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

 

[marees]

That's why I am mentioning a more scalable solution, primarly down-scaling from the current PT implementation so that more GPUs can actually run it at decent framerates.

I would also not base a GPU buying decision on PT performance. SW and algorithms can improve that performance much more than HW TFLOPS. Unreal Engine 5 is an example in that direction. Not actually path tracing but coming closer to it and at the same time quite good performance.

Yes Senua's Saga Hellblade 2 is example of well optimized UE 5 game

Vulkan based Id tech engine also performs well

Another option is Voxel based Cone Tracing implemented in Kingdom Come Deliverance 2

 

[adroc_thurston]

Yes Senua's Saga Hellblade 2 is example of well optimized UE 5 game

Well, uh, not much of a game there.

Vulkan based Id tech engine also performs well

Indeed.

Another option is Voxel based Cone Tracing implemented in Kingdom Come Deliverance 2

SVOGI isn't RT tho.

 

[TESKATLIPOKA]

I mean given the node differences that seems pretty trivial to overcome. Also N33 has a 128KB VREG vs N32/N31 192KB VREG so no idea how that plays a part in performance but the difference is there. It is not like CUs are the only difference.

7600(XT) was already working at very high clocks and 9700XT has 30.4% higher TFLOPs compared to 7800XT.
9600XT would need 3462-3593MHz to have the same increase in TFLOPs over 7600(XT). Not having castrated CUs like N33 and 20gbps will help in reducing this clock to more conservative levels.

Point is you can get into the nitty gritty if you want but it seems like massively overkill for working out the rough performance level a 9060XT might have. It should go without saying the error bars are quite large but I guess I will say it explicitly. 7700XT +/- 15% is probably the error bars you are looking at which is a range from 4060Ti at the lower end and 6800XT at the upper end. It will depend on final clocks, the final ram spec, tdp choices and so on.

I don't see how N44 could reach 6800XT level of performance. Even at Full HD that's still 57% difference and at 4K It's +68.4%.

N48 is a more competitive product than N32 so can command a higher price.

N44 will be in the 7700XT class +/- a step. The 7700XT can be had for around £380 in the UK and the 7600XT is £300. That means the max price for a 9060XT is probably around £320 if you want a 20% perf/$ advantage. At current exchange rates that would be around $340 USD when you take off VAT from the UK price.

If you think it will be $380 then the only way I can see that being the case is if performance is towards the upper bound of what I said which is 6800XT performance and that seems optimistic to me.

N44 16GB is also more competitive product than N33 16GB, which cost $329 at launch.
9700XT($600) vs 7800XT($440), perf/$ is only 8% better at 4K according to TPU.
If I calculate a 40-50% increase over 7600XT and add 8% better perf/£ I end up with £389-417.
If I calculate a 40-50% increase over 7600XT and add 20% better perf/£ I end up with £350-375.
If you want to claim that's too high because 7700XT costs only £380, then you should looks at the other advantages like more Vram, lower power consumption, better RT, better features.

Selling price does not always have a lot to do with BOM cost. N33 was also cheaper than N23 was but has sold for more than N23 parts did despite AMD having the option to keep margins steady and sell at a lower price.

I am sure N44 will probably cost more than N33 just through wafer pricing it will also be cheaper than N32 was so there is room for them to sell N44 for less than the 7700XT and still keep the same margins or slightly better than on N32.

That was my point from the beginning. AMD won't sell It for cheap, unless there is no demand or Nvidia suddenly starts a price war and not because they can't but because they don't want to.

 

[moonbogg]

Personally, I don't understand why one would allow path-tracing to influence current purchasing decisions.
It reduces $1600 GPUs (RTX 5080, RTX 4090) to sub 60 FPS at 1080p. Even if you're not using a poverty GPU you get a poverty GPU experience.

That leaves you with only one GPU to buy which allows a playable experience at an absurd price and power consumption. If you really want to use path-tracing you will need to buy a new GPU in a generation or two anyway.

