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]

True, but still I'm not sure they couldn't keep up had they followed single-die design path

Keeping up isn't enough. You gotta win when you're behind.
You're not winning against NV with a monodie cuz NV also loves shipping big.

 

[Bryo4321]

Radeon 9070xt prime oc 3x-8power like iv been yapping about

View attachment 114355

Hunh, I didn’t really expect 3x 8 pin. Curious to see how it scales with extra power.

 

[Heartbreaker]

I don't know, it just feels like a half assed release, it just doesn't make sense overall.
GDDR7 brings a huge bandwidth increase, it is looking like a necessity, I never saw GPU vendors being so stingy with bus width, I mean, Nvidia is selling RTX 4070 Mobile cards with a 128 bit bus ffs! it probably scales terrible with new nodes, so bandwidth provided by memory seems very important. How is AMD going to compete with their 128 bit cards against Nvidia's 5060s 128 bit cards?

Waiting until NVidia Launches and significantly undercutting them on price?

Lot of reviews were very positive on the Intel B580 (yeah I know about driver overhead), because of it's price.

 

[tajoh111]

AMD might as well give up discrete or atleast trying to get marketshare. Missing CES for laptop makers is a big no no. AMD needed to launch their mobile variants if they wanted to gain significant markethare. With all companies announcing laptops, if AMD wanted to gain marketshare they needed to announce the product now so laptop makers can pair up their new laptops announced at CES with RDNA 4. By ignoring this market, it's no surprise AMD is in the low to mid teens for discrete.

Global gaming PCs and laptops revenue 2018-2029 | Statista

In 2024, gaming laptop revenues were estimated at over 31 billion U.S.

www.statista.com

IN Gaming laptops vs desktop revenue, laptops have higher numbers in both revenue and growth. So if Nvidia takes all of laptop discrete aside from strix halo portion, it just leaves AMD struggling to fight for 45% of the market. If Nvidia takes most of the desktop market, what is there left for AMD? Considering the size of RDNA4 chips, this should have been a market AMD prioritized into penetrating.

 

[adroc_thurston]

AMD needed to launch their mobile variants if they wanted to gain significant markethare

not after RDNA3.

 

[MrTeal]

Hunh, I didn’t really expect 3x 8 pin. Curious to see how it scales with extra power.

Be interesting to see what power draw actually is. There were plenty of 300W+ 3080s with 2x8pins, while some of the AIB 3x8pin models like my FTW3 could pull 400W gaming.

If these cards pull 350W+ and can't reach 4070 Ti Super levels of performance, that's going to be pretty disappointing.

 

[blckgrffn]

AMD might as well give up discrete or atleast trying to get marketshare. Missing CES for laptop makers is a big no no. AMD needed to launch their mobile variants if they wanted to gain significant markethare. With all companies announcing laptops, if AMD wanted to gain marketshare they needed to announce the product now so laptop makers can pair up their new laptops announced at CES with RDNA 4. By ignoring this market, it's no surprise AMD is in the low to mid teens for discrete.

Global gaming PCs and laptops revenue 2018-2029 | Statista

In 2024, gaming laptop revenues were estimated at over 31 billion U.S.

www.statista.com

IN Gaming laptops vs desktop revenue, laptops have higher numbers in both revenue and growth. So if Nvidia takes all of laptop discrete aside from strix halo portion, it just leaves AMD struggling to fight for 45% of the market. If Nvidia takes most of the desktop market, what is there left for AMD? Considering the size of RDNA4 chips, this should have been a market AMD prioritized into penetrating.

They launched some beefcake APUs instead, which honestly is better for them.

That way its not a system that could accommodate a "GPU" that almost is always an nvidia one but rather with that APU, no GPU needs to be added at all.

To OEMs that likely has some appeal.

And these APU's can't have worse attach rate than AMD mobile GPUs, can they?

 

[adroc_thurston]

To OEMs that likely has some appeal.

no the appeal is Mx Pro compete.

 

[Meteor Late]

But if I wanted better graphics before, I had the option to use an Nvidia laptop GPU, Apple Mx didn't do much in that regard.
Why would I buy an expensive APU like Strix halo to have 4060 like performance that can probably be had for like half the price? (we will see pricing).
Aside from the RAM appeal for some AI tasks, what else is there that makes it enticing?
And if it's an Mx Pro competitor, why is the name AI Max and not AI Pro?

 

[adroc_thurston]

not AI Pro?

Because AI Pro is already the name for their commercial parts.

 

[GodisanAtheist]

It's been a helluva rollercoaster today. Definitely going down as one of the strangest, underwhelming GPU announce (?) anyone has ever seen.

Nvidia couldn't have crashed AMD's hype train harder than AMD has at this point...

 

[adroc_thurston]

Nvidia couldn't have crashed AMD's hype train harder than AMD has at this point...

There wasn't any since no halo parts.

