Ashes of the Singularity Coming, with DX12 Benchmark in thread.

thebennyboy · 24 Feb 2016 at 23:24

Did Nvidia ever "fix" the issues with Async compute? I don't think there is a lot they can do as it's a hardware limitation.

CAT-THE-FIFTH · 24 Feb 2016 at 23:34

Anandtech test:

http://www.anandtech.com/show/10067/ashes-of-the-singularity-revisited-beta/6

Update 02/24: NVIDIA sent a note over this afternoon letting us know that asynchornous shading is not enabled in their current drivers, hence the performance we are seeing here. Unfortunately they are not providing an ETA for when this feature will be enabled.

I thought the last time the benchmark was run weren't they saying a similar thing??

D.P. · 24 Feb 2016 at 23:58

Dygaza said:
AMD just released 16.2 hotfix drivers that has performance upgrades for this benchmark version.

It is an AMD PR stunt form their marketing division so i expect nothing less.

Dygaza · 25 Feb 2016 at 00:17

D.P. said:
It is an AMD PR stunt form their marketing division so i expect nothing less.

That driver did had other fixes aswell, so it wasn't dedicated for this game alone.

CAT-THE-FIFTH · 25 Feb 2016 at 00:21

D.P. · 25 Feb 2016 at 00:39

Dygaza said:
That driver did had other fixes aswell, so it wasn't dedicated for this game alone.

But they specifically gave it to Anandtech, no doubt asking Anandtech to do such a review. Thats fine, you just have to be careful what you read on the internet and take such benchmarks with a large mountain of salt. Its just a pitty AMD are not so willing to make driver hotfixes for AAA game releases.

Anyway, it would be nice to get a clearer picture of the Nvidia driver status. An AOTS developer claimed the newest Nvidia drivers had async enabled, but nvidia are claiming not for AOTS yet. Until that is resolved such benchmarks are not that useful at this stage sadly.

CAT-THE-FIFTH · 25 Feb 2016 at 00:48

The pcgameshardware test does not use the 16.2 driver for AOS,but the earlier 16.1 one which is not optimised for the test:
http://www.pcgameshardware.de/Ashes...8/Specials/Benchmark-Test-DirectX-12-1187073/
http://support.amd.com/en-us/kb-art...rimson-Edition-16.1-Hotfix-Release-Notes.aspx

Nvidia was supposed to have Async working the last time the benchmark was in the news which was in October 2015 and still nothing.

Waiting for the "async" driver for my GTX960 is like Waiting for Godot at this rate. Personally I don't think we will see anything for another few months and by then Pascal will be out and it will probably be some pipe dream.

Personally I am ****ed off Nvidia,as they CBA to do anything about it.

D.P. · 25 Feb 2016 at 01:03

I doubt it is because they CBA dealing with it' it just not a priority since there are basically zero fully released Dx12 games. It's like the DX12 Fermi drivers, they said they would be released when DX12 games are available.

I also suspect that the driver solution might have to be tuned to specific games more closely. We know form fable of legends that maxwell async performance can be very good if done in consideration for the architecture.

CAT-THE-FIFTH · 25 Feb 2016 at 01:07

Its not the point,the major strategy game me and my mates play at LANs is Sins of a Solar Empire followed by the various versions of Supcom.

This is probably going to be the next major LAN game we get - ATM,we have nothing but Nvidia empty promises for like 4 months.

Give it 4 more months and out pops Pascal,and why would they even care at that point?? If I don't see them release an "Async" driver in the next month or two ,I think I will just give up all hope of it ever happening. Six months is more than enough time for a company like Nvidia to be able to make an "Async" capable driver.

I am not happy with the driver support in the last 4 to 5 months or so. FFS,ARK ran WORSE on me and my mates GTX960 cards than a rebranded HD7950 with less VRAM which is a worse card,and its bloody Nvidia sponsored. Plus where is the DX12 patch for that too?? I could have done with the performance boost. I ended up refunding the game since it was pointless.

Dave2150 · 25 Feb 2016 at 01:13

D.P. said:
But they specifically gave it to Anandtech, no doubt asking Anandtech to do such a review. Thats fine, you just have to be careful what you read on the internet and take such benchmarks with a large mountain of salt. Its just a pitty AMD are not so willing to make driver hotfixes for AAA game releases.

Anyway, it would be nice to get a clearer picture of the Nvidia driver status. An AOTS developer claimed the newest Nvidia drivers had async enabled, but nvidia are claiming not for AOTS yet. Until that is resolved such benchmarks are not that useful at this stage sadly.

D.P, time to stop trolling.

AMD have been quite good driver wise lately, releasing a hotfix patch for the recent Fallout 4 patch (that broke the crossfire profile) extremely quickly.

Just because AMD are winning this benchmark at the moment does give you the right to enter troll mode.

drunkenmaster · 25 Feb 2016 at 01:23

D.P. said:
I doubt it is because they CBA dealing with it' it just not a priority since there are basically zero fully released Dx12 games. It's like the DX12 Fermi drivers, they said they would be released when DX12 games are available.

I also suspect that the driver solution might have to be tuned to specific games more closely. We know form fable of legends that maxwell async performance can be very good if done in consideration for the architecture.

Nvidia said there would be Fermi DX12 drivers at Windows 10 launch, then they said later, then they said end of the year NOW they say when a DX12 game is ready... but what makes that more believable than the other multiple claims about an ETA for them? The question is still what will change between a DX12 game being released and the release of Win 10/DX12, the answer is nothing at all. Even for what they claim for DX12 support for Fermi is extremely minimal, the driver will be defaulting to dx11 modes for the majority of stuff it's asked to do within DX12 so there is very little to support there.

