Discussion RDNA4 + CDNA3 Architectures Thread

[DisEnchantment]

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

 

[GodisanAtheist]

5700xt was already competitive with 2070 and better than the 2060S.

-But it was a long way off from the 2080Ti.

How would that work? 5700xt at ~250mm^2 would be competitive with 2070 but 6800XT with a big die and new uarch was going to competitive with 3070?

AND just to show that I'm not talking in hindsight, here are posts that I made in September/October 2020: So AMD targeted big Navi (500+mm^2 7nm card) to fight a 3070 (390mm^2 8nm)

- The fact that RDNA2 was stuck on the same node and had relatively narrow bus widths did it for a lot of people. Infinity Cache really came out of nowhere.

Also, let us please not ignore the deep persistent irony of RDNA3, which ended up with 500mm^2+ of RDNA3 breaking even against 370mm^2 of Ada (and losing in RT)... exactly the argument that was supposedly outlandish when talking about RDNA2...

Please know that when I use the word "everyone" as in "everyone thought" its just rhetorical flair/hyperbole. I'm just suggesting that it was a very popular position among the masses, even if folks such as yourself called the play correctly.

 

[exquisitechar]

It became obvious that RDNA2 was going to be a big leap over RDNA1 when we got details about the PS5. A huge increase in clock speeds/power efficiency was confirmed. But it is true that a ton of people were in denial for some reason and kept saying that it would be 2080ti level up until almost the official reveal.

Don’t know how good the 9070 series is going to be as a product, but RDNA4 also seems almost certain to be a very good microarchitecture at this point. Big mistake that AMD isn’t releasing a top to bottom stack, they would have been significantly more competitive than they were with even RDNA2.

 

[gdansk]

It's not a mystery. Big Navi 4x would have been massive. They cancelled it because Radeon never builds chips that big. Nvidia does (when they need to).

 

[Hail The Brain Slug]

reaper

I really like the Nitro+ except they've gone with 12v2x6 and hidden it at the end of the PCB underneath the backplate. This means you are forced to have a tight 90-degree bend in the cable, which we all know by now is a bad idea. Why Sapphire engineers went this route, I have no idea. They wanted to be cute and have a hidden power cable?

Will it even have current monitoring and load balancing across the pins?

 

[adroc_thurston]

Also, let us please not ignore the deep persistent irony of RDNA3, which ended up with 500mm^2+ of RDNA3 breaking even against 370mm^2 of Ada (and losing in RT)... exactly the argument that was supposedly outlandish when talking about RDNA2...

Oh no brother that's not how it works.
N31 is a tiled design, 225mm^2 of it are N-1.

They cancelled it because Radeon never builds chips that big.

Nope.
Wasn't a 'chip' either, a package really.

 

[RnR_au]

I really like the Nitro+ except they've gone with 12v2x6 and hidden it at the end of the PCB underneath the backplate. This means you are forced to have a tight 90-degree bend in the cable, which we all know by now is a bad idea. Why Sapphire engineers went this route, I have no idea. They wanted to be cute and have a hidden power cable?

Will it even have current monitoring and load balancing across the pins?

Does 90 deg bends matter in the 300W class for the 12v2x6 power connector? For 600W... sure... you are right up against the spec limits... but 300W? I don't think I would care.

 

[exquisitechar]

Oh no brother that's not how it works.
N31 is a tiled design, 225mm^2 of it are N-1.

True, and the tiled design itself came with an overhead. Still, RDNA3 was not very close to Ada.

 

[adroc_thurston]

Still, RDNA3 was not very close to Ada.

Wasn't super far off either.

 

[GodisanAtheist]

Oh no brother that's not how it works.
N31 is a tiled design, 225mm^2 of it are N-1.

- I know brother.

And if you're going to try and sell the idea that AMD wanted 500mm2 of chiplet based silicon to go against 370mm2 of monolithic die (and still lose in RT) I suggest you sell your bridges elsewhere.

We were all here for the soaring hype and crushing disappointment of the RDNA3 launches, based on paper specs.

 

[gaav87]

I really like the Nitro+ except they've gone with 12v2x6 and hidden it at the end of the PCB underneath the backplate. This means you are forced to have a tight 90-degree bend in the cable, which we all know by now is a bad idea. Why Sapphire engineers went this route, I have no idea. They wanted to be cute and have a hidden power cable?

Will it even have current monitoring and load balancing across the pins?

Ye i also wanted nitro+ but then i saw it... Still the xfx mercury and sapphire pure looks good... Also taichi has h++

 

[Kronos1996]

Indeed but sanity sucks.

I guess.

NV will ramp volumes soon enough.
They just had a late prod kickoff for Blackwell since the thing is rife with issues on all levels.

Any idea how/why Nvidia screwed the pooch with Blackwell? I assumed changing nodes during design didn’t help but Zen 5 managed to turn out fine albeit slightly disappointing.

 

[adroc_thurston]

And if you're going to try and sell the idea that AMD wanted 500mm2 of chiplet based silicon to go against 370mm2 of monolithic die (and still lose in RT) I suggest you sell your bridges elsewhere.

Total die area for tiled parts is whatever.
You gotta look at SM/CU count and GCD/SED area and that's the thing where RDNA3 failed.
96CU and 384b G6 and it's barely competitive against AD103.
Fortunately, it's the opposite now.

