Windows Hardware Accelerated GPU Scheduling Benchmarks (Frametimes & FPS)
video description
Date: 2020-07-04
Related videos
Comments and reviews: 10
DVTye
I would love to be able to try out Windows Hardware Acceleration with my MSI GTX 1070 Gaming Z to see how it affects some games; but Microsoft's machine learning (ML) based process hasn't flagged my current PC (ASUS X570 Crosshair VIII Hero and 3900x) for an update to version 2004. Honestly though, I don't know weather or not if I want to though; as I've been reading and hearing a fair bit of information from articles, forums, and friends who have updated to version 2004 experience problems. Predominantly those with AMD platforms. One such friend had to rollback his windows to a previous version as he experienced problems with CPU scheduling in the Adobe Suite (more specifically it barely utilized half of his CPU and premiere starts dropping frames ). While another friend with a Zen build (3900x) experienced no issues and even stable higher overclocks on his GPU. I would love to know your perspective as to how Microsoft's algorithm chooses systems to receive the 2004 update when we have three similar PC builds with three distinct outcomes (mine won't even update even with a click on Check for Updates , while two of my friends can update however with contrasting experiences to the update).
reply
I would love to be able to try out Windows Hardware Acceleration with my MSI GTX 1070 Gaming Z to see how it affects some games; but Microsoft's machine learning (ML) based process hasn't flagged my current PC (ASUS X570 Crosshair VIII Hero and 3900x) for an update to version 2004. Honestly though, I don't know weather or not if I want to though; as I've been reading and hearing a fair bit of information from articles, forums, and friends who have updated to version 2004 experience problems. Predominantly those with AMD platforms. One such friend had to rollback his windows to a previous version as he experienced problems with CPU scheduling in the Adobe Suite (more specifically it barely utilized half of his CPU and premiere starts dropping frames ). While another friend with a Zen build (3900x) experienced no issues and even stable higher overclocks on his GPU. I would love to know your perspective as to how Microsoft's algorithm chooses systems to receive the 2004 update when we have three similar PC builds with three distinct outcomes (mine won't even update even with a click on Check for Updates , while two of my friends can update however with contrasting experiences to the update).
reply
Clayton
You need multiple processes competing for GPU resources in order for changes affecting scheduling of multiple processes to make any difference. Testing with one process should reveal no change. As has been stated below, an example of multiple processes utilizing the GPU would be streaming a game using GPU video encoding. I would suggest you run your tests while having a stream and recording using GPU encoding at the same time. That might show an improvement, as it is a situation that where there is actually some scheduling going on. Simply running a game by itself is practically no scheduling load. This is akin to benchmarking a six core CPU versus an eight core CPU that are both at the same clock speed for single thread workloads with a single-threaded workload. Of course they ll give results within the margin of error. You need to test multiple threads at the same time to reveal differences between six and eight cores. Test with one and two encoding jobs.
reply
You need multiple processes competing for GPU resources in order for changes affecting scheduling of multiple processes to make any difference. Testing with one process should reveal no change. As has been stated below, an example of multiple processes utilizing the GPU would be streaming a game using GPU video encoding. I would suggest you run your tests while having a stream and recording using GPU encoding at the same time. That might show an improvement, as it is a situation that where there is actually some scheduling going on. Simply running a game by itself is practically no scheduling load. This is akin to benchmarking a six core CPU versus an eight core CPU that are both at the same clock speed for single thread workloads with a single-threaded workload. Of course they ll give results within the margin of error. You need to test multiple threads at the same time to reveal differences between six and eight cores. Test with one and two encoding jobs.
reply
The
Would changing the driver specific setting of amount of pre rendering frames to zero might be a way to find out more about the possible differences of this feature?
As someone whos using windows insider fast ring on a 1080 (non TI) and 8700k i have not experienced any differences with this feature enabled (which btw has enabled itself automatically and i did not turn it on manually).
