What exactly is hardware GPU scheduling?

How is this superior to software scheduling currently?

https://www.tomshardware.com/news/am...20-beta-driver

Yousuf Khan wrote:

How is this superior to software scheduling currently?

https://www.tomshardware.com/news/am...20-beta-driver

That article links to:

https://www.tomshardware.com/news/gp...eduling-tested

Not much to care there. Hardware scheduling has been a firmware feature
of GPUs for awhile. Now they're adding an API for drivers and games to
access it. Unfortunately, the benchmarks at the 2nd site never mention
CPU load and how its affected based on the number of cores.

On 2020-06-29 2:08 p.m., Yousuf Khan wrote:
How is this superior to software scheduling currently?

https://www.tomshardware.com/news/am...20-beta-driver


Hi Peter and Paul, I followed the link you gave and ended up in the
app store , searched for and found it somehow after a few screwy pages
clicked GET and it downloaded , installed and launched in a Command box
as advertised.

The version number is 0.0.11.761.0, You need Windows 2004 19041.xxx to
run it.
Something screwy about getting it as I seemed to require a little
runaround in the store.
But it is there


Rene

On 6/29/2020 7:10 PM, Rene Lamontagne wrote:
On 2020-06-29 2:08 p.m., Yousuf Khan wrote:
How is this superior to software scheduling currently?

https://www.tomshardware.com/news/am...20-beta-driver


Hi Peter and Paul,Â*Â* I followed the link you gave and ended up in the
appÂ* store , searched for and found it somehow after a few screwy pages
clicked GET and it downloaded , installed and launched in a Command box
as advertised.

The version number is 0.0.11.761.0, You need Windows 2004 19041.xxx to
run it.
Something screwy about getting it as I seemed to require a little
runaround in the store.
But it is there

What are you talking about?

Yousuf Khan

Yousuf Khan wrote:
On 6/29/2020 7:10 PM, Rene Lamontagne wrote:
On 2020-06-29 2:08 p.m., Yousuf Khan wrote:
How is this superior to software scheduling currently?

https://www.tomshardware.com/news/am...20-beta-driver


Hi Peter and Paul, I followed the link you gave and ended up in the
app store , searched for and found it somehow after a few screwy
pages clicked GET and it downloaded , installed and launched in a
Command box as advertised.

The version number is 0.0.11.761.0, You need Windows 2004 19041.xxx to
run it.
Something screwy about getting it as I seemed to require a little
runaround in the store.
But it is there

What are you talking about?

Yousuf Khan


That's a mis-post, attached to the wrong thread.

It was intended for the WinFR thread.

Paul

On 2020-06-30 9:50 a.m., Paul wrote:
Yousuf Khan wrote:
On 6/29/2020 7:10 PM, Rene Lamontagne wrote:
On 2020-06-29 2:08 p.m., Yousuf Khan wrote:
How is this superior to software scheduling currently?

https://www.tomshardware.com/news/am...20-beta-driver


Hi Peter and Paul,Â*Â* I followed the link you gave and ended up in the
appÂ* store , searched for and found it somehow after a few screwy
pages clicked GET and it downloaded , installed and launched in a
Command box as advertised.

The version number is 0.0.11.761.0, You need Windows 2004 19041.xxx
to run it.
Something screwy about getting it as I seemed to require a little
runaround in the store.
But it is there

What are you talking about?

Â*Â*Â* Yousuf Khan


That's a mis-post, attached to the wrong thread.

It was intended for the WinFR thread.

Â*Â* Paul

Yes I reposted properly at 6:43 p.m. yesterday.

Rene

On 6/30/2020 11:07 AM, Rene Lamontagne wrote:
On 2020-06-30 9:50 a.m., Paul wrote:
Yousuf Khan wrote:
On 6/29/2020 7:10 PM, Rene Lamontagne wrote:
On 2020-06-29 2:08 p.m., Yousuf Khan wrote:
How is this superior to software scheduling currently?

https://www.tomshardware.com/news/am...20-beta-driver


Hi Peter and Paul,Â*Â* I followed the link you gave and ended up in
the appÂ* store , searched for and found it somehow after a few
screwy pages clicked GET and it downloaded , installed and launched
in a Command box as advertised.

