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

 

[adroc_thurston]

Wonder what could happen if you slap a 28G GDDR7 onto Navi48 and ramp the TBP to 350W.
Wouldn't that be really funny.

 

[msj10]

yes. per-CU perf is waaaaay up.

so it caught up to Ada's PPA 2 years later.
congrats I guess

 

[adroc_thurston]

so it caught up to Ada's PPA 2 years later

Did it?

 

[msj10]

Did it?

if it's 4080 performance at 390 mm2 then yeah?

 

[adroc_thurston]

if it's 4080 performance at 390 mm2 then yeah?

It's not 390mm^2.

 

[SolidQ]

DLSS4 MFG is pure ML slop.

for RDNA5 matrix cores?

if you slap a 28G GDDR7 onto Navi48 and ramp the TBP to 350W.

or just 80CU with same clocks, would be interesting too see

 

[adroc_thurston]

for RDNA5 matrix cores?

ughhh no?

or just 80CU with same clocks, would be interesting too see

bandwidth matters here.

 

[SolidQ]

ughhh no?

AMD gonna answer to MFG, only question in RDNA4 or RDNA5

 

[Bryo4321]

Wonder what could happen if you slap a 28G GDDR7 onto Navi48 and ramp the TBP to 350W.
Wouldn't that be really funny.

I would upgrade my 7900xtx is what. Yes please.

Really really impressive. It seemed to hard to believe with a 64cu part, but dang. Those numbers are impressive.

 

[adroc_thurston]

AMD gonna answer to MFG, only question in RDNA4 or RDNA5

they can do it with FSR3 lmao.

I would upgrade my 7900xtx is what

Wouldn't even be big of an upgrade perf-wise, but still cool.

Really really impressive. It seemed to hard to believe with a 64cu part, but dang. Those numbers are impressive.

Yea. Turns out they don't always fail.

 

[SolidQ]

they can do it with FSR3 lmao.

I know, but without ML it would be terrible quality i'd assuming

 

[MarutTheSlayer]

AMD gonna answer to MFG, only question in RDNA4 or RDNA5

AMD must double down on more graphical compute power for superior raster. If they do another decent step in RT, that is a double win - it is perfect for consoles, as well as gamers who don't want to bother with latency and lag, and too much fake frames.
Frame gen is useful for CPU limited scenarios or 144-165 Hz refresh, but DLSS makes for lazy optimization and unfinished products that are never quite fully fixed even a year after launch.

 

[Tup3x]

If what you say is true the 5070 is DoA, which it should be regardless because of its crippled VRAM.

If they lower the price to 349-399 € (inc. VAT) and rename it to 5060 Ti, it would be fine but not like this. Card with 12 GB of VRAM is so crippled that >400 € pricing is ridiculous. AMD is very likely trying to make it obsolete with their pricing.

 

[Timorous]

yes. per-CU perf is waaaaay up.

yea, it's exactly ~7900XTX perf now.

ding-ding-ding, with Blackwell being what it is (poo), AMD can price whatever. $549 is the optimistic pick.

Given how far off RDNA3 leaks were and how Zen 5 performance did not transfer from server to client for reasons I am waiting for benchmarks.

The raw specs look like it will be 7800XT +20% ish which is between 7900GRE and 7900XT. If it is faster then great that will be a nice surprise.

 

[adroc_thurston]

I am waiting for benchmarks.

They got poasted like 2 pages ago.

The raw specs look like it will be 7800XT +20% ish which is between 7900GRE and 7900XT.

Turns out new uarch means new uarch in this case.

If it is faster then great that will be a nice surprise.

You can just scroll up and the numbers are right there.

 

[DavidC1]

If those leaks are true then AMD has a clearly superior product.

Yea a halo part would have done Nvidia in.

 

[gaav87]

For ppl who are to lazy to scroll
AIB to the moon hoo hoo

 

[adroc_thurston]

AIB to the moon hoo hoo

Nope.
But if they refresh it with GDDR7 and higher TBPs? Ayup.

 

[SolidQ]

But if they refresh it with GDDR7

I'd assuming that possible, but first need to see 5080 result's, if real 4080 + 15%, should be easy for AMD.
Hitman 3 and F12023, maybe driver will improve perf

 

[adroc_thurston]

I'd assuming that possible, but first need to see 5080 result's, if real 4080 + 15%, should be easy for AMD.

GDDR7 will at best give N48 10% more perf, but perf is perf and it makes the world go round.

 

[SolidQ]

GDDR7 will at best give N48 10% more perf

Call it 9070 XT super or 9070XT Max?
Wonder how many % add final driver to December leak

 

[Timorous]

They got poasted like 2 pages ago.

I saw them. I also saw the RDNA 3 benchmarks AMD showed at their reveal event, they were also very misleading so I have little faith in the leaks above

Turns out new uarch means new uarch in this case.

Well I hope it turns out well, I just won't be riding the hype train like last time. I can wait for 3rd party benchmarks.

 

[del42sa]

https://videocardz.com/newz/asus-co...n-greek-retailer-list-january-23-availability

23th January ? withouth official launch and specs ? weird ...

 

[SolidQ]

I also saw the RDNA 3 benchmarks AMD showed at their reveal event

That different story AMD RDNA3 was like "UP x 1.7" no numbers
Here is FPS numbers and remember it's leak, not official announce from AMD

 

[adroc_thurston]

Call it 9070 XT super or 9070XT Max?

Whatever.

Wonder how many % add final driver to December leak

5% tops.

I also saw the RDNA 3 benchmarks AMD showed at their reveal event, they were also very misleading so I have little faith in the leaks above

This isn't any reveal event, it's their internal testing against comp and their previous gen.

I can wait for 3rd party benchmarks.

This cope never works when you have concrete FPS numbers 2 pages down.

That different story AMD RDNA3 was like "UP x 1.7" no numbers
Here is FPS numbers and remember it's leak, not official announce from AMD

It loses to N31 in 2 bandwidth-bound titles anyway.