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 · Feb 6, 2025

Bigos said:
"soon" -> "in a month"?

Rumors are pointing to a dedicated event in late February.

PJVol · Feb 6, 2025

marees said:
Yes, full details are coming soon.

Missed «™»

gaav87 · Feb 6, 2025

Soo

branch_suggestion said:
Now I'm starting to remember some interesting research from years ago.
If elements of this shared L1 paradigm are in RDNA4 that would be very cool indeed.

https://adwaitjog.github.io/docs/pdf/sharedl1-pact20.pdf

Interesting so instead of going full shared l1 cache with its negatives they introduced gl1 buffer to mitigate the interconnect bandwidth/latency hit and data replication ?

Z O X · Feb 6, 2025

gaav87 said:
Soo

Interesting so instead of going full shared l1 cache with its negatives they introduced gl1 buffer to mitigate the interconnect bandwidth/latency hit and data replication ?

Don't understand why the performance drop for some compute.
Is sampling running always, or on compute switch (e.g. different shaders running if I understand correctly)?
If it's running once, why should there be a drop?

branch_suggestion · Feb 6, 2025

Solving the Dense Geometry Problem

Discover how AMD's Dense Geometry Compression Format (DGF) revolutionizes graphics by compressing complex models for efficient real-time rendering, bridging the gap between rasterization and ray tracing.

gpuopen.com

Hopefully it is accelerated in RDNA4.
And arguably the bigger tidbit is the desire to implement it in APIs, seeing a lot of recent momentum for features to be implemented in future DX12/DXR versions.

RnR_au · Feb 6, 2025

Frank is teasing with a single slide... grr...

https://twitter.com/x/status/1887560644787920961

poke01 · Feb 6, 2025

RnR_au said:
Frank is teasing with a single slide... grr...

https://twitter.com/x/status/1887560644787920961

Annoying

blckgrffn · Feb 6, 2025

poke01 said:
Annoying

You gotta admit, if they can stretch out this limelight from nvidias underwhelming launch they are wise to. They need a long runway for the hype train for the greater gamerdom.

Then we can all dump on this thread for my birthday! 😂 (early March)

beginner99 · Feb 7, 2025

blckgrffn said:
They need a long runway for the hype train

The longer the runway the bigger the chance it derails at the end.

GodisanAtheist · Feb 7, 2025

blckgrffn said:
You gotta admit, if they can stretch out this limelight from nvidias underwhelming launch they are wise to. They need a long runway for the hype train for the greater gamerdom.

Then we can all dump on this thread for my birthday! 😂 (early March)

-Yeah if NV had stuck their launch then AMD's shens would have gone down as legend, but NV managed to screw up just about everything from performance to (street) price to availability.

NV botched this launch so bad that honestly I think most people actually forgot about AMD withdrawing RDNA 4 from computex at this point (in an outrage sense).

If AMD sticks the landing with their release (and that's a mighty big IF) then they walk away looking like heroes.

gaav87 · Feb 7, 2025

Z O X said:
Don't understand why the performance drop for some compute.
Is sampling running always, or on compute switch (e.g. different shaders running if I understand correctly)?
If it's running once, why should there be a drop?

Coz of interconnect bandwidth and data replication. So with GL1 buffer u get rid of the replication and bandwidth problem. So the cu dont need to go to l2 to fetch and just store it in the gl1 bufer and get it there. Idk if im right making it super fast. So the l1 stays almost the same but can also fetch from the gl1 buffer that acts like shared l1 from the video i think it works like this would be logical.

SolidQ · Feb 7, 2025

Listening? Let's see soon™, how they listening

https://twitter.com/x/status/1887200486408798643

soresu · Feb 7, 2025

SolidQ said:
Listening? Let's see soon™, how they listening

https://twitter.com/x/status/1887200486408798643

Lol "we are listening" sounds a whole lot creepier in the information age 😅

blckgrffn · Feb 7, 2025

soresu said:
Lol "we are listening" sounds a whole lot creepier in the information age 😅

Like... right now? WHAT ELSE HAVE YOU HEARD?!?

eek2121 · Feb 7, 2025

SolidQ said:
Listening? Let's see soon™, how they listening

https://twitter.com/x/status/1887200486408798643

I missed half the stuff on this forum due to being sick for weeks (flu, ugh), but last I checked, this launch was still rather comical:

Original price/launch date was $479-$549+ for AMD/partner models and a 1/23 launch date, according to 2 partners of a certain company making cards. This was further backed up by a retailer who claimed to have cards (I saw no evidence, only a claim) (note: the price leak actually had some credibility)
Performance was around 7900XTX
Videocardz stated that a leaker stated that the cards were delayed to march. (IIRC, but I saw the March 22nd/23rd date somewhere)
New price is ??? since NVIDIA botched the 5070 launch (high prices, limited uplift) (AMD should really just let this thing undercut the 5070. They supposedly were going to anyway. I suspect they’ll bump up prices now, however)
Performance is right up there close to the 5070ti (if this is the case, imagine if they launched at $479 🤣)
New launch date is supposedly Feb 23rd (I think?)
AMD is spending the additional time to optimize software and wait for the dust to settle.
AMD had no idea NVIDIA would botch things this badly (5070-5080 performance vs previous gen)

Did I miss anything?