It's a technology development initiated by Nvidia to create FOMO and poor GPU performance to ensnare more gamers into frequent GPU upgrades. The holy grail of high FPS and smoothness was already met for the foreseeable future. GPUs were lasting a long time in terms of raster performance, so Nvidia had to ruin that by creating this other garbage.

 

[moonbogg]

Well, uh, not much of a game there.

Indeed.

SVOGI isn't RT tho.

OK dude. What the hell is SVOGI? How many people here know what SVOGI stand for? I feel like you talk like that just to make people feel stupid.

 

[Timorous]

7600(XT) was already working at very high clocks and 9700XT has 30.4% higher TFLOPs compared to 7800XT.
9600XT would need 3462-3593MHz to have the same increase in TFLOPs over 7600(XT). Not having castrated CUs like N33 and 20gbps will help in reducing this clock to more conservative levels.

I don't see how N44 could reach 6800XT level of performance. Even at Full HD that's still 57% difference and at 4K It's +68.4%.

N44 16GB is also more competitive product than N33 16GB, which cost $329 at launch.
9700XT($600) vs 7800XT($440), perf/$ is only 8% better at 4K according to TPU.
If I calculate a 40-50% increase over 7600XT and add 8% better perf/£ I end up with £389-417.
If I calculate a 40-50% increase over 7600XT and add 20% better perf/£ I end up with £350-375.
If you want to claim that's too high because 7700XT costs only £380, then you should looks at the other advantages like more Vram, lower power consumption, better RT, better features.

That was my point from the beginning. AMD won't sell It for cheap, unless there is no demand or Nvidia suddenly starts a price war and not because they can't but because they don't want to.

N33 was overpriced in both flavours, should have been $250 / $300 at most.

The 7800XT otoh was the best performance/$ RDNA3 part.

As for clock speeds the 6600XT was half the size of the 6900XT and had 300Mhz higher clocks with a 160W TDP.

The 9070XT hits around 3.1Ghz so if the half size N44 with a 160W TDP manages the same absolute uplift N23 managed it would be ~3.4Ghz.

The other factor is that NV left the door open for the 9070XT to land in, that won't be quite so easy for the 9060XT with a 5050, 5060, 5060Ti 8GB and 5060Ti 16GB in the $250 - $500 bracket.

I don't think $330-$350 is cheap tbh, it is inline with the 9070XT maybe a bit better which you expect from the cheaper parts.

 

[adroc_thurston]

What the hell is SVOGI?

Voxel-based conetracing.

How many people here know what SVOGI stand for?

well it's the graphics section, I expect people to have the basic understanding of computer gfx.

 

[moonbogg]

Voxel-based conetracing.

well it's the graphics section, I expect people to have the basic understanding of computer gfx.

SVOGI means voxel-based cone tracing? What does the S, O, G, and I stand for?

Standard Voxel Original God Imaging?

 

[gaav87]

Anyway wonder how gigabyte aorus elite scores.
Asrock cheaped out on thermal pads and did thermal paste instead of ptm... 40C diff between gpu temp and hotspot is crazy.
Also 414W xD

 

[ryanjagtap]

SVOGI means voxel-based cone tracing? What does the S, O, G, and I stand for?

Standard Voxel Original God Imaging?

Sparse Voxel Octree Global Illumination (SVOGI)

 

[GodisanAtheist]

Looks like Adroc got another one.

@moonbogg don't pay Adroc too much mind, he's just the resident Autist Extraordinaire.

The superior attitude and curt/dismissive responses are just part of the territory.

 

[GaiaHunter]

Quick AMD is doing decent RT and upscaler. The new talking point is Path Tracing!

Anyone remember the most important future technology PhysX?

 

[blackangus]

...in the future

As long as there are not more GPUs with more performance or a more scalable path tracing solution, adoption will be slow. Game developers prioritize wide usability over maximum graphical fidelity.