Concerning Red Dead 2 ...that game is such a shitshow of quirkyness especially in online mode that its really hard to tell what is what when it comes to strange behaviour. All i can say that the performance of that game on a 1080 has always been a thing constantly changing and altering itself on a miracolous basis....and my suspicions are far more directed towards the used DRM technology for a lot of the strange ultrications i have encountered with Rockstar Games horrible Launcher and the game itself.
reply
Would changing the driver specific setting of amount of pre rendering frames to zero might be a way to find out more about the possible differences of this feature?
As someone whos using windows insider fast ring on a 1080 (non TI) and 8700k i have not experienced any differences with this feature enabled (which btw has enabled itself automatically and i did not turn it on manually).
Concerning Red Dead 2 ...that game is such a shitshow of quirkyness especially in online mode that its really hard to tell what is what when it comes to strange behaviour. All i can say that the performance of that game on a 1080 has always been a thing constantly changing and altering itself on a miracolous basis....and my suspicions are far more directed towards the used DRM technology for a lot of the strange ultrications i have encountered with Rockstar Games horrible Launcher and the game itself.
reply
Kitkat5335
So mostly the typical Here is new technology for future use. Don't expect anything yet...we just implemented it, expect future functional use since most programs up to this point was programed without this feature available. Sounds better than the way that Nvidia Launched RTX and said 'It just works.' ( Note that it works once it gets added to things, don't expect it to work with anything that doesn't know how to make it work. In the mean time we will work on trying to make it work better then launch a refresh 6 months down the road). However, Microsoft isn't being as vague about it by saying expect future utility of it over current utility while also telling people stop getting over hyped...it isn't all that in use so you won't see it yet, but it is there (TL:DR version: It just works, but you won't see it work yet).
reply
So mostly the typical Here is new technology for future use. Don't expect anything yet...we just implemented it, expect future functional use since most programs up to this point was programed without this feature available. Sounds better than the way that Nvidia Launched RTX and said 'It just works.' ( Note that it works once it gets added to things, don't expect it to work with anything that doesn't know how to make it work. In the mean time we will work on trying to make it work better then launch a refresh 6 months down the road). However, Microsoft isn't being as vague about it by saying expect future utility of it over current utility while also telling people stop getting over hyped...it isn't all that in use so you won't see it yet, but it is there (TL:DR version: It just works, but you won't see it work yet).
reply
Raletia
I have a Fury so can't test the Scheduling yet, I'm curious if it gives any improvements to the longstanding issue that 2004 update FINALLY fixed: Window's handling of different refresh rates on multiple monitors. Before 2004, DWM would drop to the lowest refresh rate when any app on a lower Hz screen used any GPU acceleration. But now, DWM renders at the highest Hz and basically turns off VSync on the lower refresh rate monitors. Sometimes I've noticed some hitching in games that didn't exist before (On the latest optional AMD driver 20.5.1), wondering if with GPU scheduling it's any smoother with mixed refresh rate screens? (In my case, I have a 144hz main screen, a 60hz vertically on the right, and a 60 on the left, and another 60hz above the main.)
reply
I have a Fury so can't test the Scheduling yet, I'm curious if it gives any improvements to the longstanding issue that 2004 update FINALLY fixed: Window's handling of different refresh rates on multiple monitors. Before 2004, DWM would drop to the lowest refresh rate when any app on a lower Hz screen used any GPU acceleration. But now, DWM renders at the highest Hz and basically turns off VSync on the lower refresh rate monitors. Sometimes I've noticed some hitching in games that didn't exist before (On the latest optional AMD driver 20.5.1), wondering if with GPU scheduling it's any smoother with mixed refresh rate screens? (In my case, I have a 144hz main screen, a 60hz vertically on the right, and a 60 on the left, and another 60hz above the main.)
