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

Josh128 · Mar 18, 2025

Kepler_L2 said:
Navi4C was cancelled in late January 2023.

Which logically would have taken place once they determined from developing RDNA3 that chiplet GPUs still have too many challenges. Cancelling chiplet 4C was likely a great call because of that. Chiplet= no bueno vs mono.

Kepler_L2 · Mar 18, 2025

Josh128 said:
Which logically would have taken place once they determined from developing RDNA3 that chiplet GPUs still have too many challenges. Cancelling chiplet 4C was likely a great call because of that. Chiplet= no bueno vs mono.

There is no relationship at all between Navi4C and Navi31/32.

gaav87 · Mar 18, 2025

Anyway cant post screenshots for some reason.
Anyone with 9070xt if you want to see how broken the overclocking is
go to heaven benchmark near the baloon (free camera)
Set -25mV or -30mv or -35mV and +275mhz clock offset (leave the power and memory stock)
U will boost to 3550mhz then try -40mV or 0mV BAM 3300mhz
I crash at 3650mhz at -30mV or -35mV any other mV settings im at either 3250 or 3300mhz... Broken as hell...

gaav87 · Mar 18, 2025

- Same thing in helldivers 2...
I can fiddle with the offset and mV and somehow trick "force" the card to run at 3550mhz stable for few min...
Sometimes the card is stuck at 3150mhz and 320W even with +10% power limit... should be up to 410W (highest i saw in furmark)
hotspot at 75C (forced fans at 80%)... .
- I force reset drivers and same spot on the ship in helldivers bam 374W 3550mhz and crash after few min xD
FPS with stuck boost clocks 144 at 3150mhz and 157-158 with ~3550mhz "unstuck"

- steel-nomad 3dmark seem to be broken... You can have 2950mhz at 290W or 378W+ with exact same perf. Its like the clock is stuck at 2950mhz and 890mV !!!
Like what the heck is this. Sometimes im able to "unstuck" it and gain like 5% perf its boosting to 3150mhz netting me low top100 leaderboard...

-Same thing i did in furmark when i somehow unstuck the clock and did 5% faster than rtx 4090D performance
furmark: Stock 343fps, normal overclock 373fps, unstuck oc 400fps
This is madness hope they fix this...

- I have also noticed that the voltage is also stuck... Card is doing 1050mV at 3150mhz and u can go 270W or 330W and its still stuck. So its not powerlimited. Something is sreaming broken... Why doesnt it always go 1.175V and 3550mhz ? (talking about helldivers) its driving me nuts... Its like the driver is forcing the card to undervolt itself.
When the clock and voltage is stuck at for example 1050mV at 3150mhz i can go -100mV w/o crash and the voltage does not budge... but if i go -105/-110mV the clocks and the voltage decrease, a little with crash in next few sec if i dont increase the voltage again...

Josh128 · Mar 18, 2025

Kepler_L2 said:
There is no relationship at all between Navi4C and Navi31/32.

How so?

eek2121 · Mar 18, 2025

adroc_thurston said:
They can build it any way they want.
It's just that they're not gonna.

adroc_thurston said:
Doesn't exist.

Also doesn't exist.
Not even in JESD250D.

PS5 is like 14Gbps tho.

Never say never. Current roadmaps are not set in stone. There is absolutely a higher clocked part with more memory at AMD HQ (not as high of clocks as some of these rumors have suggested, but there is a bit of a bump). To my knowledge it is just a toy at the moment, and nothing has made it to partners...unless things have changed. My own speculation was that it meant as a response to the 5xxx super series. I hope they do launch it.

igor_kavinski · Mar 18, 2025

gaav87 said:
I can fiddle with the offset and mV and somehow trick "force" the card to run at 3550mhz stable for few min...

What are the temps like during that period of high frequency?

A modded BIOS could unlock that frequency permanently but it would probably degrade the silicon quicker since if that frequency had been safe, why would AMD limit it? UNLESS they have plans to release a halo part with that frequency...

tsamolotoff · Mar 19, 2025

igor_kavinski said:
why would AMD limit it

Because it's amd , they've locked PPT for no reason on n31 (and it can run daily with like 800W power budget with no damage to memory or compute chiplets). With nVidia there are at least some gpus with unlimited PL vbios, but for AMD you have to use evc2se or other hw mods.

Racan · Mar 19, 2025

https://videocardz.com/newz/amd-unreleased-radeon-rx-9070-oem-reference-design-pictured

It's a shame we didn't get this AMD design imo.

SolidQ · Mar 19, 2025

Racan said:
It's a shame we didn't get this AMD design imo.

Maybe 2 coolers not enough was?

