I asked the question below on a project forum, but since this venue is where a lot of the power users with heavy experience visit I will ask here as well.

On Windows 10 version 2004 Hardware accelerated GPU scheduling was added if your Video card supported it and you had at least (NVIDIA version 451.48, or AMD version 20.5.1 Beta) installed. Theory being you can off loading the CPU of the GPU scheduling that the GPU could do for it's self.

For the Gamer's it might add 2-4 frames ps if you had a newer card last year.

Has anyone played with it to see if it would help with computational work like what we do ?

Thanks
Bill F

I asked the question below on a project forum, but since this venue is where a lot of the power users with heavy experience visit I will ask here as well.

On Windows 10 version 2004 Hardware accelerated GPU scheduling was added if your Video card supported it and you had at least (NVIDIA version 451.48, or AMD version 20.5.1 Beta) installed. Theory being you can off loading the CPU of the GPU scheduling that the GPU could do for it's self.

For the Gamer's it might add 2-4 frames ps if you had a newer card last year.

Has anyone played with it to see if it would help with computational work like what we do ?

Thanks
Bill F

It's called CUDA, and it's been in existence for many years now.
You still can't offload everything.

CUDA only applies to Nvidia GPUs and AMD probably have something akin to it.

CUDA only applies to Nvidia GPUs and AMD probably have something akin to it.

Google search
CUDA is a parallel computing platform and application programming interface model created by Nvidia. It allows software developers and software engineers to use a CUDA-enabled graphics processing unit for general purpose processing – an approach termed GPGPU. Wikipedia

As was mentioned already CUDA and the CUDA toolkit have been around for years. What Microsoft has released is fairly new 2020 and it enables GPU cards to do their own scheduling rather than having the System CPU do the Scheduling. I would think that for BOINC users doing mixed work with Tasks from one project CPU based and other Tasks that are GPU based there would be a benefit to the CPU side tasks and possibly on the GPU side as the GPU card no longer has wait for the System CPU.

Bill F

---

In October 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.

What Microsoft has released is fairly new 2020 and it enables GPU cards to do their own scheduling rather than having the System CPU do the Scheduling.

Was that Microsoft release aimed at developers/programmers, or at end users?

Do you have a link?

What Microsoft has released is fairly new 2020 and it enables GPU cards to do their own scheduling rather than having the System CPU do the Scheduling.

Was that Microsoft release aimed at developers/programmers, or at end users?

Do you have a link?

There are plenty of how to,,, links but this one was out of the Microsoft DirectX group I believe.

https://devblogs.microsoft.com/directx/hardware-accelerated-gpu-scheduling/

Thanks
Bill F

https://devblogs.microsoft.com/directx/hardware-accelerated-gpu-scheduling/

Thanks for that.

So, we need a late-revision Windows 10 version (May 2020 update or later); a 'recent GPU that has the necessary hardware'; and 'a WDDMv2.7 driver that exposes this support to Windows'. Sorry - that rules me out, probably on all three counts.

And there's nothing about whether this will enhance the computational use of GPUs. The only technical commentary concentrates on frame buffering, user input, and latency: in other words, gaming.

Unless we have any Windows 'Insider' participants reading here (and not under NDA), we'll just have to wait until the promised

The goal of the first phase of hardware accelerated GPU scheduling is to modernize a fundamental pillar of the graphics subsystem and to set the stage for things to come… but that’s going to be a story for a another time 😊.

And there's nothing about whether this will enhance the computational use of GPUs. The only technical commentary concentrates on frame buffering, user input, and latency: in other words, gaming.

In that case one may want to remove the newest Windows Update: https://www.windowslatest.com/2021/04/22/nvidia-suggests-uninstalling-latest-windows-10-updates-to-fix-gaming-issues/
https://nextvame.com/kb5001330-and-kb5001337-updates-cause-problems-with-windows-10-and-games/

A lot of users have reported that having installed Windows 10 KB5001330 cumulative update, they have also experienced some bad stuttering and FPS drop when playing games.

Sub-Reddit R / Windows10 Gamers have reported that the KB5001330 Cumulative Update for Windows 10 2004 (“20H1”) and Windows 10 20H2 and KB5001337 for Windows 10 1903 and 1909 can actually cause blue screens and error messages 0x800f081f, 0x800f0984, or 0x800f0922 during installation.
Error 0x800f081f occurs when installing updates

In this context, current Nvidia GeForce 466.11 WHQL graphics driver and Intel 27.20.100.9466 graphics driver are frequently cited as the cause of crashes. Crashes during installation also occur with AMD graphics cards, regardless of whether they are Core i series or Ryzen.

Sorry - that rules me out, probably on all three counts.

Turns out I spoke too soon. I woke up the Windows 10 boot drive on a perfectly ordinary desktop machine, did a year's worth of updates (the machine runs Windows 7 day-to-day), and now that hardware acceleration setting is active.

The machine is an Intel i5, with HD 530 iGPU and NVidia GTX 1050 Ti - both drivers from Microsoft, so I get OpenCL support from Intel. At least five years old, and not the sharpest knife in the box even then. It all seems to run, but I don't think I can run a useful performance test.

Sorry - that rules me out, probably on all three counts.

Turns out I spoke too soon. I woke up the Windows 10 boot drive on a perfectly ordinary desktop machine, did a year's worth of updates (the machine runs Windows 7 day-to-day), and now that hardware acceleration setting is active.

The machine is an Intel i5, with HD 530 iGPU and NVidia GTX 1050 Ti - both drivers from Microsoft, so I get OpenCL support from Intel. At least five years old, and not the sharpest knife in the box even then. It all seems to run, but I don't think I can run a useful performance test.

Congratulations.... My system is too long in the tooth as well.
Intel(R) Xeon(R) CPU E5620 @ 2.40GHz with NVIDIA GeForce GTX 1060 6GB (4095MB) driver: 461.40 OpenCL: 1.2. The GPU driver is a Studio and one step off current.

I have reviewed some of the gaming site comments and many of those who had video issues resolved them by getting the current driver for their video card.

Bill F

---

In October 1969 I took an oath to support and defend the Constitution of the United States against all enemies, foreign and domestic;
There was no expiration date.

I believe Nvidia is researching in technologies to host a quad core ARM CPU inside the GPU, to offload that one CPU thread per GPU for deep learning.
Both AMD and NVidia are working towards having the GPU access the RAM directly, rather than invoke the CPU to transfer between RAM and GPU.
I read somewhere, AMD already uses it in some of their latest GPUs, but I don't know all the details about it.

If a project has both CPU and GPU projects, I would assume that a CPU project on a GPU will be slower than running the GPU optimized projects.