reply
Yellow
What i am wondering is if it impact power draw at all as well, since it's a task that change the demand of the hardware in question. There might be a small but interesting difference depending on gpu and cpu's mode of operation or simply type. And on that note, i have to ask if the change is worth it or not as well. And the same goes if one run multiple pieces of hard or soft hitting software on a machine. This also make a significant impact, worthy of a brisk test. (if a larger change is seen in terms of wattage pull, it might be telling if a game or combination is more or less efficient prior to the change, for the worse or better. I think it would be interesting or completely mundane but therefore worth a checkup if a game engine behave differently)
reply
What i am wondering is if it impact power draw at all as well, since it's a task that change the demand of the hardware in question. There might be a small but interesting difference depending on gpu and cpu's mode of operation or simply type. And on that note, i have to ask if the change is worth it or not as well. And the same goes if one run multiple pieces of hard or soft hitting software on a machine. This also make a significant impact, worthy of a brisk test. (if a larger change is seen in terms of wattage pull, it might be telling if a game or combination is more or less efficient prior to the change, for the worse or better. I think it would be interesting or completely mundane but therefore worth a checkup if a game engine behave differently)
reply
Nismo
Having tested it myself, I found within margin of error results but some people plead that it's an amazing difference.
From Microsoft themselves;
-What to expect when switching to the new GPU scheduler
-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.
-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 .
reply
Having tested it myself, I found within margin of error results but some people plead that it's an amazing difference.
From Microsoft themselves;
-What to expect when switching to the new GPU scheduler
-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.
-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 .
reply
BlackDahlia1147
with my asus strix 1080Ti, after enabling GPU scheduling my gaming performance has been relatively identical, BUT watching videos on my second monitor while gaming on the main monitor causes heavy stutters or freezes to the video feed. This is with two 1440P monitors hooked up via display port.
It seems to be a more obvious problem when gaming in borderless fullscreen rather than exclusive fullscreen. Games include Dead by Daylight, Squad, Call of Duty Warzones, and Detroit Become Human.
Kinda hoping this feature will get better with time, either on the driver side or the game engine side, it's a bit disappointing as is.
reply
with my asus strix 1080Ti, after enabling GPU scheduling my gaming performance has been relatively identical, BUT watching videos on my second monitor while gaming on the main monitor causes heavy stutters or freezes to the video feed. This is with two 1440P monitors hooked up via display port.
It seems to be a more obvious problem when gaming in borderless fullscreen rather than exclusive fullscreen. Games include Dead by Daylight, Squad, Call of Duty Warzones, and Detroit Become Human.
Kinda hoping this feature will get better with time, either on the driver side or the game engine side, it's a bit disappointing as is.
reply
tarfeef
What might be an interesting rest:
Measure input latency in a few games (or settings combinations) with low, medium, and high framerates, specifically with buffering disabled. This is realistic because some people, especially in competitive games, want every latency advantage possible.
Then, enable this feature, and see if you can achieve better performance at the same latency, better latency at the same performance, etc.
If you can also use this with buffering, that's a candidate for equal latency with higher performance.
It's almost assuredly going to be a wash, but I'm sure some are interested to see (myself included)
reply
What might be an interesting rest:
Measure input latency in a few games (or settings combinations) with low, medium, and high framerates, specifically with buffering disabled. This is realistic because some people, especially in competitive games, want every latency advantage possible.
Then, enable this feature, and see if you can achieve better performance at the same latency, better latency at the same performance, etc.
If you can also use this with buffering, that's a candidate for equal latency with higher performance.
It's almost assuredly going to be a wash, but I'm sure some are interested to see (myself included)
reply
ecvent0r
It would be interesting to see how this feature compares to enabling low latency mode(aka old max prerender frames = 1 setting) in the gpu driver input latency wise.
I'd guess the negative scaling in some of the games comes from the fact that with hw scaling batching is inherently disabled.
Typically enabling low latency mode in the driver has the same slightly negative effect on the fps however the input latency improvement is huge to the point that I pretty much run any game with it enabled except maybe turn based games. (running 6700K + GTX 1080 btw)
reply
It would be interesting to see how this feature compares to enabling low latency mode(aka old max prerender frames = 1 setting) in the gpu driver input latency wise.
I'd guess the negative scaling in some of the games comes from the fact that with hw scaling batching is inherently disabled.
Typically enabling low latency mode in the driver has the same slightly negative effect on the fps however the input latency improvement is huge to the point that I pretty much run any game with it enabled except maybe turn based games. (running 6700K + GTX 1080 btw)
reply
Add a review, comment
Other channel videos