Other things I picked up on trying to catch up:

Note that at least one GPU vendor is working to make MFG automatic and transparent, meaning supposedly that you won’t be able to tell if the game is using it or not. (vendor unnamed) 😡

marees · Feb 7, 2025

eek2121 said:
Did I miss anything?

Might launch for $650 for 9070xt & $480 for 9070

gaav87 · Feb 7, 2025

eek2121 said:
I missed half the stuff on this forum due to being sick for weeks (flu, ugh), but last I checked, this launch was still rather comical:

Original price/launch date was $479-$549+ for AMD/partner models and a 1/23 launch date, according to 2 partners of a certain company making cards. This was further backed up by a retailer who claimed to have cards (I saw no evidence, only a claim) (note: the price leak actually had some credibility)

Performance was around 7900XTX

Videocardz stated that a leaker stated that the cards were delayed to march. (IIRC, but I saw the March 22nd/23rd date somewhere)

New price is ??? since NVIDIA botched the 5070 launch (high prices, limited uplift) (AMD should really just let this thing undercut the 5070. They supposedly were going to anyway. I suspect they’ll bump up prices now, however)

Performance is right up there close to the 5070ti (if this is the case, imagine if they launched at $479 🤣)

New launch date is supposedly Feb 23rd (I think?)

AMD is spending the additional time to optimize software and wait for the dust to settle.

AMD had no idea NVIDIA would botch things this badly (5070-5080 performance vs previous gen)

Did I miss anything?

Other things I picked up on trying to catch up:

Note that at least one GPU vendor is working to make MFG automatic and transparent, meaning supposedly that you won’t be able to tell if the game is using it or not. (vendor unnamed) 😡

Well pretty much. I expect 5070ti -5/10%RT -20/30%PT and +5/10%Raster some announcements on 12th or 19th launch same time as 5070ti.

itsmydamnation · Feb 7, 2025

soresu said:
Lol "we are listening" sounds a whole lot creepier in the information age 😅

edit: better add some words , because words are important

edit2: damn AMD is the Ferrari of the GPU world , next Generation (tm)

marees · Feb 8, 2025

The upcoming RTX 5070 series is also likely to experience bottlenecks at the beginning. At least things are looking much better for the AMD cards.

https://twitter.com/x/status/1887955138309963847

Heartbreaker · Feb 8, 2025

eek2121 said:
I missed half the stuff on this forum due to being sick for weeks (flu, ugh), but last I checked, this launch was still rather comical:

Original price/launch date was $479-$549+ for AMD/partner models and a 1/23 launch date, according to 2 partners of a certain company making cards. This was further backed up by a retailer who claimed to have cards (I saw no evidence, only a claim) (note: the price leak actually had some credibility

That's a rather selective choice of rumor. The Vast majority have the 9070 XT launching at $600 or higher.

IEC · Feb 8, 2025

If anything I bet the MSRP goes up because nV fumbled so hard on performance and gimped lower tier cards so much.

But after serial disappointment from AMD I am happy to be disappointed in a good way. And if they actually have stock so I can get some on launch day over nV unobtanium...

Thunder 57 · Feb 8, 2025

Heartbreaker said:
That's a rather selective choice of rumor. The Vast majority have the 9070 XT launching at $600 or higher.

It would be nice to see at $600 but I'm guessing $650.

GTracing · Feb 8, 2025

At this point I think it's unlikely the 9070/XT will launch below $500/$650 in any meaningful quantities. If AMD prices them super low, like the rumored $480/$550 pricepoint, the vast majority of the stock will be bought by scalpers. AMD would miss out on a several million dollars in revenue and it wouldn't really benefit gamers.

I think the most likely scenario is that they'll have a "fake" MSRP so that they look good in reviews, and then after the first batch of cards, the only available SKUs will all be $100+ above MSRP.

gaav87 · Feb 8, 2025

GTracing said:
At this point I think it's unlikely the 9070/XT will launch below $500/$650 in any meaningful quantities. If AMD prices them super low, like the rumored $480/$550 pricepoint, the vast majority of the stock will be bought by scalpers. AMD would miss out on a several million dollars in revenue and it wouldn't really benefit gamers.

I think the most likely scenario is that they'll have a "fake" MSRP so that they look good in reviews, and then after the first batch of cards, the only available SKUs will all be $100+ above MSRP.

Scalping amd mid-range gpu u mad ?

Heartbreaker · Feb 8, 2025

gaav87 said:
Scalping amd mid-range gpu u mad ?

Look back at Crypto Mining booms. Everything was getting scalped.

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Senior member

Senior member

Senior member

Junior Member

Senior member

Platinum Member

Diamond Member

Diamond Member

Diamond Member

Diamond Member

Senior member

Golden Member

Diamond Member

Diamond Member

Diamond Member

Senior member

Senior member

Diamond Member

Senior member

Diamond Member

Elite Member

Diamond Member

Senior member

Senior member

Diamond Member