NaCL-support for GPU in future?

Moderators: Site Moderators, PandeGroup

Re: NaCL-support for GPU in future?

Postby Joe_H » Sat Mar 04, 2017 9:10 pm

This is getting off-topic from the original question about NaCl support for GPU's. But, there have been a number of posts before explaining the difference in this forum.

The nVidia driver support places a CPU core in a loop waiting for OpenCL instructions to transfer data to and from the GPU. This was how nVidia chose to implement their support for OpenCL.

AMD's drivers have been reported to wait for an interrupt before requiring CPU cycles. So the CPU appears to be less used.
Image

iMac 2.8 i7 12 GB smp8, Mac Pro 2.8 quad 12 GB smp6
MacBook Pro 2.9 i7 8 GB smp3
Joe_H
Site Admin
 
Posts: 3823
Joined: Tue Apr 21, 2009 4:41 pm
Location: W. MA

Re: NaCL-support for GPU in future?

Postby bruce » Sat Mar 04, 2017 11:48 pm

A spin-wait uses CPU resources to accelerate the initiation of the PCIe I/O by avoiding the overhead of an interrupt driven response.

Case 1: (NV)
> Is there IO to do?
> If no, branch to previous statement.
> (If yes) handle I/O

Case 2: (AMD)
> Is there I/O to do?
> If no, set a path to capture an interrupt.
> Wait.
..
..
..
> On Interrupt: sort out interrupt
> If it's ours, Branch to I/O handler code.

Setting the interrupt takes a number of instructions, as does decoding the interrupt whenever it happens.
bruce
 
Posts: 21293
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: NaCL-support for GPU in future?

Postby foldinghomealone » Sun Mar 05, 2017 12:38 am

Does that mean that if AMD had chosen a similar approach like NVIDIA for OpenCL support then AMD GPUs would be faster?

I'm sorry, I know it's OT, but it's so interesting to understand more and more how it works.
foldinghomealone
 
Posts: 50
Joined: Wed Feb 01, 2017 7:07 pm

Re: NaCL-support for GPU in future?

Postby bruce » Mon Mar 06, 2017 3:50 am

Probably. in cases where transferring data to/from the screen is a limiting factor. If 3D computations by the GPU are the limiting factor, probably not.

For a long time FAH was limited by the compute speed and the common statement was that the speed of the PCIe bus didn't matter. Many of the proteins being processed today are huge in comparison and in those cases, the PCIe bus speed is beginning to matter. (There are a number of discussion on this forum associating numbers with specific configurations.)

The word "speed" consists of a combination of certain latency while the CPU is preparing itself to send/receive the data as well as the transfer speed, once the data starts flowing. Latency for interrupt driven I/O is greater than latency for spin-wait I/O.

All this discussion is certainly off-topic, since we're talking about Windows/Linux/MacOS rather than NaCl.
bruce
 
Posts: 21293
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: NaCL-support for GPU in future?

Postby foldinghomealone » Sun Mar 12, 2017 8:51 pm

As my questions went OT long ago I started a related topic but answers given have been more opinions than facts.
viewtopic.php?f=38&t=29789
If you could give some more facts I would be thankful.
Thank you for your great contribution and help.
foldinghomealone
 
Posts: 50
Joined: Wed Feb 01, 2017 7:07 pm

Previous

Return to NaCl client (Chrome);

Who is online

Users browsing this forum: No registered users and 1 guest

cron