift from RDNA2 to RDNA3...

forums.anandtech.com

ToTTenTranz · Apr 9, 2024

Aapje said:
That is only true from the consumer's perspective, if you are blind to what companies need to survive (in the long term).

A product whose value is low (relative to production costs, the competition, etc) in the eyes of consumers is a bad product, because the company can't sustainably sell products like that.

At the end of the day, we are all best off if companies make products that have lots of value to us and preferably cost them much less to make than how much we value it.

I'm far from blind to what companies need to survive on the long term. All discrete GPUs being sold nowadays have ridiculous margins. The GPU performance/price ratios for consumers went way down during the crypto craze and only recently it went above late 2020 levels.
The RX 6800 XT released for $650 in November 2020 and it's still hard to find it substantially cheaper, and all the while N7 wafers and 16GB of GDDR6 16GT/s got much cheaper.

AMD and Nvidia are making bank on GPUs.

AMD's reluctance in lowering their dGPU prices probably has more to do with brand value recognition than profit margins. But my guess is they're not exactly happy with the swift downturn in marketshare from the last couple of years. Which led to RDNA4 being a Polaris rehash (product line made for marketshare recoup).
The decisions to keep N3x prices up are perhaps more related to AMD cancelling RDNA4's high-end than N31/32's failed power/clock targets.

SolidQ said:
Reddit so funny. RDNA 3.5 is 4070ti super

Isn't RDNA3.5 only on APUs? Strix Point and Strix Halo. Even the PS5 Pro has RDNA4's raytracing architecture so it's a step above that.
I'm hoping for N48 to get around 4070 Ti Super performance at a lower (~$500?) cost, though.

Mahboi · Apr 9, 2024

branch_suggestion said:
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.

I've assumed for some time that RDNA 3's success greatly hinged on RDNA 2's as well. RDNA 2 polished AMD's public image, and RDNA 3 enjoyed the afterglow of that.
How much are we talking accelerating RDNA 5? Dropping all the 3 top SKUs really accelerates next gen by 6 months, by a year?

Mahboi · Apr 9, 2024

ToTTenTranz said:
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).

You make a fair point. I do have to point out that if they did that, it's likely that they would have released those RDNA3+ cards around 1 year later or so, probably around the same delay NV released the Supers. So 1y3m?
That would put the release of RDNA 3+ around last month.I feel like the timetable might've been a little bad wrt RDNA 4.

Also, if branch_suggestion is correct and they cleaned the clocks of everyone at RTG, then it's not impossible that AMD focused on actually moving on from the RDNA 3 debacle also to focus on reviewing the entire division.
If I were a manager at RTG/AMD brass and I got told "Mr Papermaster, there's a problem with RDNA 3, apparently nobody thought of checking the voltages and the arch may perform 75% worse in power draw than expected", I might have raised an eyebrow, got very annoyed that the promised cheap-and-efficient RDNA 3 ended up so dysfunctional, and decided that "we stop all current projects, no Halo product, no respins, no refreshes, because we're going to take the time to clean up the division and not see this repeat again". Losing one very promising gen is bad, but keeping a wet dog of a division that bumps into things at the office is another.
Just speculation ofc. But putting RTG onto a lower intensity of releases to heavily review everyone's work with ample time wouldn't surprise me.
(I'm obviously actually Mark Papermaster)

branch_suggestion said:
Wafers are not the bottleneck, main bottleneck is HBM now, then adv packaging.

Looks like Norrod's statement about "substrates being the problem" is going to remain true for a long time yet.

Aapje · Apr 9, 2024

SolidQ said:
N36 was discussed here. Only i don't remember topic. RDNA3 or RDNA4

Apparently it was mentioned once by RGT and then Adroc claimed it was true. But it never even got on Wccftech, so if even they don't believe the rumor...

Page 18 - Discussion - RDNA4 + CDNA3 Architectures Thread

Page 18 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

SolidQ · Apr 9, 2024

Aapje said:
But it never even got on Wccftech

Maybe wctech didn't watch those video.

Nvidia reddit pretty funny

not competitive (c)

adroc_thurston · Apr 9, 2024

SolidQ said:
Nvidia reddit pretty funny

plebbit hates Radeon with a fire of a thousand suns.
w/ever

SolidQ · Apr 9, 2024

adroc_thurston said:
plebbit hates Radeon with a fire of a thousand suns.

Never understand it. I had Riva TNT first NV card, Radeon 9000 first ATI/AMD card, etc doesn't have problem with both. Except got 3 died cards from NV, and 0 from ATI/AMD. I'm fan price/perfomance ratio(not those upscalers/RT) and decent Vram.
Will take N44(if 200-250$) for limited time, until 9700XT, and can forget about upgrade for next 7-8 years

adroc_thurston · Apr 9, 2024

SolidQ said:
Never understand it.