The version number is 0.0.11.761.0, You need Windows 2004 19041.xxx
to run it.
Something screwy about getting it as I seemed to require a little
runaround in the store.
But it is there

What are you talking about?

Â*Â*Â* Yousuf Khan


That's a mis-post, attached to the wrong thread.

It was intended for the WinFR thread.

Â*Â*Â* Paul

Yes I reposted properly at 6:43 p.m. yesterday.

Rene

Okay.

On 6/29/2020 3:08 PM, Yousuf Khan wrote:
How is this superior to software scheduling currently?

https://www.tomshardware.com/news/am...20-beta-driver

Now, according to some of the websites, hardware scheduling has been
available in GPU's for a long time now, but it's only now that the
support for the feature has been implemented in DirectX 12. But
according to this AMD read.me the new driver is only compatible with the
Radeon RX 5000-series (RDNA), no mention of previous architectures like
GCN. Does this mean that this scheduler is not available in the older
architectures, or just that the current implementation isn't of the
drivers is limited to RDNA?

https://www.amd.com/en/support/kb/re...0-5-1-ghs-beta

It seems like HW scheduling might be a godsend for people with older
machines which have a reputation of being somewhat high-latency, such as
my FX-8300 CPU.

Yousuf Khan

Yousuf Khan wrote:
On 6/29/2020 3:08 PM, Yousuf Khan wrote:
How is this superior to software scheduling currently?

https://www.tomshardware.com/news/am...20-beta-driver


Now, according to some of the websites, hardware scheduling has been
available in GPU's for a long time now, but it's only now that the
support for the feature has been implemented in DirectX 12. But
according to this AMD read.me the new driver is only compatible with the
Radeon RX 5000-series (RDNA), no mention of previous architectures like
GCN. Does this mean that this scheduler is not available in the older
architectures, or just that the current implementation isn't of the
drivers is limited to RDNA?

https://www.amd.com/en/support/kb/re...0-5-1-ghs-beta

It seems like HW scheduling might be a godsend for people with older
machines which have a reputation of being somewhat high-latency, such as
my FX-8300 CPU.

Yousuf Khan

The biggest win on DX12 versus DX11, is DX12 uses display lists
and DX11 uses "objects". To display a scene in DX11 required
dispatching thousands of "objects". Whereas in DX12, it's bad to
the old days, where a developer simply lards up a display list
and dispatches the whole thing. This is why DX12 is claimed to
be 30% faster than DX11.

The transition from DX11 to DX12 is intended to put the
developer back in control, in a sense.

The benchmarks for the scheduling alternative, don't seem all that
impressive. Perhaps because it's early days for the effort. And
probably every generation of hardware needs separate optimization.
And since it seems to make squat difference, this is presumably
"hard work".

The games being run, the developer likely developed the game, without
this feature in mind.

The question becomes, after a while, exactly how much driver level
fiddling do you want to saddle developers with. There will be
some who welcomed the opportunity to program their own display list.
And others who just wanted to make a game, and are less obsessed
with game engines.

Paul

On 7/2/2020 12:56 PM, Paul wrote:
The benchmarks for the scheduling alternative, don't seem all that
impressive. Perhaps because it's early days for the effort. And
probably every generation of hardware needs separate optimization.
And since it seems to make squat difference, this is presumably
"hard work".

I think the websites have so far only tested with high-end hardware,
high-end processors (9900K) and high-end GPU's (2080Ti) where it
probably makes no difference whether you use software or hardware
scheduling. The real benefits should come with low-end hardware, both
CPU's and GPU's that are not top-of-the-line anymore.

Yousuf Khan