 

[GodisanAtheist]

Total die area for tiled parts is whatever.
You gotta look at SM/CU count and GCD/SED area and that's the thing where RDNA3 failed.
96CU and 384b G6 and it's barely competitive against AD103.
Fortunately, it's the opposite now.

- The whole argument was "it's silly to think a 500mm2 AMD part would compete with a 390mm^2 Nvidia part" and yet that's exactly what happened.

Retroactive precedent

 

[Mopetar]

4) What we are seeing are post-tariff prices, and the reason AMD delayed release was primarily to earn money by holding their stock of pre-tariff imports and selling them at the higher price level caused by tariffs.

That's not how tariffs work. AMD makes the same amount of money and the consumer pays the additional cost of the tariff. It's no different than any other tax in that way. Similarly, AMD doesn't make more money holding off sales until a VAT increase went into effect for example.

Perhaps you could argue that by holding off they don't get put into an uncomfortable position where they launch at some price and then have a 10%+ jump in price when tariffs would go into effect.

If anything they'd still want to launch early as higher prices would limit sales in the long run. It doesn't matter much if like Nvidia they have a limited supply that will sell out or even drive the market rate above MSRP, but generally speaking lower prices means higher sales.

 

[GodisanAtheist]

That's not how tariffs work. AMD makes the same amount of money and the consumer pays the additional cost of the tariff. It's no different than any other tax in that way. Similarly, AMD doesn't make more money holding off sales until a VAT increase went into effect for example.

Perhaps you could argue that by holding off they don't get put into an uncomfortable position where they launch at some price and then have a 10%+ jump in price when tariffs would go into effect.

If anything they'd still want to launch early as higher prices would limit sales in the long run. It doesn't matter much if like Nvidia they have a limited supply that will sell out or even drive the market rate above MSRP, but generally speaking lower prices means higher sales.

- I thought the implication there was AMD gets to hide a price increase behind tariffs. Ye old repackaged "greedflation".

 

[marees]

Supposed to be 304W TBP for the 2970MHz MBA version of XT. 3.1GHz will probably be in the 360W+ range.

https://videocardz.com/newz/amd-radeon-rx-9070-xt-confirmed-as-304w-card-rx-9070-non-xt-is-220w

Those clocks seem low

 

[marees]

40% higher vs the 7900 GRE

 

[Jan Olšan]

Hopefully that means the price is in the same realm.

Well TPU has the XTX 37% faster than the GRE at 4K.
Therefore it is a 16GB N31 at lower power with far better RT and low precision accel using much less Si.
Aka the best uplift of any GPU architecturally since Maxwell.

Careful with comparing a percentage from one source to others. It can have up to double digits error margin because games are varied that much (and that's putting aside possible cherry picking and other distortions to the original percentage).

I tried to fit the leaked numbers to ComputerBase and TPU and ended up with 9070XT possibly being anywhere between 88-98% of RTX 5070 Ti performance. It looks like XT won't be faster/matched with it, that's probably certain now.

 

[Jan Olšan]

wait they teased MBAs and canned them? Impressive.

AMD has done 3D model only (I assume) designs many times. The reason to make one even if you aren't going to sell your own cards is to have something AIB-neutral to use in your promo materiels, for artworks on your site, PDFs and so on.

 

[marees]

40% higher vs the 7900 GRE

View attachment 117813

$650 ???

https://twitter.com/x/status/1684981936849100801

 

[SolidQ]

$650 ???

That what i was trying to find original price. Unless AMD compare it with global 550$ price

 

[Hail The Brain Slug]

$650 ???

https://twitter.com/x/status/1684981936849100801

AMD Radeon RX 7900 GRE review: the lowest binned Navi 31 variant is now globally available, starting at $549

Navi 31 hops on a golden rabbit.

www.tomshardware.com

It launched globally at $549

​

 

[Mopetar]

- I thought the implication there was AMD gets to hide a price increase behind tariffs. Ye old repackaged "greedflation".

I mean they can blame tariffs for an increase in price, but it's not as though consumers can't look at the price and the tariffs to figure out what the cost would be without the tariff. It's also a cost increase that affects the entire market (unless someone starts manufacturing and assembling GPUs in the USA) so it's a wash as far as competitive advantage/disadvantage is concerned.

Also greedflation is an idiotic term, particularly when the market price is exceeding MSRP for a product. Yes, NVidia/AMD have raised prices, but that's natural given that there's a new class of consumer that values GPUs far above what gamers do. No one would call you or anyone else greedy for taking a new job that paid you 50% more than what you currently earn even though the cost of your labor has gone up 50% over what it was previously.

 

[gdansk]

It looks like XT won't be faster/matched with it, that's probably certain now.

They really need a fake MSRP for reviews then. I had assumed they would need a MBA model to do that but Nvidia showed how a fake MSRP can still be done with only AIB cards.

 

[Shmee]

AMD Radeon RX 7900 GRE review: the lowest binned Navi 31 variant is now globally available, starting at $549

Navi 31 hops on a golden rabbit.

www.tomshardware.com

It launched globally at $549

​

​

Yup I used a 7900GRE in a build for someone, it was around $520. Worked out well, huge upgrade over their old computer. Got them a 7800X3D, 32GB RAM, 7900GRE, and I think a 2TB NVMe. They were coming from Sandy Bridge with I think only 8GB of RAM, and an R9 290 I believe.