Kids justifying their purchases.
Plus, NV successfully astroturfed the PCMR plebbit marketing wave starting around ~2014 or so.

blckgrffn · Apr 9, 2024

adroc_thurston said:
Kids justifying their purchases.
Plus, NV successfully astroturfed the PCMR plebbit marketing wave starting around ~2014 or so.

Oh, they'd never pay to astroturf.

Rollo? Where you at? Keys? You still reading?

Saylick · Apr 9, 2024

adroc_thurston said:
Kids justifying their purchases.
Plus, NV successfully astroturfed the PCMR plebbit marketing wave starting around ~2014 or so.

If you market your product as being "the best" and people buy your product with the assumption they are buying "the best", those people will push back on competing products which disrupt that preconception because they want to keep their status of having "the best". People don't like how it makes them feel when the Nvidia GPU they paid big bucks for all of a sudden isn't "the best".

SolidQ · Apr 9, 2024

adroc_thurston said:
NV successfully astroturfed the PCMR plebbit marketing wave

Yeah, very often still hear. AMD is bad cards, drivers doesn't work, AMD have bad encode, AMD have bad temps etc. 80% of my friends have AMD cards and don't have any problems.

People don't like how it makes them feel when the Nvidia GPU they paid big bucks for all of a sudden isn't "the best".

Like overpriced 4080?

If you market your product as being "the best

Some people saying Intel + Nvidia was always golden standard

gdansk · Apr 9, 2024

blckgrffn said:
Oh, they'd never pay to astroturf.

Rollo? Where you at? Keys? You still reading?

They don't need to anymore and with the wisdom that comes with age Jensen realized customers will spread FUD for free.

It is bizarre that the 6000 series was so close and they'll admit it sometimes but everyone still acts like Radeon has never been competitive and was never an option but they were sold out for a year too.

Mahboi · Apr 9, 2024

SolidQ said:
Never understand it.

It's very simple: NV marketing is really good at creating FOMO and customer jealousy.
They market the product as "better in every way" and "the thing everyone wants to own". This is why they're so obsessed with Steam HW survey numbers, or RT performance. It's to keep proving that you're "smart enough to buy Nvidia" or just too dumb to understand.
Doesn't work too well on actual smart people, but for sheeple, it is an incredibly effective tactic. "Oh no, everyone is saying this is the smart thing to buy, I don't want to be the idiot that bought AMD". It's Emperors New Clothes, except it's the Emperor's new GPU.

NV has managed to weaponise their customers into being cult members who actually white knight for GPU Princess Jensen Huang.
It is disgusting, but it reinforces the moat with incredible effectiveness. You'll find people moaning about FSR's image quality when it will be 99.9% the same as DLSS, because in their heads they HAVE to prove that Nvidia is just superior in all cases and anyone who didn't understand that is intellectually inferior to them.

Flatter your customers' ego, reinforce their sense of self-worth, make them believe that it's a sign of intelligence and being in the right place to buy Nvidia...and they just always sheeple back to Jensen's yard.

I would spit at the marketing for this horrendous behaviour, but the truth is, AMD has been politely bending forward and lowering their pants in front of NV's marketing. Makes it a bit difficult for me to find excuses for them.

adroc_thurston · Apr 9, 2024

blckgrffn said:
Oh, they'd never pay to astroturf.

Of course, Digital Foundry does it for ~free~.

Saylick · Apr 9, 2024

SolidQ said:
Yeah, very often still hear. AMD is bad cards, drivers doesn't work, AMD have bad encode, AMD have bad temps etc. 80% of my friends have AMD cards and don't have any problems.

Like overpriced 4080?

Some people saying Intel + Nvidia was always golden standard

Pretty much. People would rather pay increasingly exorbitant prices for Nvidia GPUs so long as they know they will never experience buyers remorse because there’s no competing option.

SolidQ · Apr 9, 2024

Mahboi said:
NV marketing is really good at creating FOMO and customer jealousy.

Yeeah i know that. Most is affected who is 15-30 year range. I remember when, in 2002 i'm finished the school. There was heavy marketing about GeForce FX, was thinking it's good, but thanks to Administrator from PC club, he's saying buy better 9700pro, he's also saying buy athlon thoroughbred b 1.5v(not 1.6v) and oc'ed it to moon. So i'm buyed it, and was very happy.

You'll find people moaning about FSR's image quality when it will be 99.9% the same as DLSS

One of reviewer was saying, when it compare in real gameplay, (not like 4x Zoom and 6-25% speed). So answer was is "it's hard notice difference in real gameplay"
Overall based on my expirience FSR need only fix Lines, like Hair or Grass
Example about hairs

Tuna-Fish · Apr 9, 2024

ToTTenTranz said:
Which led to RDNA4 being a Polaris rehash (product line made for marketshare recoup).

