Folding Forum

You might come across a new item in the Windows 10 version of the NVidia Control Panel, if you installed the most recent drivers and O/S updates.

Essentially, it is saying that Windows O/S now manages selection of the graphics processor. And gives you a link to the Windows 10 Graphics Settings.

You are then given a choice as to whether or not you want to use Hardware Accelerated GPU Scheduling. I won't tell you how to proceed, as you know your configurations better than I do. However, proceed with caution and give your feedback as to whether or not turning it on helps (should you do it).

Paul

Note: That may also impact those w/AMD GPU's too.

Paul

Assuming you're talking about this feature: https://devblogs.microsoft.com/directx/ ... cheduling/ I chatted with some fellow folders and it seems to have no impact currently. However, YMMV

Maybe it has an impact on latency and/or throughput when you're GPU folding and gaming at the same time? It would be nice if this is a path to actual GPU process priority.

We will all have to see. I put it to on, and will spot check the FAH logs as often as I can.

Paul

Is this just a new name for the feature that I'll call "Should the GPU be used to render the desktop or do you want to use CPU rendering for the desktop?"

Ask yourself: What would Microsoft do with a GPU if it had access to one?

psaam0001 wrote:You are then given a choice as to whether or not you want to use Hardware Accelerated GPU Scheduling.

Just a wild guess, but I'll bet the feature only applies to Microsoft's own Desktop, not to foreign apps like FAH.

This may also include rendering your MS browser if it happens to carry the Microsoft brand name, but if you run FireFox/Opera/etc. they have their own settings.

If my guess is right, it's not a new feature. It's just been renamed. The feature has been part of Windows for as long as I've had a GPU.

Hardware Accelerated GPU Scheduling(HAGS) applies to all applications using GPU because it reduces overhead and latencies in Windows Display Driver Model (WDDM). But it has only a small effect on games or CUDA - like 1%.

The transition should be transparent, and users should not notice any significant changes. Although the new scheduler reduces the overhead of GPU scheduling, most applications have been designed to hide scheduling costs through buffering.

https://devblogs.microsoft.com/directx/ ... cheduling/

Thanks for that. The documentation you reference is really good information ... and my earlier assumptions are wrong (NP).

FAH is designed to buffer every frame so as soon as frame N has been processed, the data needed to process frame (N+1) is already in GPU memory (NVidia) simply queued for processing (AMD). That means that you won't see a change in the processing of WUs unless there's a fundamental change to how the FAHCores manage their data.

Sometimes, when a "slow" GPU is folding, there is a noticeable screen-lag. Hopefully this feature will alleviate this problem. In those cases, there is a noticeable lag when you move your mouse or you type something. This feature SHOULD allow windows to preempt the busy queue and get immediate high priority access to the GPU. This is worthy of a test, but you'd have to figure out if your (old) "slow" GPU has drivers that can support this feature and then test it twice, with the new feature on and off.

Q1) Can FAH dispense with the 1-cpu support processing thread that is always reserved to support FAH?
Q2) Which GPUs will have drivers that will support this new feature and which will not?

When a FAHCore is running on an nVidia GPU, portions of the WU or all of the WU are preloaded (cached) to GPU memory and it works on it from there. AMD advertises that it can work on data directly from Main RAM without need for the caching.

I wonder if the FAHCores are smart enough to take advantage of that difference.

P13428, 2:07 TPF prior to enabling Hardware Accelerated GPU Scheduling; 1:58 TPF after enabling. GTX 1070SC, game ready driver 456.55, W10-64 2004, my only windows folding box. Totally worth the two minutes spent.
How to: https://www.ghacks.net/2020/07/03/why-y ... indows-10/

That sound good is it reproducable with other work units too? Then this really would be a recommendation to update Windows 10 to 2004 or newer. Is the option enabled by default or the users need to switch it on?

foldy wrote:That sound good is it reproducable with other work units too? Then this really would be a recommendation to update Windows 10 to 2004 or newer. Is the option enabled by default or the users need to switch it on?

It appears to be reproducible with other work units and I saw an immediate bump of about 10% in my PPD once enabled. Will continue to monitor. Users will need to switch it on.

tried on one of mine and for 13428, TPF improvement of 10s.

Does this give performance improvement for both Nvidia CUDA and AMD OpenCL with FAH?

personally I can't say, I only have Nvidia cards, definitely works on a 1070.