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

 

[gdansk]

Oh wow I'M SAVED

Spoiler

Not a single model of the new GPUs will fit in my tiny 315mm length mATX case
yayyyy.............................................................................

https://www.powercolor.com/product-detail212.htm

Reaper?

 

[Heartbreaker]

Youtube! Leaked, not licked!

FSR4 has fallback mode for non-90 series, so technically FSR4 works with other cards.

 

[Mahboi]

https://www.powercolor.com/product-detail212.htm

Reaper?

Oh wow, it does fit, I checked the website but missed it. I think it's a lil' dinky of a cooler for 300W if I'm completely honest though.

 

[gdansk]

Oh wow, it does fit, I checked the website but missed it. I think it's a lil' dinky of a cooler for 300W if I'm completely honest though.

Yeah, it's probably pretty loud. There's also the ASRock Steel Legend which might fit.

ASRock Radeon RX 9070 XT Steel Legend 16GB

<b>Clock: GPU / Memory</b>, Boost Clock<span style=color:red;>*</span>: Up to 2970 MHz / 20 Gbps, Game Clock<span style=color:red;>*</span><span style=color:red;>*</span>: 2400 MHz / 20 Gbps<br /><b>Key Specification</b>, AMD Radeon™ RX 9070 XT GPU, 16GB GDDR6 on 256-Bit Memory Bus, AMD RDNA™ 4...

www.asrock.com

And the Asus Prime

ASUS Prime Radeon RX 9070 XT OC Edition - Tech Specs

Detailed specification of PRIME-RX9070XT-O16G Tech Specs，Graphic Engine:AMD Radeon™ RX 9070 XT、Memory Speed:20 Gbps、Bus Standard:PCI Express 5.0、OpenGL:OpenGL®4.6、Video Memory:16GB GDDR6、Engine Clock:TBD、Stream Processors:4096、Memory Interface:256-bit、Resolution:Digital Max Resolution 7680 x...

www.asus.com

 

[Mahboi]

Yeah, it's probably pretty loud. There's also the ASRock Steel Legend which might fit.

ASRock Radeon RX 9070 XT Steel Legend 16GB

<b>Clock: GPU / Memory</b>, Boost Clock<span style=color:red;>*</span>: Up to 2970 MHz / 20 Gbps, Game Clock<span style=color:red;>*</span><span style=color:red;>*</span>: 2400 MHz / 20 Gbps<br /><b>Key Specification</b>, AMD Radeon™ RX 9070 XT GPU, 16GB GDDR6 on 256-Bit Memory Bus, AMD RDNA™ 4...

www.asrock.com

And the Asus Prime

ASUS Prime Radeon RX 9070 XT OC Edition - Tech Specs

Detailed specification of PRIME-RX9070XT-O16G Tech Specs，Graphic Engine:AMD Radeon™ RX 9070 XT、Memory Speed:20 Gbps、Bus Standard:PCI Express 5.0、OpenGL:OpenGL®4.6、Video Memory:16GB GDDR6、Engine Clock:TBD、Stream Processors:4096、Memory Interface:256-bit、Resolution:Digital Max Resolution 7680 x...

www.asus.com

Oh it's perfect...And we're back to the original problem: STOP MAKING ME BUY

 

[Josh128]

https://www.powercolor.com/product-detail212.htm

Reaper?

Seems like the Reaper has OEM clocks? Maybe this one will actually be MSRP? Surely AMD consulted with the AIBs to make sure at least a couple of them had at least 1 SKU at MSRP before making a worldwide make or break announcement about it?

 

[gdansk]

Seems like the Reaper has OEM clocks? Maybe this one will actually be MSRP? Surely AMD consulted with the AIBs to make sure at least a couple of them had at least 1 SKU at MSRP before making a worldwide make or break announcement about it?

It seems like there are some MSRP cards in China. If there aren't any in your market it may be a government trade policy problem.

 

[Saylick]

FSR4 has fallback mode for non-90 series, so technically FSR4 works with other cards.

What does that entail? It becomes FSR3.1 in fallback mode? Or does it revert to a lighter weight AI inferencing model that can be run on hardware with limited TOPS?

 

[In2Photos]

Liking all the different models available including several white models. The hellhound spectral has my eye if I decide to upgrade. Probably getting my son one first though since he has a GTX1660.

