NVidia does not support asynchronous compute on DX12

[Cry Havoc]

http://hardforum.com/showthread.php?t=1873640

https://www.reddit.com/r/pcgaming/co...o_not_support/

Getting really interesting right now for NVidia.

Current indications are performance gains of up to 70% for AMD GPUs in Direct X 12 (you read that right).

NV GPUs don't see nearly this gain. Meaning in some cases under DX12, a 290x is actually outperforming the 980ti.





Obviously nothing set in stone yet but definitely worth watching.

[baseline bum]

http://hardforum.com/showthread.php?t=1873640

https://www.reddit.com/r/pcgaming/co...o_not_support/

Getting really interesting right now for NVidia.

Current indications are performance gains of up to 70% for AMD GPUs in Direct X 12 (you read that right).

NV GPUs don't see nearly this gain. Meaning in some cases under DX12, a 290x is actually outperforming the 980ti.





Obviously nothing set in stone yet but definitely worth watching.

Where are you seeing the 290x outperforming the 980 Ti? In the Ashes of Singularity benchmark done by PC Perspective they show a 390x outperforming a 980 in DirectX12 mode.



But there is a pretty large difference in power between a 980 and 980 Ti and also a non-negligible difference between a 290x and 390x also. It will be interesting to see if this trend holds in other DX12 games. I'd find it kind of funny after Nvidia bought out Aegia to make PhysX a proprietary NV feature if AMD comes and rapes their physics performance with asynchronous computing.

AMD seems to be going for broke on DX12 right now while ignoring DX11, I wonder if this is why.

[DJR210]

typical AMD propaganda

will always be behind

AMD is such a great GPU maker on paper

"Mantle will revolutionize PC gaming"

"TressFX will revolutionize the way developers implement physics into games"

"HBM will revolutionize VRAM in GPU's"

AMD..bless their hearts

[baseline bum]

typical AMD propaganda

will always be behind

AMD is such a great GPU maker on paper

"Mantle will revolutionize PC gaming"

"TressFX will revolutionize the way developers implement physics into games"

"HBM will revolutionize VRAM in GPU's"

AMD..bless their hearts

Yeah, I'm a little skeptical until I see results with more DX12 les.

[DJR210]

Havoc

[baseline bum]

What games are supposed to be DX12 this Christmas season anyways? The two I'm looking forward to are Fallout 4 and the new Tomb Raider.

[baseline bum]

Ah ok, looked it up and the list looks like

Arma III - desperately needed considering how CPU limited that game is

Ark Survival Evolved - This game looks like a steaming turd, that

Star Citizen - Supposed to release in Fall 2016, but I doubt it. It'll probably just be a little above average graphically by the time the ing thing actually releases.

So I guess I'm good with DirectX11 still for another year even if Maxwell does end up sucking at DX12.

[Cry Havoc]

What games are supposed to be DX12 this Christmas season anyways? The two I'm looking forward to are Fallout 4 and the new Tomb Raider.

The majority probably won't be until next year. Still, if the numbers hold, 290x, 390, and 390x look like fantastic buys right now for the value they'll give. Even if the numbers aren't as gaudy as it seems now (and probably won't be), there seems to be a lot of indications that DX12 is going to be a pretty big win for AMD in the performance department.

[ElNono]

Not really sure it's a hardware issue, tbh... AFAIK, both NVidia and AMD upload signed binary blobs with microcode to their cards.

I think this is the case of DX12 just happening to be closer architecturally to Mantle (and Metal, Vulkan, etc), and so AMD is currently a better fit.

On the other hand, NVidia DX11 code is top notch and pretty much unrivaled at this point.

NVidia might have to trade off some DX11 performance in order to boost DX12 and they might just be unwilling to do that at this point.

It will certainly be interesting to follow how this moves along.

[illusioNtEk]

lol AMD using Intel CPUs to showcase there new GPUs

[baseline bum]

lol AMD using Intel CPUs to showcase there new GPUs

Hard to believe only ten years ago AMD's CPUs all over Intel's garbage processors.

[Cry Havoc]

Hard to believe only ten years ago AMD's CPUs all over Intel's garbage processors.

[Cry Havoc]

http://www.overclock.net/t/1572716/d...crowd-sourcing

This is a great post and gives a TON of info about why this could be, and likely *will be* a big deal.

[baseline bum]

http://www.overclock.net/t/1572716/d...crowd-sourcing

This is a great post and gives a TON of info about why this could be, and likely *will be* a big deal.

I don't know how much Fable Legends uses async compute, but strike another win for AMD with the $330 R9 390 outperforming the $500 GTX 980 in the Fable Legends DX12 benchmark.

[Cry Havoc]

I don't know how much Fable Legends uses async compute, but strike another win for AMD with the $330 R9 390 outperforming the $500 GTX 980 in the Fable Legends DX12 benchmark.

Also note that it's a 1080p benchmark. AMD's new cards really shine at 1440p and 4k.

Edit: I'm stupid and didn't see the 2nd graph. Doh.

[baseline bum]

Also note that it's a 1080p benchmark. AMD's new cards really shine at 1440p and 4k.

Edit: I'm stupid and didn't see the 2nd graph. Doh.

I wonder how much of this is hardware on Nvidia cards and how much is them not giving a about DX12 right now and not writing good drivers for it while AMD is pushing DX12 hard. From what I have read this benchmark doesn't seem to make much use of async compute and AMD is still winning big here.

[m>s]

ing nerd get a real gf

[Cry Havoc]

I wonder how much of this is hardware on Nvidia cards and how much is them not giving a about DX12 right now and not writing good drivers for it while AMD is pushing DX12 hard. From what I have read this benchmark doesn't seem to make much use of async compute and AMD is still winning big here.

They've actually looked at the way the hardware is designed. Even with optimal drivers, Nvidia cards *cannot* make good use of asynch. It's not possible with any of their cards atm. AMD is going to have a huge lead from basically now until the next GPU cycle in performance.

[ElNono]

They've actually looked at the way the hardware is designed. Even with optimal drivers, Nvidia cards *cannot* make good use of asynch. It's not possible with any of their cards atm. AMD is going to have a huge lead from basically now until the next GPU cycle in performance.

They actually looked at how NVidia said the hardware is currently designed to operate. Now it could perfectly be that indeed there's a hardware barrier there, but I think what's being overlooked is that NVidia actually uploads a digitally signed firmware blob to the cards (and we know this because of the Linux drivers), which actually is the OS for the card. So it's yet to be seen whether the limitation can be overcome that way. Time will certainly tell.