The O2-All Sky Gravitational Wave Search on GPUs - discussion thread.

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4312
Credit: 250425925
RAC: 35123

There is now version 1.08 

There is now version 1.08  which has a tiny change in the OpenCL code and is otherwise identical to 1.07. This has a data-dependent effect on the result, it is completely unknown how large this is on a larger scale. We'll gather some data over the weekend and see how validation develops.

BM

Matt White
Matt White
Joined: 9 Jul 19
Posts: 120
Credit: 280798376
RAC: 0

Bernd Machenschalk

Bernd Machenschalk wrote:
There is now version 1.08  which has a tiny change in the OpenCL code and is otherwise identical to 1.07. This has a data-dependent effect on the result, it is completely unknown how large this is on a larger scale. We'll gather some data over the weekend and see how validation develops.

I just noticed them in my work queue, thanks, Bernd. Looking forward to seeing the results.

Clear skies,
Matt
Matt White
Matt White
Joined: 9 Jul 19
Posts: 120
Credit: 280798376
RAC: 0

My first 2 v1.08 tasks are in

My first 2 v1.08 tasks are in the pending queue.

Clear skies,
Matt
DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 0

A bit belated because I've

A bit belated because I've been busy, my windows/nvidia results on 1.07 are:

WIN 10, i7-4790K. GF 1080

1x task 14-37% GPU load, 4800 seconds

2x task 38-67% GPU load, 7100 seconds for an ~50% speedup

3x task 55-84% GPU load (generally above 80 when not doing startup stuff), but 9600 second run times so no actual speedup.

 

I also had 2 systems running 1070s that I only ran single tasks on.

i7-4770k which did single tasks in 5100 seconds.

i7-930 which took 11000 seconds.

 

Between the 1080 offering barely any speedup over the 1070 (and by an amount small enough it might be down to CPU clocks), and my old 930 box taking nearly twice as long are still heavily CPU bound, and not just eating full cores because NVidia's openCL driver sucks.

 

 

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4312
Credit: 250425925
RAC: 35123

It looks like all results of

It looks like all results of 1.08 are 'invalid'. Retracted / deprecated.

 

BM

Matt White
Matt White
Joined: 9 Jul 19
Posts: 120
Credit: 280798376
RAC: 0

Bernd Machenschalk wrote:It

Bernd Machenschalk wrote:

It looks like all results of 1.08 are 'invalid'. Retracted / deprecated.

 

Well now, that's not good news. Not totally unexpected based considering I had no valid results after a day of crunching.

I aborted the balance of v1.08 tasks in my work queue.

Clear skies,
Matt
Betreger
Betreger
Joined: 25 Feb 05
Posts: 992
Credit: 1588272403
RAC: 759980

My GTX1060 produces a fair

My GTX1060 produces a fair number of invalid tasks, is this helping the project or just wasting electricity? It doe also produce valid results. When running BTP tasks it is totally reliable. 

cecht
cecht
Joined: 7 Mar 18
Posts: 1533
Credit: 2899765560
RAC: 2161471

My one host had regularly

My one host had regularly been getting usually two new O2AS20-500 v1.07 tasks about every 1 hr 15 min, but stopped getting new work at 8:19am CDT ("Scheduler request completed: got 0 new tasks"). My other host continues to get O2AS20-500 v1.07 tasks. (I just now reset Project preferences for the "lazy host" and it can get binary pulsar #1 GPU work, which it is running now.) The lack of v1.07 work appears to be a host-specific issue, but I can't figure it out. help?

Ideas are not fixed, nor should they be; we live in model-dependent reality.

Jim1348
Jim1348
Joined: 19 Jan 06
Posts: 463
Credit: 257957147
RAC: 0

My only AMD card is an RX 570

My only AMD card is an RX 570 on Windows.  It has failed (with invalids) all the 1.07 and 1.08.  How about a CUDA version?  I don't know if that would simplify development, but working Nvidia cards are more useful than non-working AMD cards.

I am perfectly willing to buy either one by the way. 

 

archae86
archae86
Joined: 6 Dec 05
Posts: 3157
Credit: 7220524931
RAC: 973070

Jim1348 wrote:My only AMD

Jim1348 wrote:
My only AMD card is an RX 570 on Windows.  It has failed (with invalids) all the 1.07 and 1.08.  How about a CUDA version?  I don't know if that would simplify development, but working Nvidia cards are more useful than non-working AMD cards.

I have two RX 570 Windows systems.  Both are getting very high (>95%) validation rates on 1.07, once I stopped running at multiplicities of 3X or 4X, both of which failed 100%.  Both systems also failed 100% of their 1.08 work.

But the reason your RX 570 Windows system has failed 100% of 1.07 tasks is not the brand of AMD on the card, nor the model.  Based on elapsed times, I don't think you are running 3X or 4X, and don't have a guess as to what might be wrong.

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.