Discussion RDNA4 + CDNA3 Architectures Thread

DisEnchantment · Mar 23, 2022

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

igor_kavinski · May 13, 2024

marees said:
if they drop the ball on discreet gaming then Intel could swoop in for a console win (Xbox?), assuming Intel has more money to burn

No, I don't think Intel can give a decent APU to console makers at desirable prices. If RTG dies, console makers may go for ARM SoCs.

Mahboi · May 13, 2024

Wait until they announce LIsA Su, for Large Intelligence's Artificial Superior.

igor_kavinski · May 13, 2024

The LeeSA PeeCee!

ToTTenTranz · May 13, 2024

Mahboi said:
A very interesting thing to observe will be how often Strix Point is accepted as "solo" vs "driver of an NVidia GPU".
Already we've seen 4050s into STX lappies and I don't see it as a very good sign.

I think Strix Point is probably going to be regarded as having the best laptop CPU out there, considering its 12x Zen5 cores with performance and thermals superior to anything Intel will have available this year. AMD also equipped it with 16x lanes PCIe 4.0 which is good enough for any dGPU.

That said, I'm fully expecting many of the higher end laptops with >RTX 4070 to use Strix Point, regardless of its iGPU.

branch_suggestion said:
I hope Halo does make its way into machines that would usually have a low/mid spec dGPU.

Strix Halo is targetting RTX 4070 laptop performance in rasterization. It's not coming to machines with low/mid spec dGPUs. I doubt AMD will want Strix Halo in sub $1200 laptops, at least its full 256bit + 20WGP variant.

marees said:
I still see lots of scope for AMD to sell discrete GPUs. (Loss of laptop market is a big deal tho)

This time AMD developed 3 different SoCs with RDNA3.5 iGPUs covering all performance brackets from the RTX2050 to the RTX4070. One of them is an unforeseen MCM solution using 256bit LPDDR (which is a first for laptop x86 IIRC).

Why are you claiming they're losing the laptop market if they never invested this many resources into specific laptop solutions?

marees · May 13, 2024

ToTTenTranz said:
I think Strix Point is probably going to be regarded as having the best laptop CPU out there, considering its 12x Zen5 cores with performance and thermals superior to anything Intel will have available this year. AMD also equipped it with 16x lanes PCIe 4.0 which is good enough for any dGPU.

That said, I'm fully expecting many of the higher end laptops with >RTX 4070 to use Strix Point, regardless of its iGPU.

Strix Halo is targetting RTX 4070 laptop performance in rasterization. It's not coming to machines with low/mid spec dGPUs. I doubt AMD will want Strix Halo in sub $1200 laptops, at least its full 256bit + 20WGP variant.

This time AMD developed 3 different SoCs with RDNA3.5 iGPUs covering all performance brackets from the RTX2050 to the RTX4070. One of them is an unforeseen MCM solution using 256bit LPDDR (which is a first for laptop x86 IIRC).

Why are you claiming they're losing the laptop market if they never invested this many resources into specific laptop solutions?

Bulk of the laptop market is Intel CPUs. & AMD has lost the GPU for that market because of RDNA 3's poor efficiency vs Ada Lovelace

adroc_thurston · May 13, 2024

Kepler_L2 said:
For now OEMs will stick with ol' green reliable.

ugh no, -halo has design wins.

igor_kavinski said:
I don't think RTG was ever properly integrated into AMD.

yes they are.

igor_kavinski said:
They are probably treated as 2nd class citizens compared to the CPU teams.

no they're stars of the show.

del42sa said:
neverending story and it´s hardly change unless RTG changes significantly their attitude ..

nope?
unless you consider wccf articles real.

maddie said:
but revenue is possibly even more critical for continued existence

no, margin is the only thing that matters.

ToTTenTranz said:
That said, I'm fully expecting many of the higher end laptops with >RTX 4070 to use Strix Point, regardless of its iGPU.

the luggables will use fire range.
everyone else halos or dies. simple!

marees said:
assuming Intel has more money to burn

they do not.
Capex-heavy company with no revenue streams of note.

Mahboi · May 13, 2024

adroc_thurston said:
nope?
unless you consider wccf articles real.

Don't make me go back to paint.net and make another "wccf reported as not real, all articles and posters are actually creations from Silent Hill meant to torture tech people".

maddie · May 13, 2024

adroc_thurston said:
no, margin is the only thing that matters.

Nope.

adroc_thurston · May 13, 2024

maddie said:
Nope.

Yeah.
Why do you think products like strix-halo exist?

igor_kavinski · May 13, 2024

adroc_thurston said:
no they're stars of the show.

