Discussion in 'Videocards - NVIDIA GeForce Drivers Section' started by Dipankar sanyal, Sep 27, 2015.
All this fighting about intel vs amd in the Nvidia driver section go figure.
IIRC they said the feature was being exhibited as being available/enabled when it wasn't and said it'll come in a future driver. But I could be mixing things up.
The way it's currently implemented in the driver it can actually cost performance instead of gain some.
Which is not the case with AMD GCN GPUs, developers already have experience from the PS4 and AMD gains performance with it (a Frostbite developer said e.g. ambient occlusion for free under certain circumstances etc.).
The feature has to be enabled for a driver to be DX12 compatible. All DX12 drivers must have async compute feature exposed. There is no way of having it missing and being a DX12 driver. The feature itself however does not preclude any way of handling it on the hardware. Meaning that there is no "feature was being exhibited as being available/enabled when it wasn't" - this is a lie on Oxide's part. If this feature is exhibited then it is exhibited. If you loose performance when you use the feature - this is the way it is handled by the h/w in question. There is no "proper way" of handling async compute.
What AoS has done is disabled the _usage_ of async compute feature in the game itself on NV GPUs. Why? Probably because NV felt that this would be easier to handle while they are working on enabling parallel async compute on Maxwell 2. Or maybe because AoS is using the feature in such a way that NV GPUs will always be better off without it - don't forget that we're talking about an AMD sponsored game made for Mantle in the first place.
Running compute in parallel to graphics can easily cost performance on any GPU, including AMD GPUs. Consoles and whatever whoever achieved on them with async compute have little to do with PCs because on a console you can hand tune to the static h/w and s/w configurations - this is completely impossible on PC. The question of how much performance will async compute even bring on PC even for AMD h/w is still open.
All the dx12 threads are pretty much the same...
As I already said, we'll probably see in ~2 weeks.
Question is not really if Nvidia profits the same like AMD, but how much more AMD profits by AC.
Like it or not, doesn't change a thing.
you should actually take a look on that
Don’t rely on the driver to parallelize any Direct3D12 works in driver threads
On DX11 the driver does farm off asynchronous tasks to driver worker threads where possible – this doesn’t happen anymore under DX12
While the total cost of work submission in DX12 has been reduced, the amount of work measured on the application’s thread may be larger due to the loss of driver threading. The more efficiently one can use parallel hardware cores of the CPU to submit work in parallel, the more benefit in terms of draw call submission performance can be expected.
this pretty much seals the case about the driver..
We probably won't see anything in ~2 weeks and it will probably take us at least half a year before any statistically measurable data will be collected. Like it or not but there are like five or so DX12 titles coming between now and May next year. All the rest of them are DX11.
This has nothing to do with async compute as it describes job submission to the GPU by the CPU. It is perfectly possible that a future Maxwell 2 driver will enable parallel async compute.
i call it bs
my rig with oced sb at 4.4ghz and oced 7870 tahitile at 1.020g runs fine to this moment. PSU have 3-4y. Played Warthunder, WOT, BF3, BFBC2, Witcher2, 3, and so one, so no miner in use. Working at last 70-80h weekly.
550 watt psu is more than enough for an intel cpu and a 970 with a 145 watt tdp.
Is sli possible for Async Compute?
Could sli performance make it "do-able" while single card owners may not benefit due to the deficit?
Either way, 30fps for dx12 is low...bring on next gen.
Asking them to fix async compute is like asking them to fix the 512mb on the gtx970.
Sorry guys but i never go back to AMD even if they perform few fps better.
All because drivers witch go flanky. They cant make stable drivers after 2 years of card release.
Which is probably why they are asking developers not to use/abuse certain features/methods:
DX12 Do's and don'ts
Common...insisting with the AMD drivers is bs. Nvidia showed they can mess up drivers too. I have both nvidia (in my sig) and amd (HD7850 in another computer) currently and didn't have any driver issues at all.
I don't have any noticeable AMD driver issues. The excuse not to go with AMD because of drivers is getting tiresome.
Yeah, it's pretty mind boggling why people still claim that.
In actuality, AMD drivers have gotten pretty good, while Nvidia's have gotten worse over the last year or so. The only real problems I've been having lately on AMD systems was with Crossfire X compatibility/profiles, but the same can be said with SLI on the Nvidia side (just that it's easier to find/create a working SLI profile your own with Nividia Inspector than it is for AMD Crossfire X using AMD Control Panel/RadeonPro although there are more and more games where neither SLI nor CrossfireX work properly). Right now I'd say both companies' drivers are on roughly the same level, although AMD's seem to be a bit more optimized for Windows 10 to me.
You guys fail to understand that Async compute is a "new" feature exclusive for their pascal gpus. If you want Async compute you will have to purchase a new pascal gpu!
Bloody hell. Async compute is a feature of any DX12 driver. In no way does this feature imply any particular way of execution. Depending on the underlying h/w architecture it may actually be better to execute stuff launched asynchronously in a serial fashion.
And you don't know **** about Pascal.