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

 

[jpiniero]

I wouldn't surpise if AMD repeat same mistake like 7700XT/7800Xt 50$ difference

That wasn't a mistake. They wanted you to buy the 7800 XT and not the 7700 XT. Either that's the reason (they want you to buy the 9070 XT) or the 9070 XT will have a Fake MSRP.

I still say they should only launch one model, one or the other.

 

[adroc_thurston]

or AMD was thinking about 550 for 9070. I wouldn't surpise if AMD repeat same mistake like 7700XT/7800Xt 50$ difference

That wasn't a mistake brother, you're supposed to buy 7800XT.
Chop was uneconomical.

 

[blckgrffn]

That wasn't a mistake brother, you're supposed to buy 7800XT.
Chop was uneconomical.

That seemed pretty obvious from the jump. It wasn’t really supposed to replace the 6700xt, they never made a purpose built SKU for that. Unless the 7600XT was supposed to be a lot faster than it was…

 

[adroc_thurston]

Unless the 7600XT was supposed to be a lot faster than it was…

No it was supposed to be more efficient.

 

[SolidQ]

Please tell me these are just placeholders…

it's. Looks at this prices between different sku's

 

[HuesToo4]

Sorry brother you ain't touching that DVFS network.

Is it locked on just the 9070 or 9070 XT as well? If so it looks like hardware modding is the only way.

 

[adroc_thurston]

Is it locked on just the 9070 or 9070 XT as well

Yes. As it should be.

If so it looks like hardware modding is the only way.

I think the SMU will whack you for doing that.

 

[HuesToo4]

I think the SMU will whack you for doing that.

It's possible on the 6900XT and 7900XTX with EVC2SE mod, but we'll see.

 

[ToTTenTranz]

Vega 56 all over again, but AMD will probably stop the poors from having fun.

N48 does not support FP4, only INT4.

Is there even any point in FP4?

I mean.. besides the literal floating point..

 

[jpiniero]

Is there even any point in FP4?

I mean.. besides the literal floating point..

AI

 

[gaav87]

9070 non-xt = 5070 price
xt= 599$
my last price ramble and il shut it about price.

 

[PJVol]

I would not compare radeonsi with Windows drivers.

Sure, and still can't say I'm very impressed

I think the SMU will whack you for doing that.

For the time being it was successfully fooled by tapping into SVI pipe.

 

[marees]

9070 non-xt = 5070 price
xt= 599$
my last price ramble and il shut it about price.

$500 & $650 for the base models

 

[exquisitechar]

$500 & $650 for the base models

The 9070 won’t be that much cheaper than the 9070 XT.

 

[Thunder 57]

Still going with $550/$650. Ok, maybe $500/$650, but I doubt that. It'd kill XT sales.

 

[PJVol]

Still going with $550/$650. Ok, maybe $500/$650, but I doubt that. It'd kill XT sales.

Ehh...

Radeon RX Vega Unveiled: AMD Announces $499 RX Vega 64 & $399 RX Vega 56

 

[Josh128]

^^
Yeah that was like 8 years ago, those days are gone, never to return my dude. Also was on the el cheapo GloFo / Samsung 14nm process.

$599/$699 still my prediction.

 

[MrTeal]

Still going with $550/$650. Ok, maybe $500/$650, but I doubt that. It'd kill XT sales.

Obviously Nvidia's MSRP is a very very loose suggestion, but the 5080 is only like 15% faster than a 5070 Ti and is 33% more expensive. If the 9070 XT ends up being ~15% faster than a 9070, 500/650 would be similar.
I'd also be surprised though, unless yields are atrocious and they can't get enough top bin dies.

 

[blackangus]

All for it, as long as it isn't the Voodoo of 3dfx kind ...

RGSS was awesome!

 

[blckgrffn]

Still going with $550/$650. Ok, maybe $500/$650, but I doubt that. It'd kill XT sales.

I don’t think anything will kill XT sales at or less and 5070ti MSRP other than actual stock availability of the 5070ti somewhere near its MSRP.

Piles of 5070 cards at $600 and less will also change the dynamic. Right now you can’t buy any Nvidia card north of the 4060 reliably.

I feel @coercitiv was spot on with this, the market is crazy. Literally what they have will sell, that’s my $.02.

 

[IEC]

$629/$699 my prediction. Because AMD never misses an opportunity to miss an opportunity on Radeon pricing.

 

[gdansk]

I predict Nvidia -$50. So 700/500.

 

[Saylick]

Is there even any point in FP4?

I mean.. besides the literal floating point..

If you want to really get into it...

**1. Same Number, Different Flavors**When documentation mentions 4 bits, it’s like saying you’re using “flour.” But *which* flour?- **INT4**: The all-purpose flour—straightforward and widely used- **FP4**: A specialty flour—same size bag, but with different properties

- 4-bit formats come with intriguing trade-offs: - **INT4**: Excellent for weight storage in large models (e.g., GPT-Q 4-bit quantization), capturing patterns in weights effectively. - **FP4**: Offers a broader numeric range, making it great for activation values that need more dynamic flexibility.

https://twitter.com/x/status/1876804594979987768

 

[Makaveli]

View attachment 118209View attachment 118210

7900XTX vs 9070XT

20732 vs 20904

0.8% difference so basically tie.


If we ignore win vs Linux

Open GL
XTX has a 17% lead



Direct X
XTX has a 20% lead

 

[SolidQ]

Price not bad for whole PC

https://videocardz.com/newz/first-gaming-pcs-with-radeon-rx-9070-xt-listed-on-newegg-price-starts-at-1799