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

 

[MrTeal]

Hah, they're up on videocardz now.

Prices are off reddit, and there's no VAT in them. There is a 5% country-wide tax and provincial taxes between 0% and 8%, but those are calculated at checkout.

Re: from @adroc_thurston's regular point and on the same bapccanada subreddit. Typical GPU buyer.

 

[gdansk]

He's just pissy he didn't get what he wanted at the price he wanted. He's not done, he'll just buy Nvidia a month or two later. As usual.

 

[blckgrffn]

He's just pissy he didn't get what he wanted at the price he wanted. He's not done, he'll just buy Nvidia a month or two later. As usual.

And likely post something about regretting not just taking that ASUS card when he had the chance. smh.

 

[DAPUNISHER]

He's just pissy he didn't get what he wanted at the price he wanted. He's not done, he'll just buy Nvidia a month later. As usual.

Exactly. Whingers will only buy from one IHV; then wonder why prices are absurd.

 

[MrTeal]

He's just pissy he didn't get what he wanted at the price he wanted. He's not done, he'll just buy Nvidia a month or two later. As usual.

"I had to pay $1700 for a freaking 80 series. This is all AMD's fault."

 

[blckgrffn]

"I had to pay $1700 for a freaking 80 series. This is all AMD's fault."

"When will they ever get their act together and get nvidia to give me a better deal?"

Round and round we go.

 

[DAPUNISHER]

"I had to pay $1700 for a freaking 80 series. This is all AMD's fault."

It would not make me want to liver kick them, if it were not for supposed professionals echoing the same BS. Tim from HUB parrots this nonsense.

 

[ToTTenTranz]

Exactly. Whingers will only buy from one IHV; then wonder why prices are absurd.

Why does it hurt every time I shoot myself on the foot?

 

[linkgoron]

They could've said nothing. Then again, this forum would've criticized them for that too.

They should have said nothing. However, as they've answered I assume that they'll show that the perf/price is better than Nvidia and then say "SEE WE'RE LISTENING".

 

[gaav87]

I see on canada computers 5080 prices are ~19% above msrp this would mean 599$ for the XT would align more with the rumored 499$+100$ iv been hearing.

 

[gdansk]

They'll be at a competitive disadvantage (in reviews at least) if they don't copy Nvidia's fake MSRP technology.

 

[MrTeal]

I see on canada computers 5080 prices are ~19% above msrp this would mean 599$ for the XT would align more with the rumored 499$+100$ iv been hearing.

Not sure what you mean by this, that "MSRP" will be $499 but AIBs sell for an extra $100? Or that AMD shifted MSRP up $100 from their original plan to $599?

There are some MSRP models for 5080's as well, the vast majority are just 20% over MSRP.

https://www.canadacomputers.com/en/powered-by-nvidia/268152/zotac-gaming-geforce-rtx-5080-solid-16gb-gddr7-zt-b50800d-10p.html

Similar to Newegg (canada and US), where there are US$1000 MSRP 5080's, but they're 18th and 21st sorted by best sellers. All the ones selling ahead of them are 30%+ over MSRP.

Possible that none of these 9070XT's are MSRP, but the Reaper is Powercolor's base model.

 

[gaav87]

Not sure what you mean by this, that "MSRP" will be $499 but AIBs sell for an extra $100? Or that AMD shifted MSRP up $100 from their original plan to $599?

There are some MSRP models for 5080's as well, the vast majority are just 20% over MSRP.

https://www.canadacomputers.com/en/powered-by-nvidia/268152/zotac-gaming-geforce-rtx-5080-solid-16gb-gddr7-zt-b50800d-10p.html

Similar to Newegg (canada and US), where there are US$1000 MSRP 5080's, but they're 18th and 21st sorted by best sellers. All the ones selling ahead of them are 30%+ over MSRP.

Possible that none of these 9070XT's are MSRP, but the Reaper is Powercolor's base model.

Price for XT poor edition was suppose to be 499$ at least in late december.

 

[jpiniero]

Price for XT poor edition was suppose to be 499$ at least in late december.

I still assume it was intended to be $599 for the XT, with it being a 5070 competitor at $699.

 

[Win2012R2]

If there were no SKUs planned with GDDR7 then memory controller won't support it - it's not free, especially validating it with new VRAM, there is zero chance RDNA4 supports it, it was meant to be a stop gap measure until RDNA5/UDNA anyway, why would they waste money that they don't have?