With very few DX12 features it's sure taking a hell of a long time to support it in drivers.

As for DX12 games being out, at least one would be out today if Nvidia hadn't worked with a company to remove DX12 features from a shipping game.

D.P. · 25 Feb 2016 at 01:57

Dave2150 said:
D.P, time to stop trolling.

AMD have been quite good driver wise lately, releasing a hotfix patch for the recent Fallout 4 patch (that broke the crossfire profile) extremely quickly.

Just because AMD are winning this benchmark at the moment does give you the right to enter troll mode.

I'm not trolling anyone :rolleyes:

I really don't care about this benchmark in the slightest, the truth about async compute is far too complex to be captured by a single benchmark by an A!D sponsored developer. Nvidia's architecture is very different to AMDs in term of multi-engine support, with Maxwell more design around a CUDA approach that does not translate efficiently to AMD's approach. You can make async compute capabilities run very well on either architecture but mostly likely with different approaches. GCN hardware better matches the multi-engine API in DX12, maxwell will be support an approach as used for the likes of physx.

If AMD have been releasing game ready drivers on release then I take that comment back.

Mtom · 25 Feb 2016 at 07:31

D.P. said:
But they specifically gave it to Anandtech, no doubt asking Anandtech to do such a review. Thats fine, you just have to be careful what you read on the internet and take such benchmarks with a large mountain of salt. Its just a pitty AMD are not so willing to make driver hotfixes for AAA game releases.

Anyway, it would be nice to get a clearer picture of the Nvidia driver status. An AOTS developer claimed the newest Nvidia drivers had async enabled, but nvidia are claiming not for AOTS yet. Until that is resolved such benchmarks are not that useful at this stage sadly.

Guru3d tested with the 16.1 Crimson, not the new driver
.

Dygaza · 25 Feb 2016 at 09:20

D.P. said:
But they specifically gave it to Anandtech, no doubt asking Anandtech to do such a review. Thats fine, you just have to be careful what you read on the internet and take such benchmarks with a large mountain of salt. Its just a pitty AMD are not so willing to make driver hotfixes for AAA game releases.

Anyway, it would be nice to get a clearer picture of the Nvidia driver status. An AOTS developer claimed the newest Nvidia drivers had async enabled, but nvidia are claiming not for AOTS yet. Until that is resolved such benchmarks are not that useful at this stage sadly.

Anandtech actually used 16.1.1 hotfix drivers.

About nvidia, Kollock said that latest nvidia shipping drivers had support for async. Does that actually mean public driver or special driver for devs?

fs123 · 25 Feb 2016 at 09:24

This benchmark has to favour AMD with thier Async Compute hardware. If it doesn't give noticeable gains then all the DX12 hype was for nothing.

flopper · 25 Feb 2016 at 09:32

Dygaza said:
Anandtech actually used 16.1.1 hotfix drivers.

About nvidia, Kollock said that latest nvidia shipping drivers had support for async. Does that actually mean public driver or special driver for devs?

LOL.
Nvidia just got owned totally by AMD a 390x goes agaisnt a 980ti.
You do know that AMD support stuff with hardware not software like Nvidia does with its recent 980ti?

Fury built for the future as well the new Polaris brighter than ever

Thunnylaf · 25 Feb 2016 at 09:38

Apparently, async isn't enabled in the latest drivers.

"Update 02/24: NVIDIA sent a note over this afternoon letting us know that asynchornous shading is not enabled in their current drivers, hence the performance we are seeing here. Unfortunately they are not providing an ETA for when this feature will be enabled."
http://www.anandtech.com/show/10067/ashes-of-the-singularity-revisited-beta/6

Mauller · 25 Feb 2016 at 10:18

Funny how people wet themselves now the furyx has shown a 20-30% lead in places. Yet the fury x has always been the better specced card when it comes to compute. With a reasonable amount of async, the fury should always be ahead of the 980ti. It was showing that 20-30% lead over a 1.3ghz 980ti, with the furyx at stock. Just goes to show how well it can work when its hardware is being used considering its peak theoretical is 8Tflops sp while the 980ti is around 5.5-6.

Another major point is how it has that lead while having better IQ, the nvidia cards appear to have a lot of lighting effects missing if you watch comparison videos.

Dygaza · 25 Feb 2016 at 10:32

Mauller said:
Another major point is how it has that lead while having better IQ, the nvidia cards appear to have a lot of lighting effects missing if you watch comparison videos.

With all honesty, that comparision video doesn't tell the whole truth, as if you look at them, you notice that video with nvidia card is being ran on older version of the game.

0.62.14002 for 390x
0.61.13854 for 980

eddyr · 25 Feb 2016 at 10:35

Well their was the whole thing about latest Nvidia drivers accepting async commands but implementing a workaround. That is both working and not working depending on whether you ask the question in the context of support or implementation. Marketing can respond as it suits the situation.

Mauller said:
Another major point is how it has that lead while having better IQ, the nvidia cards appear to have a lot of lighting effects missing if you watch comparison videos.

Yup makes you wonder about the changes IHV's were able to submit into their game branches, as is mentioned on Anandtech.

Definitely a lot of lighting differences, in both dx 11 and 12 from this video posted earlier.
https://www.youtube.com/watch?v=o2b6Nncu6zY