I'd like to note that trying to divine AMD intentions from the product line is a bit fraught because the same rumors that led us to believe that big RDNA4 was cancelled made pains to point out that this lineup was not what AMD chose to do, it was what they were forced into by external conditions. Supposedly, AMD wanted to release a world-beating stacked chiplet monstrosity, but for some reason could not.

If this is true, I would expect that the reason is that packaging resources were limited and they had to make hard choices between MI300 and RDNA4, and the consumer cards lost. If this is what happened, it pays to keep in mind that AMD did not choose this lineup as a strategy, it's what they were left with.

adroc_thurston · Apr 9, 2024

ToTTenTranz said:
Which led to RDNA4 being a Polaris rehash (product line made for marketshare recoup).

oh but it is not.
Not at all.
Polaris was bigger dies for less money. This is the direct opposite!

Tuna-Fish said:
If this is true, I would expect that the reason is that packaging resources were limited and they had to make hard choices between MI300 and RDNA4, and the consumer cards lost

Nope, RDNA4 uses no CoWoS substrate, it's all hybrid bonded.

Tuna-Fish said:
but for some reason could not.

The reason is simple, gotta focus on execution.

Saylick · Apr 9, 2024

Mahboi said:
It's very simple: NV marketing is really good at creating FOMO and customer jealousy.
They market the product as "better in every way" and "the thing everyone wants to own". This is why they're so obsessed with Steam HW survey numbers, or RT performance. It's to keep proving that you're "smart enough to buy Nvidia" or just too dumb to understand.
Doesn't work too well on actual smart people, but for sheeple, it is an incredibly effective tactic. "Oh no, everyone is saying this is the smart thing to buy, I don't want to be the idiot that bought AMD". It's Emperors New Clothes, except it's the Emperor's new GPU.

NV has managed to weaponise their customers into being cult members who actually white knight for GPU Princess Jensen Huang.
It is disgusting, but it reinforces the moat with incredible effectiveness. You'll find people moaning about FSR's image quality when it will be 99.9% the same as DLSS, because in their heads they HAVE to prove that Nvidia is just superior in all cases and anyone who didn't understand that is intellectually inferior to them.

Flatter your customers' ego, reinforce their sense of self-worth, make them believe that it's a sign of intelligence and being in the right place to buy Nvidia...and they just always sheeple back to Jensen's yard.

I would spit at the marketing for this horrendous behaviour, but the truth is, AMD has been politely bending forward and lowering their pants in front of NV's marketing. Makes it a bit difficult for me to find excuses for them.

Social acceptance is a hellova drug. Nvidia does a wonderful job tapping into that by convincing people to believe that if you buy Nvidia, you're part of the in-group where no one will criticize your purchase, irrespective of how expensive it was, because you also think that everyone buys Nvidia. It's like how Intel dominated for so long in the server business, even when AMD had better server CPUs, because "no one ever got fired for buying Intel". It's not until you stick your neck out and decide to deviate from the status quo that you start getting flak and people are not willing or comfortable being in that position, so they'd rather pay the status quo tax than to be deemed an outsider or rebel.

AtenRa · Apr 9, 2024

ToTTenTranz said:
View attachment 96646

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 slide says up to 1.8x RT and there are games that 7900XTX is up to 1.7-1.8x faster against the RX6900XT in RT

SolidQ · Apr 9, 2024

Interesting when RDNA4 will be announce Computex or Gamescon?

maddie · Apr 9, 2024

adroc_thurston said:
oh but it is not.
Not at all.
Polaris was bigger dies for less money. This is the direct opposite!

Nope, RDNA4 uses no CoWoS substrate, it's all hybrid bonded.

The reason is simple, gotta focus on execution.

I have to ask. Any idea as to price range? N44 & N48?

adroc_thurston · Apr 9, 2024

maddie said:
Any idea as to price range? N44 & N48?

Probably 300 and 500 bucks.
Maybe 329 and 529.

Timorous · Apr 9, 2024

adroc_thurston said:
Probably 300 and 500 bucks.
Maybe 329 and 529.

Cheaper than I am expecting on N48 if it manages 7900XT perf or near enough.

N44 I expect around 330 if it fits in that gap between 6700XT and 7700XT that the 3070 occupies.

adroc_thurston · Apr 9, 2024

Timorous said:
Cheaper than I am expecting on N48 if it manages 7900XT perf or near enough.

N44 I expect around 330 if it fits in that gap between 6700XT and 7700XT that the 3070 occupies.

peanut dies, pretty easy pricing decisions.

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Member

Senior member

Senior member

Golden Member

Senior member

Platinum Member

Senior member

Platinum Member

Diamond Member

Diamond Member

Senior member

Platinum Member

Senior member

Platinum Member

Diamond Member

Senior member

Golden Member

Platinum Member

Diamond Member

Lifer

Senior member

Diamond Member

Platinum Member

Golden Member

Platinum Member