You may be referring to the enterprise guys responsible for MI300X. Is the consumer GPU architecture team well respected there too?

Mahboi · May 13, 2024

adroc_thurston said:
Yeah.
Why do you think products like strix-halo exist?

You don't seriously think they'll be making mad margin on this thing.

TESKATLIPOKA · May 13, 2024

adroc_thurston said:
ugh no, -halo has design wins.

Halo is an APU and @Kepler_L2 was talking about mobile dGPUs.

blckgrffn · May 13, 2024

Mahboi said:
You don't seriously think they'll be making mad margin on this thing.

Compared to small die, small run dGPU margins that get slim to none design wins? I would think yeah, definitely, especially as AMD APUs seem quite prevalent in laptops - certainly compared to just a few years ago.

dGPU outside of the lugger class laptops dying is written on the wall, IMO.

If Intel wants to compete in the laptop space, an APU style approach with a much bigger "integrated" GPU is probably their meal ticket too.

The 2050/3050/4050 class of GPUs are on borrowed time now, again IMO.

adroc_thurston · May 13, 2024

igor_kavinski said:
You may be referring to the enterprise guys responsible for MI300X.

No.
Just GPU guys.

igor_kavinski said:
Is the consumer GPU architecture team well respected there too?

yes they just have to execute.
It's not some miracle or magic. Just deliver on what's promised.
more RDNA2's and less RDNA3's. Pretty simple right?

Mahboi said:
You don't seriously think they'll be making mad margin on this thing.

the whole selling point.

TESKATLIPOKA said:
Halo is an APU and @Kepler_L2 was talking about mobile dGPUs.

strictly pitched as discrete class graphics replacement.
Please don't argue laptops with me. Futile.

TESKATLIPOKA · May 13, 2024

adroc_thurston said:
strictly pitched as discrete class graphics replacement.

That is still limited to 4070 level of performance and Nvidia's next gen probably won't be too far away from Halo's release, at least I think so.
So I have to wonder how will It fare against GB207 for example.

adroc_thurston · May 13, 2024

TESKATLIPOKA said:
That is still limited to 4070 level of performance and Nvidia's next gen probably won't be too far away from Halo's release

Yeah.

TESKATLIPOKA said:
So I have to wonder how will It fare against GB207 for example.

Client blackwells can't spam Si, they're doomed to incremental perf gains.

maddie · May 13, 2024

adroc_thurston said:
Yeah.
Why do you think products like strix-halo exist?

Might as well ask why do the console chips exist?

Joe NYC · May 13, 2024

TESKATLIPOKA said:
That is still limited to 4070 level of performance and Nvidia's next gen probably won't be too far away from Halo's release, at least I think so.
So I have to wonder how will It fare against GB207 for example.

What percentage of mobile dGPUs are <= 4070 in your opinion?

My guess is about ~75% units, if not more.

So there is more than enough market potential for Strix Halo to compete for. No point to even think about higher tier of dGPU market before enough of this ( <= 4070) market is gained.

adroc_thurston · May 13, 2024

maddie said:
Might as well ask why do the console chips exist?

Because it's a zero-effort captive market.
Please just listen to Lisa.

Joe NYC said:
What percentage of mobile dGPUs are <= 4070 in your opinion?

My guess is about ~75% units, if not more.

So there is more than enough market potential for Strix Halo to compete for. No point to even think about higher tier of dGPU market before enough of this ( <= 4070) market is gained.

ohhhh look someone here is smart enough to grasp the whys of things.

blckgrffn · May 13, 2024

adroc_thurston said:
Because it's a zero-effort captive market.
Please just listen to Lisa.

ohhhh look someone here is smart enough to grasp the whys of things.

Hey now. There are ~~thousands~~ ~~hundreds~~ tens of us.

Mahboi · May 13, 2024

blckgrffn said:
Compared to small die, small run dGPU margins that get slim to none design wins? I would think yeah, definitely, especially as AMD APUs seem quite prevalent in laptops - certainly compared to just a few years ago.

dGPU outside of the lugger class laptops dying is written on the wall, IMO.

If Intel wants to compete in the laptop space, an APU style approach with a much bigger "integrated" GPU is probably their meal ticket too.

The 2050/3050/4050 class of GPUs are on borrowed time now, again IMO.

How does that work exactly?
Laptop dGPUs are at least away from the CPU, so you can cool each with a different fan/heatsink. APUs would have to take all the heat to one spot, that sounds like it'll be harder to make it live than 2 dies.
Also APUs are pretty quickly limited, and that includes Halo. We're looking at 40 CUs tops, while 4090 mobile is at 76 SMs.