 

[RnR_au]

Liking all the different models available including several white models. The hellhound spectral has my eye if I decide to upgrade. Probably getting my son one first though since he has a GTX1660.

idk... seems to me from the beautiful pictures of the AMD reference cards, that the AIB's strong armed AMD into cancelling their reference efforts...

AIB's: "Too stylish!! Noone would buy our RGB rainbows! Please fire the only person with a clue in your famous Marketing department!".

 

[Thunder 57]

While that is generally true, we shouldn't wait for drivers to mature. AMD is a multi billion Dollar company, if drivers are garbage again than that is close to unacceptable. What you wrote basically sounds like we are at a point where it is acceptable for AMDs drivers to be bad at the beginning and it's great when they mature sometime in the future. However that is not what we should expect from a company like that.

Also looking at RDNA2: 6900XT had 11% more CUs than 6800XT without any other changes: Same clocks, same TDP. Yet it delivered like 8-9% perf at 4K over 6800XT at that same TDP. So RDNA4 scaling seems to be quite meh for now, we will see how N44 turns out later this year.

For now it could be a first indication why other RDNA4 was cancelled, at least seems more likely than just "GB202 too strong, we give up".

And just how well are Black(screen)well drivers doing? That old drivers argument is so tired it hasen't been true in 20 years.

 

[blackangus]

Im bummed, I really like the Sapphire Nitro + but I cant swing only 2 displayport ports.
Most of the rest of the cards are too gaudy for me.
The renders above would be right up my alley.

 

[poke01]

Pathetic.

https://imgur.com/OnhAJMK

These retailers have no stock and also no shame. The distributors raise prices. A trillion dollar company cannot control its own supply chain

 

[SolidQ]

These retailers have no stock and also no shame

5090?

 

[adroc_thurston]

A trillion dollar company cannot control its own supply chain

They had a late prod kickoff and no will to delay the launch any further. It happens.

 

[soresu]

but I cant swing only 2 displayport ports

Can't use MST?

 

[Kepler_L2]

Let's what they cook with RDNA5, that should be another major arch rework

I'm hoping it's not a major rework, given RDNA1 and RDNA3 both had big issues and RDNA2/4 were much better despite fewer major changes.

 

[SolidQ]

I'm hoping it's not a major rework, given RDNA1 and RDNA3 both had big issues and RDNA2/4 were much better despite fewer major changes.

Shoud be ok, if RDNA5 is PS6 GPU also

 

[Kepler_L2]

Not really.
Remember the quirk: for quite a lot of cases (particularly load-balanced PS waves) 6800XT behaves like a 64CU GPU.

Not the case with RDNA4 anymore, they support round-robin scheduling now

 

[adroc_thurston]

I'm hoping it's not a major rework, given RDNA1 and RDNA3 both had big issues and RDNA2/4 were much better despite fewer major changes.

Lmao it's RDNA4 that is massive.
They've bolted one limited OoO straight from CA-510 onto a GPU.

 

[Kepler_L2]

Lmao it's RDNA4 that is massive.
They've bolted one limited OoO straight from CA-510 onto a GPU.

Yeah but gfx13 does the real cocaine stuff like register renaming.

 

[adroc_thurston]

Yeah but gfx13 does the real cocaine stuff like register renaming.

you sure?
shieeeeeeeeeee

 

[Josh128]

Past my silly posts, I do like the huge effort where AMD was really lacking: media consumption, RT, better idle power draw...there's really a lot to like, all the little things that they were behind have been shored up seemingly.
My only 3 metrics left are:
- VR support speed (took SEVEN MONTHS for RDNA 3 to function like RDNA 2 did)
- ROCm support speed (Phoronix is already lamenting that months is too much, and it will be months)
- Broader RDNA 4 features support in 3rd party elements

For the latter, I'm really hopeful that some of the most well-known weaknesses in say Blender actually get pounded away, and that several of the other titles get support before 3 months have to pass. Premiere Pro, Photoshop, etc...

What was said about idle power draw? I must've missed it?

 

[branch_suggestion]

Lmao it's RDNA4 that is massive.
They've bolted one limited OoO straight from CA-510 onto a GPU.

Yeah but gfx13 does the real cocaine stuff like register renaming.

How long until we can start calling it a manycore CPU wrt scheduling paradigm?

 

[adroc_thurston]

How long until we can start calling it a manycore CPU wrt scheduling paradigm?

revenge of the Larrabee (good ending).