 

[Tuna-Fish]

Hunh, I didn’t really expect 3x 8 pin. Curious to see how it scales with extra power.

Same here. N48 is expected (and now officially marketed by AMD) to land somewhere in the 7900XT ballpark. 7900XT has a 315W TDP. They are moving from chiplet to unified, and getting a small boost from better litho. How in the world are they expecting to burn so much more power?

 

[adroc_thurston]

Same here. N48 is expected (and now officially marketed by AMD) to land somewhere in the 7900XT ballpark. 7900XT has a 315W TDP. They are moving from chiplet to unified, and getting a small boost from better litho. How in the world are they expecting to burn so much more power?

They don't but you gotta sell expensive 3 slot AICs somehow.

 

[Saylick]

It's been a helluva rollercoaster today. Definitely going down as one of the strangest, underwhelming GPU announce (?) anyone has ever seen.

Nvidia couldn't have crashed AMD's hype train harder than AMD has at this point...

I just hope coming out of this that the average enthusiast learn to stop blaming AMD for Nvidia's exorbitant pricing because AMD is simply too incompetent to make an impact. Sometimes, people play both sides of the argument where AMD is somehow simultaneously competent enough to make Nvidia lower prices but yet incompetent where they end up buying Nvidia over AMD anyways, which I find to be disingenuous and a thinly-veiled cop out for their desire to strictly use AMD as leverage for cheaper Nvidia GPUs. A part of me wants to see what consumers would said if AMD and Intel pulled out of the GPU market completely, because then there'd be no other excuse for Nvidia jacking up prices on people besides greed.

 

[SolidQ]

3x8 pin? If perfomance 4070ti super, that massive L for AMD.
RTX 4070ti super 285W and ADA is energy efficiency.

 

[adroc_thurston]

3x8 pin

That's not the ref.

 

[linkgoron]

Pathetic by AMD. RDNA2 was so promising, but it just went into the gutter after.

I can only assume that AMD is waiting for Nvidia to release their pricing so they can undercut by $50 or whatever. Jebaited!

 

[SolidQ]

Pathetic by AMD. RDNA2 was so promising, but it just went into the gutter after.

Seems like Jack is Raja2

That's not the ref.

Doesn't matter for that perf, too much. Unless 3x8 pin verison like 4080

 

[adroc_thurston]

Doesn't matter for that perf, too much

It's not a ref.

Unless 3x8 pin verison like 4080

It performs the exact same.

 

[Tuna-Fish]

AMD apparently did a round table with gaming press immediately after the event, with one subject "why did we not talk about RDNA4 yet?"

Ian has a writeup on it, but the short version is that AMD claims right now was just not the right time in the product lifecycle to do a major announcement. Still reading it, but seems fairly content-free.

 

[techjunkie123]

AMD apparently did a round table with gaming press immediately after the event, with one subject "why did we not talk about RDNA4 yet?"

Ian has a writeup on it, but the short version is that AMD claims right now was just not the right time in the product lifecycle to do a major announcement. Still reading it, but seems fairly content-free.

They also say the performance leaks are wrong lol....

I take this to mean no 4070 Ti Super levels of raster or RT.

 

[SolidQ]

from TPU
AMD's 3rd Gen Ray accelerator is expected to reduce the performance cost of ray tracing, by putting more of the ray tracing workload through dedicated hardware, offloading the SIMD engine. Lastly, AMD is expected to significantly upgrade the media acceleration and display I/O of its GPUs.

 

[Bryo4321]

AMD apparently did a round table with gaming press immediately after the event, with one subject "why did we not talk about RDNA4 yet?"

Ian has a writeup on it, but the short version is that AMD claims right now was just not the right time in the product lifecycle to do a major announcement. Still reading it, but seems fairly content-free.

According to Frank it sounds like there will be a near future presentation (videos?) on the architecture so that will be interesting.

 

[SolidQ]

not only 9070. N44 also

Tim Schiesser – Hardware Unboxed​

​

So with the FSR4 slide, there was an interesting footnote talking about some sort of feature that's going to be exclusive to the 9070 XT. Sites like Videocardz interpreted that to mean that FSR4 would be exclusive to the 9070 XT or 9070 series products (compared to 9060). Is that accurate reporting?

David McAfee​

​

It's more just those top tier platforms. The content that was shown during the CES presentation was on the 9070 XT, which is why the footnote said that specifically. There is nothing about FSR4 that makes it exclusive to a single model. It is very much architecture optimized at this point in time.

One of the major advancements in RDNA 4 is the MLOps being able to drive compute efficiently through the graphics engine is a massive generational improvement. I think what you should expect from AMD is as we roll out FSR4 is that it will lean into those capabilities of the Navi4 architecture.

They also say the performance leaks are wrong lol....

I take this to mean no 4070 Ti Super levels of raster or RT.

I don't think it's about 4070ti, i think it's more, looking how massive cards and 3x8 pins