But AMD could improve that:
- Provide something like Ray Reconstruction as an open solution to GPUOpen. RR improves quality and provides an additional scalability lever (less samples-per-pixel rendered, but compensated to a good degree by RR)
- Release a scalable PT GI solution akin to ReSTIR, but improved (e.g. Area ReSTIR + ShaRC + scalable samples-per-pixel). Successor to their GI-1.0-Sample or Brixelizer GI (see GPUOpen)

ReSTIR is scalable today. But only 1...N samples-per-pixel and not less. At least in their RTXGI git sample it is like that.

And Nvidia could improve the situation by ceasing to create a walled feature garden and contribute to open solutions, arguably far more than AMD ever could because Nvidia is the market leader.
But...

 

[gaav87]

Anyway 9070xt can boost to 3725mhz... overclocking is bugged i figured out how the stupid offset works.
Clock are tied to hot-spot temp. Ofset at +-0mhz =3250mhz boost.
I can do 3625mhz w/o crashing in some games and 3650mhz is no go.
Stupid asrock i have 40C delta between gpu temp and hotspot... 40C !!

 

[blackangus]

Why nobody does PT if not sponsored by Nvidia?

Because currently path tracing is a marketing mechanism designed to invoke FOMO.
Developers dont have time for that kind of stuff without any real return on the investment.

Edit: LOL! I posted this before I saw Moonboggs post above

 

[gdansk]

Stupid asrock i have 40C delta between gpu temp and hotspot... 40C !!

You mean they didn't remove the hotspot temperature reporting? Was that Nvidia who did that then?

 

[blackangus]

4090 had great sales and 5090 would have had too had they hit at least +50% real perf, and more than that in ray tracing.

Id say it sold pretty poorly at a whole .7% of the market! =)

 

[blackangus]

well it's the graphics section, I expect people to have the basic understanding of computer gfx.

You expect too much, this isnt B3D.
And... I mean... that is pretty esoteric =)

 

[adroc_thurston]

You expect too much, this isnt B3D.

Yeah but SVOGI is so easy to google too.
The first result is Crytek's docs no less.

 

[blackangus]

Yeah but SVOGI is so easy to google too.

This I do not deny =)

 

[gaav87]

Because currently path tracing is a marketing mechanism designed to invoke FOMO.
Developers dont have time for that kind of stuff without any real return on the investment.

Edit: LOL! I posted this before I saw Moonboggs post above

Fabio did optiscalar mod + path tracing 1440p Ultrawide all maxed no frame gen
66fps not even overclocked stock i can do ~77fps slap on FG and ur 144fps
on amd path tracing xD

 

[blackangus]

Fabio did optiscalar mod + path tracing 1440p Ultrawide all maxed no frame gen
66fps not even overclocked stock i can do ~77fps slap on FG and ur 144fps
on amd path tracing xD

View attachment 119688

What does any of this have to do with my comment?

 

[sandorski]

Sparse Voxel Octree Global Illumination (SVOGI)

Can that gobblygook run Crysis?

 

[inquiss]

Quick AMD is doing decent RT and upscaler. The new talking point is Path Tracing!

Anyone remember the most important future technology PhysX?

Absolutely. Only have to see the brain rot posts here occasionally talking about it like it's an important feature for this generation.

Very clever from Nvidia but very silly to gobble it up

 

[moonbogg]

Yeah but SVOGI is so easy to google too.
The first result is Crytek's docs no less.

I refuse to google anything when having a discussion. That's cheating. I either know it or I don't. Anyone can google to get an answer. SVOGI is the most ridiculous thing anyone has ever heard of. That has nothing to do with FPS. And you said you expect me to know my stuff? You can be forgiven because you probably didn't know that I'm the David Goggins of these forums. I stay hard to argue with, always. You have any idea what I did to @zinfamous ? He challenged me and I embarrassed that boy in front of everyone with a chicken foot. Also, I'm kind of cranky due to my new OMAD routine. Let me guess, you had to google that to pretend like you knew what it meant. Nice.

Members callouts are not permitted.

Mod DAPUNISHER