200k sold?

https://videocardz.com/newz/amd-reportedly-shipped-200000-radeon-rx-9070-graphics-cards-already

Racan · Mar 19, 2025

SolidQ said:
Maybe 2 coolers not enough was?

200k sold?

https://videocardz.com/newz/amd-reportedly-shipped-200000-radeon-rx-9070-graphics-cards-already

That, I think, would have been enough for the RX 9070.

Josh128 · Mar 19, 2025

AMD sold 200K RDNA 4 GPUs, apparently.

MSN

Hail The Brain Slug · Mar 19, 2025

gaav87 said:
- Same thing in helldivers 2...
I can fiddle with the offset and mV and somehow trick "force" the card to run at 3550mhz stable for few min...
Sometimes the card is stuck at 3150mhz and 320W even with +10% power limit... should be up to 410W (highest i saw in furmark)
hotspot at 75C (forced fans at 80%)... .
- I force reset drivers and same spot on the ship in helldivers bam 374W 3550mhz and crash after few min xD
FPS with stuck boost clocks 144 at 3150mhz and 157-158 with ~3550mhz "unstuck"

- steel-nomad 3dmark seem to be broken... You can have 2950mhz at 290W or 378W+ with exact same perf. Its like the clock is stuck at 2950mhz and 890mV !!!
Like what the heck is this. Sometimes im able to "unstuck" it and gain like 5% perf its boosting to 3150mhz netting me low top100 leaderboard...

-Same thing i did in furmark when i somehow unstuck the clock and did 5% faster than rtx 4090D performance
furmark: Stock 343fps, normal overclock 373fps, unstuck oc 400fps
This is madness hope they fix this...

- I have also noticed that the voltage is also stuck... Card is doing 1050mV at 3150mhz and u can go 270W or 330W and its still stuck. So its not powerlimited. Something is sreaming broken... Why doesnt it always go 1.175V and 3550mhz ? (talking about helldivers) its driving me nuts... Its like the driver is forcing the card to undervolt itself.
When the clock and voltage is stuck at for example 1050mV at 3150mhz i can go -100mV w/o crash and the voltage does not budge... but if i go -105/-110mV the clocks and the voltage decrease, a little with crash in next few sec if i dont increase the voltage again...

The Taichi review is up on Techpowerup with the teardown and PCB shots. Looks like it's one big power plane like the Nitro+.

Josh128 · Mar 19, 2025

Is 60fps V-Sync going to make a comeback or what?! I mean, sheesh, 80W vs 360W? Frame-rate caps are your AC units friend, apparently. Even if you choose something super-taboo, like 85Hz + VRR, you are still going to be saving an arse-load of power and heat generation. Its actually very impressive in this regard compared to almost every other card produced. Uses like half the power that my 6700XT does when both are throttled to 60Hz.

AsRock 9070XT Taichi

tsamolotoff · Mar 19, 2025

Josh128 said:
80W vs 360W?

60 fps in UE5 titles is like full power even in FullHD

GaiaHunter · Mar 19, 2025

I don't think the frequency offset does anything.

All it matters is the undervolt and the power target.

Z O X · Mar 19, 2025

Josh128 said:
Is 60fps V-Sync going to make a comeback or what?! I mean, sheesh, 80W vs 360W? Frame-rate caps are your AC units friend, apparently. Even if you choose something super-taboo, like 85Hz + VRR, you are still going to be saving an arse-load of power and heat generation. Its actually very impressive in this regard compared to almost every other card produced. Uses like half the power that my 6700XT does when both are throttled to 60Hz.

Or one could say:
if a GPU running e.g. Apex at 180 FPS consumes 260W, then any 9070 variant should use 180W for the same job ...

gaav87 · Mar 19, 2025

igor_kavinski said:
What are the temps like during that period of high frequency?

A modded BIOS could unlock that frequency permanently but it would probably degrade the silicon quicker since if that frequency had been safe, why would AMD limit it? UNLESS they have plans to release a halo part with that frequency...