For small dGPU with meagre margins, sure it's interesting, but considering that they're making small margins, how is Halo supposed to earn a lot more margin for roughly the same price/perf? How much cheaper is a full APU with advanced packaging over 2 small dies? How much more money can you pull away that you can't pull with 2 dies?

adroc_thurston said:
the whole selling point.

Yeah same question then. How much more money can one complex APU gain over 2 small to mid size dies in laptop? Especially with the heat concentration on one spot, etc.

Mahboi · May 13, 2024

Joe NYC said:
What percentage of mobile dGPUs are <= 4070 in your opinion?

My guess is about ~75% units, if not more.

So there is more than enough market potential for Strix Halo to compete for. No point to even think about higher tier of dGPU market before enough of this ( <= 4070) market is gained.

Same as desktop GPU where the >=80 class is not even 25% of the market.
Arguably worse even since a ton of gamers don't wanna bother themselves with a laptop with a "strong" GPU.
You can probably bet ~90% of units is 70 class down to APU and you won't be far off.

I'm not worried about the market, I'm doubtful about our resident CS:GO minded leaker who keeps mentioning margins.
APUs are not going to be a silver bullet to replace dGPUs, just another compact way of reaching the same result. Maybe with a special memory coherency between CPU and GPU, or with whatever extra stuff AMD wants to cram in them in the long run, will see dGPUs become an obsolete thing and AMD will reap the entire market.

But on the first go of a strong APU? How's that supposed to work out to good margins?
You're cramming 40 CUs into a compact die alongside 4 Z5 cores, 8 Z5c cores, 4 Z5LP cores. That is a lot of silicon to feed in a laptop even if it's a fat one, especially all under one fan.
I don't picture this thing being cheap to make, I don't picture the packaging being cheap, nor do I picture the cooling system being cheap. Maybe over time you can streamline and have a ton of reused designs, but I can't picture it being high margin on the first go, more like an expensive experimentation that'll get cheaper as time goes on.

Joe NYC · May 13, 2024

Mahboi said:
How does that work exactly?
Laptop dGPUs are at least away from the CPU, so you can cool each with a different fan/heatsink. APUs would have to take all the heat to one spot, that sounds like it'll be harder to make it live than 2 dies.
Also APUs are pretty quickly limited, and that includes Halo. We're looking at 40 CUs tops, while 4090 mobile is at 76 SMs.

For small dGPU with meagre margins, sure it's interesting, but considering that they're making small margins, how is Halo supposed to earn a lot more margin for roughly the same price/perf? How much cheaper is a full APU with advanced packaging over 2 small dies? How much more money can you pull away that you can't pull with 2 dies?

Yeah same question then. How much more money can one complex APU gain over 2 small to mid size dies in laptop? Especially with the heat concentration on one spot, etc.

Avoid duplication (memory, cooling, packaging), resulting in lower BOM.

BTW, 4060 is not trivial die size - 159 mm2. And it needs to be packaged and paired with a separate VRAM...

Mahboi · May 13, 2024

Joe NYC said:
Avoid duplication (memory, cooling, packaging), resulting in lower BOM.

Memory ok.
Cooling? How is cooling 2 60-80W chips worse than cooling 1 120W chip?
Don't you just exchange a fan for the need of a much bigger heatsink and fan?
Packaging meaning what in this context?

Joe NYC said:
BTW, 4060 is not trivial die size - 159 mm2. And it needs to be packaged and paired with a separate VRAM...

Ok for memory and extra VRMs or so, but again, if you remove that from the the dGPU, you're just setting something of roughly same power directly into the APU.

Wait now that I think on it, does STX-Halo include integrated LPDDR5 or the laptop makers need to integrate that aside? Cause if its all-in-one, even if the packaging is complicated, I can see a lot of laptop makers saying "whew, problem out of my hair" and just focus on the cooling and sending it out to sell.

adroc_thurston · May 13, 2024

Mahboi said:
How is cooling 2 60-80W chips worse than cooling 1 120W chip?

Well you need more copper for that.

Mahboi said:
Don't you just exchange a fan for the need of a much bigger heatsink and fan?

No.

Mahboi said:
you're just setting something of roughly same power directly into the APU.

no.

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Lifer

Senior member

Lifer

Member

Member

Diamond Member

Senior member

Diamond Member

Diamond Member

Lifer

Senior member

Platinum Member

Diamond Member

Diamond Member

Platinum Member

Diamond Member

Diamond Member

Platinum Member

Diamond Member

Diamond Member

Senior member

Senior member

Platinum Member

Senior member

Diamond Member