 

[ToTTenTranz]

If there were no SKUs planned with GDDR7 then memory controller won't support it - it's not free, especially validating it with new VRAM, there is zero chance RDNA4 supports it, it was meant to be a stop gap measure until RDNA5/UDNA anyway, why would they waste money that they don't have?

The initial ambitions for RDNA4 were quite different from the N44+N48 duo we ended up with. It makes sense that AMD would develop the architecture with GDDR7 in mind if N4C was set to be a 9 SE behemoth.

As for what I read:

I know that N4C was going to use GDDR7 and that N48/N44 were only validated to use GDDR6. The question is if the memory controller is physically the same as N4C and how long GDDR7 validation would take.

GDDR7 is more cost for a very mainstream part so they won't be doing it.
But the IMC supports it iirc.

 

[coercitiv]

They could've said nothing. Then again, this forum would've criticized them for that too.

Wait until launch, it will be "fantastic" to see cards supposedly built for market share simply disappear from stores, while AMD tells us that high demand has taken them by surprise.

So, my number one priority right now is to build scale, to get us to 40 to 50 percent of the market faster. Do I want to go after 10% of the TAM or 80%? I’m an 80% kind of guy because I don’t want AMD to be the company that only people who can afford Porsches and Ferraris can buy. We want to build gaming systems for millions of users.

AMD’s Jack Huynh to Tom’s Hardware

 

[GodisanAtheist]

If there were no SKUs planned with GDDR7 then memory controller won't support it - it's not free, especially validating it with new VRAM, there is zero chance RDNA4 supports it, it was meant to be a stop gap measure until RDNA5/UDNA anyway, why would they waste money that they don't have?

- We know a lot of the RDNA4 line got scrapped, so its entirely possible that GDDR7 was on the roadmap right until it wasn't. Likely more expensive to take the functionality out at that point.

 

[Bigos]

It would not make me want to liver kick them, if it were not for supposed professionals echoing the same BS. Tim from HUB parrots this nonsense.

You mean "professional influencers", i.e. people that get paid to lie? Yeah, I would not trust anyone getting sponsorship from manufacturers...

 

[Saylick]

"When will they ever get their act together and get nvidia to give me a better deal?"

Round and round we go.

Tale as old as time.

 

[Mopetar]

AMD shot themselves in the foot again. A 96CU 32gb part would’ve really won the day.

It would be an amazing card that none of us could get at MSRP because with that much VRAM someone can fit their AI model into the card's memory.

It's not quite as bad as the mining boom where even lower tier GPUs still had enough memory for mining and thus massively inflated prices, but anything with a lot of VRAM is going to be good for AI and with more companies investing there's even greater demand for GPUs.

A midrange 16 GB card is probably safe from the most egregious use cases and might be purchasable at MSRP without having to wait weeks on end for availability. Let's just hope AMD doesn't decide to jack up that MSRP to stay inline with Nvidia offerings.

 

[linkgoron]

A midrange 16 GB card is probably safe from the most egregious use cases and might be purchasable at MSRP without having to wait weeks on end for availability. Let's just hope AMD doesn't decide to jack up that MSRP to stay inline with Nvidia offerings.

Why would they? It's all about the TAM and getting more cards into gamers' hands!

So, my number one priority right now is to build scale, to get us to 40 to 50 percent of the market faster. Do I want to go after 10% of the TAM [Total Addressable Market] or 80%? I’m an 80% kind of guy because I don’t want AMD to be the company that only people who can afford Porsches and Ferraris can buy. We want to build gaming systems for millions of users.

 

[Win2012R2]

so its entirely possible that GDDR7 was on the roadmap right until it wasn't

It wasn't for those chips that were not scrapped - nobody is going to put dual mem controllers into low end that is going for value

 

[IEC]

$729 would be DoA pricing. Even if it beat 5070 Ti in everything by a few % (which it won't, because AMD doesn't have MFG) people won't accept <20% discount over comparable nVidia product because the nV mindshare and technology advantage is so strong at this point.

Would have to approach 5080 levels of performance to justify a $729 price tag, and even then some nV diehards would just wait for a cheaper 5080.

 

[jpiniero]

would just wait for a cheaper 5080.

What 5080s? lol