Normal TEMPS
Check this out 3750mhz... the VF curve is broken

-55mV 3350mhz
-40mV 3300mhz
-35mV 3575mhz
-30mV 3550mhz
-0V 3450mhz
Exact same power draw and temps in all
I can boost up to 3725-3750mhz depending on idk what tbh... sometimes i boost lower then i do pc reset and i boost higher...
Sometimes this clock behaviour also manifests itself in GAMES ! By random...
Like i set -35mV +275mhz and play game for 1-2h and out of nowhere like randy orton my card says nah RKO i hear excessive coil-whine i turn on monitoring and BAM 3550mhz and crash is imminent if i dont fiddle with the settings xD

gaav87 · Mar 19, 2025

Check this out this is not normal !!
stock 3250mhz
3350mhz +275offset -55mV 324W 79C (reported 1121mV)
3400mhz +275offset -130mV 306W 79C (reported 1086mV)
3300mhz +275offset -40mV 326W 82C (reported 1121mV)
3575mhz +275offset -35mV 336W 82C (reported 1118mV)
3475mhz +275offset +-0mV 336W 84C (reported 1120mV)
3625mhz +325offset -30mV 336W 85C+10% power
3612mhz Same with +0% power

igor_kavinski · Mar 19, 2025

@gaav87 , file a bug report with AMD and present the evidence.

gaav87 · Mar 19, 2025

igor_kavinski said:
@gaav87 , file a bug report with AMD and present the evidence.

Ye prolly will do on weekend gather more evidence. Cause ppl dont believe me...
VF curve is broken... In some benchmarks/games the card undervolts itself.

Its like this turn on any game static scene dont move. Check what voltage the card reports. lets say 1125mV
So i can do -75mV "undervolt" but the reported voltage DOES not change no matter if u do -15 -20 -25 -30 -35 etc... BUT the CLOCK increase.
So if i set -75mV i gain for example +200mhz right ? Then i start playing voltage fluctuates card wants 1135mV ? BAM CRASH

If the game static sceene wants 1125mV and i do -80mV then the voltage reporting and power draw DROPS !
People think they are doing -200/-130mV "undervolts" but in reality the card undervolts itself and the setting does nothing but somehow the clock increases. with exact same power and temps.
Idk if i explain it correctly and u get where im comming from.

OH i forgot some games or benchmark it works.
Like steel nomad 3dmark for example.
stock settings im doing 0.96V or smt
But with -55mV im doing 0.89-0.9V
Thats why there are almost no people that can go lower than -75mV in steelnomad benchmark asked few hc overclockers (i can do -65mV)

Other games/benchmarks its like i explained above !

adroc_thurston · Mar 19, 2025

gaav87 said:
VF curve is broken... In some benchmarks/games the card undervolts itself.

Oh it's not you're just touching things truly magical and you shouldn't.
AMD things exist to be kept stock.

gaav87 · Mar 19, 2025

igor_kavinski said:
@gaav87 , file a bug report with AMD and present the evidence.

Yep im not mad
Just tested in 3dmark timespy extreme.
stock +-0mV +-0offset 0% power limit
clocks at 2950-3000mhz voltage between 0.98-1.02V

I went:
-65mV 3050mhz same voltage fluctuation as in stock 0.98-1.02V
-130mV 3100-3150mhz SAME VOLTAGE 0.98-1.02V
-185mV 3300mhz SAME VOLTAGE !!! I CRASHED reported voltage 0.96-1.0V
-180mV 3250-3300mhz SAME VOLTAGE NO CRASH
all at same temps and power kinda crazy
So 1.2V -185mV = 1.015V thats why i crashed at -185mV and not at -180mV

AND i gained stock +0mv 94.6fps -180mV 100fps just by doing this........
Why does AMD always fails at something xD They had perfectly fine maximum/minimum clock setting and working undervolt setting....

Shmee · Mar 19, 2025

Is this voltage curve just for RDNA4, or will it work on RDNA3 as well?

gaav87 · Mar 19, 2025

Shmee said:
Is this voltage curve just for RDNA4, or will it work on RDNA3 as well?

Rdna4 only top is rdna2/3 bottom rdna4

Just tested forza horizon5 same exact behaviour:

-0mV 3150mhz driver reporting 1046-1066mV 107fps
anything between -0mv and -85mv = 3150mhz
-130mV 3325-3350mhz voltage reporting 1046-1066mV 115fps
-135mV 3350mhz voltage reporting 1046-1066mV 117fps <- sometimes crash i can run the benchmark 5x times
-140mV 3375mhz crash almost instant
All at 0% power limit

1200mV-130mv=1070mV "safe" spot?
1200mv -135mv = 1065mV no voltage head room from what its reporting as 1046-1066mV sometimes crash
1200mv -140mv = insta crash

only

Discussion RDNA4 + CDNA3 Architectures Thread

Golden Member

Senior member

Senior member

Senior member

Senior member

Senior member

Diamond Member

Lifer

Senior member

Golden Member

Golden Member

Golden Member

Senior member

Diamond Member

Senior member

Senior member

Diamond Member

Junior Member

Senior member

Senior member

Lifer

Senior member

Diamond Member

Senior member

Memory & Storage, Graphics Cards Mod Elite Member

Senior member