DirectX 12 Asynchronous Compute. Worth building for?

trickyd

Chieftain
Joined
Sep 17, 2016
Messages
7
Location
Tennessee
Background: I have a Mac with an i7 and NVIDIA GeForce GT 750M. My plan was Bootcamp until the Mac version is released. I figured it would be good enough.

But then I came across the following link regarding Asynchronous Compute

http://m.marketwired.com/press-rele...eiers-civilizationr-vi-nasdaq-amd-2142057.htm

This powerful feature allows for parallel execution of compute and graphics tasks, substantially reducing the time other architectures need to execute the same workloads in a longer step-by-step manner. Asynchronous compute on many Radeon™ GPUs will perfectly complement the unit-rich late game of Civilization VI.

I have not come across anything that says definitively how big of a deal this could be.

Does anyone know?

Context: I am a mid thirties gamer with a wife, a young kid and a bunch of commitments. I'd rather not spend $800-$1000 on a new build if I don't have to, but I'll probably put 250-500 hours into this game over the next 5 years.

If I can save a ton of time in the late game because of this feature it might just be compelling enough for me to go ahead with a new build.

Again, does anyone know how big of a difference async compute might make at this point?
 
i would guess nvidia has something extremely similiar in their gpus that allows them to take advantage of the directx12 feature for asynch compute. i think having directx12 is the only important thing here, then whether you have radeon or nvidia gpu shouldnt make much difference. i just upgraded from windows 7 to 10 just to get the directx12, i think the performance boost will be significant
 
Thanks for the reply, Deadstarre. I am with you that it sounds like DirectX 12 is the most important piece, but when I found this Anandtech article...

http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review/9

I came away with the impression that only the most recent Pascal architecture from NVIDIA can really do anything with async compute at the hardware level. That would leave my 750M unable to benefit.

But I do not follow this stuff closely so I could be pulling inaccurate conclusions. I would be happy to be wrong!
 
This powerful feature allows for parallel execution of compute and graphics tasks, substantially reducing the time other architectures need to execute the same workloads in a longer step-by-step manner. Asynchronous compute on many Radeon™ GPUs will perfectly complement the unit-rich late game of Civilization VI.

I have not come across anything that says definitively how big of a deal this could be.
In early testing it looks like a 16 to 20% difference.
DirectX 12 appears to help AMD by both reducing driver overhead and allowing developers to leverage GCN’s formidable asynchronous compute capabilities. It’s less clear why Nvidia continues to struggle with delivering absolute performance improvements in DirectX 12, even in titles that otherwise favor the company’s products.
It’s still too early in the DirectX 12 / Windows 10 product cycle to draw absolute conclusions about which architecture will prove definitively better and the imminent arrival of new GPUs from both companies will render the question at least somewhat moot. So far, it looks as though AMD gamers are generally better off using DirectX 12 when it’s available, while Nvidia owners may want to stick with DX11, even when gaming in Windows 10.
The quote comes from this article from ExtremeTech.

Also another article from ArsTechnica.

edit - the articles do say that the new nVidia cards may address the issue.
 
My advice is to wait to see how the game plays on your current machine. The fact that you don't see much coverage about the feature tells you that it's not going to have a night and day difference. If a turn takes 30 seconds to load, then maybe this would do it in 28-29. % wise it might be "substantial" but in real world terms it's probably nothing.

I have an oldish laptop with i5 and NVidia 335M. I'm planning to play Civ6 on that :)
 
Assume from reading the article and supporting technical write ups this is basically an advanced scheduler with memory threads using the DX libraries. Asynchronous is this context means running parallel services in memory (the pretty packages in the ARS article) not to confused with the same concept in data transport.

It also appears that the internal GPU load balancing (tricky) gives them the same kind of capabilities introduced with CPUs a while back.

If it works as advertised it could make a tremendous difference; very exciting to see this type of advances being considered.

Would guess if Radeon is doing this Nvidia will as well, rarely does one side out a advance the other for long.
 
My advice is to wait to see how the game plays on your current machine. The fact that you don't see much coverage about the feature tells you that it's not going to have a night and day difference.
And even if there would be a lot of coverage it wouldn't mean a thing.

Civ 5 was a propaganda vehicle for DirectX 11, promising this-and-that speed and graphics, where in practice it didn't make that much of a difference compared with DX9.
The biggest difference in the end being an extra menu to start up the game :p
 
My advice is to wait to see how the game plays on your current machine. The fact that you don't see much coverage about the feature tells you that it's not going to have a night and day difference.
This. Especially if you want to mind you budget (and who doesn't) then first see what it's like on your current gear. My expectation is that Civ6 will give roughly the same performance as Civ5.
 
Thanks for the links, leif erikson. Interesting articles.

That pretty much confirms that my 750M is not going to benefit from DirectX12 async. Maybe there is some benefit to be had from the multi adapter support though. My mac does also have an integrated gpu.

Good advice in here though. Async sounds like a very compelling design, but I think I will hold out to see some real world benchmarks from Civ.

As smartcunuck was saying it makes a big difference if it is 20% faster on a 30 second end of turn or something like 30% faster on a 90 second end of turn. I'd pay a premium for the latter for sure. Not positive about he former.
 
I doubt your GPU, dx11 or dx12 is going to be the issue. It is nice to have smooth frames, but the real workhorse is going to be your CPU cutting those turntimes down.
 
Is there a reason we've skipped over the obvious answer of selling the child and wife on the black market and just buying a top of the line rig?

Correct me if I'm wrong but not only would you have fewer distractions in this scenario, you'd also get the most value out of the time you have.
 
Back
Top Bottom