Rise of the Tomb Raider patch #7 adds DX12 mGPU, Async Compute support

[Arachnotronic]

Sorry but you give them too much credit.

They release yet another broken DX12 patch. From benches I've seen, minimum FPS tanks hard with Async Compute enabled. Looks like these guys just can't do DX12 right. Leave it at that. The more they try, the worse it gets.

At identical graphics settings, running on an BDW-E @ 4.3GHz and 1080 SLI...

DX11:

DX12:

I see a gigantic speedup.

 

[Flapdrol1337]

what about without sli? where you're less likely to be cpu limited.

 

[Piroko]

You did prove him right though, your averages went up but in two out of three scenes the minimums are quite a bit lower.

 

[realibrad]

what about without sli? where you're less likely to be cpu limited.

SLI would make you more cpu limited. The CPU needs to feed 2 gpus instead of one.

 

[Arachnotronic]

You did prove him right though, your averages went up but in two out of three scenes the minimums are quite a bit lower.

Main point is that Pascal supports Async Compute and it can actually deliver a meaningful speedup with it. This is contrary to what some people claim.

 

[Stuka87]

SLI would make you more cpu limited. The CPU needs to feed 2 gpus instead of one.

This should not be an issue in DX12, its whole purpose is lower CPU overhead.

 

[Stuka87]

Main point is that Pascal supports Async Compute and it can actually deliver a meaningful speedup with it. This is contrary to what some people claim.

Also contrary to what benchmarks say. Unless you know of some that state otherwise?

 

[Arachnotronic]

Also contrary to what benchmarks say. Unless you know of some that state otherwise?

I just provided benchmarks of a game in DX11 mode and a game in DX12 mode w/ Async compute, and the 1080 SLI setup sees a big speedup in DX12 mode.

 

[Piroko]

Main point is that Pascal supports Async Compute and it can actually deliver a meaningful speedup with it. This is contrary to what some people claim.

Point taken, but also replying that your point was not actually addressing the point made by Silverforce. Thus this became pointless in a way.

 

[VulgarDisplay]

SLI would make you more cpu limited. The CPU needs to feed 2 gpus instead of one.

I think what he was trying to say is that with sli he's CPU limited so run the benchmark with a single gpu to see what effect the patch has in a gpu limited scenario.

That would actually be more likely to show how well asynchronous compute is working with pascal rather than how much better dx12 is in terms of CPU overhead.

For example in bf4 with mantle which does not have asynchronous compute I saw my minimum fps numbers increase by 60 or so fps just because of the reduced CPU bottleneck on a single gpu.

Edit: actually the more I think about it the gtx1080 may just be too fast to even be able to tell the difference between a CPU based performance increase and an increase brought on by asynch.

Probably need to test it with a gtx1060 to see what's what.

 

[linkgoron]

At identical graphics settings, running on an BDW-E @ 4.3GHz and 1080 SLI...
I see a gigantic speedup.

You're not contradicting him. You're comparing DX11 and DX12.

Even so, in DX11 Syria you're getting 30% higher min FPS and in Geothermal valley you're getting 10% higher minimums in DX11.

Silver was talking about Async Compute. Can you run DX12 with Async compute and DX12 without to see the difference?

 

[Stuka87]

I just provided benchmarks of a game in DX11 mode and a game in DX12 mode w/ Async compute, and the 1080 SLI setup sees a big speedup in DX12 mode.

That in no way shows that async compute is increasing performance. We need to see a DX12 with it off, and a DX12 with on to show that it helps.

 

[realibrad]

This should not be an issue in DX12, its whole purpose is lower CPU overhead.

DX12 does not get rid of all cpu bottlenecks. It reduces DX12 overhead, but two 1080s will not be a bottlenecks in most games. That is to say, that if you were to increase the speed of the cpu, you will see more FPS than if you increased the speed of the GPUs.

 

[Sweepr]

I just provided benchmarks of a game in DX11 mode and a game in DX12 mode w/ Async compute, and the 1080 SLI setup sees a big speedup in DX12 mode.

Impressive results. Nice to see improved performance for Pascal @ DX12 in a game that's not sponsored by AMD. 🙂

 

[LTC8K6]

Do people think async compute is going to somehow save AMD or something?

 

[Kenmitch]

Do people think async compute is going to somehow save AMD or something?

I don't really think so. It's more like one of those free performance things in the end....Of course it will take proper implementation by the developers.

 

[Bacon1]

I just provided benchmarks of a game in DX11 mode and a game in DX12 mode w/ Async compute, and the 1080 SLI setup sees a big speedup in DX12 mode.

Try testing without HBAO+, just "On" for AO.

Also turn AsyncCompute off in the registry and test with it on and off to see if it is actually doing anything, or if the lower CPU overhead is why you are getting better FPS.

 

[AtenRa]

Do people think async compute is going to somehow save AMD or something?

AC alone no, its a combination of things that required in order to have a positive effect.

 

[tviceman]

Try testing without HBAO+, just "On" for AO.

Also turn AsyncCompute off in the registry and test with it on and off to see if it is actually doing anything, or if the lower CPU overhead is why you are getting better FPS.

It would be easier and more to the point just to say you don't believe the 1080's are benefiting from async.

 

[hsjj3]

Dear sirs, are there benchmarks for this latest update? I want to see how things fare on the GTX 900 series cards.

 

[Bacon1]

It would be easier and more to the point just to say you don't believe the 1080's are benefiting from async.

Well because we haven't seen them in any other games either.

In Ashes of the Singularity in sites that actually test enabled vs disabled, not just DX11 vs DX12, there is a very little gain / loss.

https://www.computerbase.de/2016-05...ramm-ashes-of-the-singularity-async-compute_2

Considering people testing on Hawaii have found there to be 1-2% perf difference from using async compute (toggling on/off) in ROTTR, I don't see how a 1080 would gain 10%+ from using it. Sounds like its just DX12 vs DX11 w/o async compute coming into play.

 

[redzo]

Dear sirs, are there benchmarks for this latest update? I want to see how things fare on the GTX 900 series cards.

Good point. We need reviews generally speaking. It's sad that the popular media didn't pick this one up.

 

[DiogoDX]

http://www.overclock3d.net/reviews/...e_tomb_raider_directx_12_performance_update/4

 

[Bacon1]

http://www.overclock3d.net/reviews/...e_tomb_raider_directx_12_performance_update/4

Nice thanks

The improvements @ 1440p are huge:

 

[crisium]

I'll give it a shot tonight. I don't bother with canned benchmarks (never once ran it for this game), but I simply go to the same area in the Soviet Installation with the lowest framerate. So far no patch has seen any improvement, and DX12 was worse than DX11.