Gamma-ray pulsar binary search #1 on GPUs

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2139
Credit: 2752977655
RAC: 1378969

Bernd Machenschalk

Bernd Machenschalk wrote:

Other than that, your result just doesn't match those of the other two closely enough for the tolerances defined in our validator - I'm not sure how that came about, but this occasionally happens to all of us.

We recently had a discussion on the validation rates for, specifically, the Binary Radio Pulsar Search (Arecibo) (BRP4) app on Intel iGPUs, and I posted some figures in comment 183546. My older, slower, iGPUs had a validation failure rate of some 3-4%, but my newest and fastest iGPU had a rate around 10%.

Since then, I've updated all my video drivers to the newest available direct from Intel, to test some complete failures at another project. Since then, my task completion times on all cards have fallen, but my validation failure rates have increased.

We have a suspicion that an OpenCL compiler optimisation flag generates faster, but less precise, binary kernels: I've heard a 'Fused Multiply-Add' opcode suggested as a culprit.

Could this be investigated, perhaps?

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33844032543
RAC: 36736388

FGRPB1G is out of work to

FGRPB1G is out of work to send. systems failed over to GW.

_________________________________________________________________________

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33844032543
RAC: 36736388

Ian&Steve C. wrote: FGRPB1G

Ian&Steve C. wrote:

FGRPB1G is out of work to send. systems failed over to GW.

and again.

_________________________________________________________________________

Tom M
Tom M
Joined: 2 Feb 06
Posts: 5589
Credit: 7675955957
RAC: 1861769

Ian&Steve C.

Ian&Steve C. wrote:

Ian&Steve C. wrote:

FGRPB1G is out of work to send. systems failed over to GW.

and again.

So you set on the website profile to send other work if no gamma ray is available?

I have not had much luck with that trick on e@h. So I run pure gr on my highest performing box.

Tom M

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33844032543
RAC: 36736388

it worked when GR was run

it worked when GR was run out. but when GR work returned, it didnt stop processing GW and flip back to GR like it's supposed to. it just kept crunching away on GW. 

 

so I just disabled that setting again.

_________________________________________________________________________

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33844032543
RAC: 36736388

Ian&Steve C.

Ian&Steve C. wrote:

Ian&Steve C. wrote:

FGRPB1G is out of work to send. systems failed over to GW.

and again.

and again.

 

seems like something is going on with the GR process or server. it's struggling to keep up the past few days. maybe the admins need to have it generate more work to have on standby, or initiate work generation sooner than when the available tasks hits zero to preempt it running dry.

_________________________________________________________________________

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33844032543
RAC: 36736388

Ian&Steve C.

Ian&Steve C. wrote:

Ian&Steve C. wrote:

Ian&Steve C. wrote:

FGRPB1G is out of work to send. systems failed over to GW.

and again.

and again.

 

seems like something is going on with the GR process or server. it's struggling to keep up the past few days. maybe the admins need to have it generate more work to have on standby, or initiate work generation sooner than when the available tasks hits zero to preempt it running dry.

and again

_________________________________________________________________________

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33844032543
RAC: 36736388

Ian&Steve C.

Ian&Steve C. wrote:

Ian&Steve C. wrote:

Ian&Steve C. wrote:

Ian&Steve C. wrote:

FGRPB1G is out of work to send. systems failed over to GW.

and again.

and again.

 

seems like something is going on with the GR process or server. it's struggling to keep up the past few days. maybe the admins need to have it generate more work to have on standby, or initiate work generation sooner than when the available tasks hits zero to preempt it running dry.

and again

and again.

_________________________________________________________________________

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3681
Credit: 33844032543
RAC: 36736388

Ian&Steve C.

Ian&Steve C. wrote:

Ian&Steve C. wrote:

Ian&Steve C. wrote:

Ian&Steve C. wrote:

Ian&Steve C. wrote:

FGRPB1G is out of work to send. systems failed over to GW.

and again.

and again.

 

seems like something is going on with the GR process or server. it's struggling to keep up the past few days. maybe the admins need to have it generate more work to have on standby, or initiate work generation sooner than when the available tasks hits zero to preempt it running dry.

and again

and again.

and again

_________________________________________________________________________

Tom M
Tom M
Joined: 2 Feb 06
Posts: 5589
Credit: 7675955957
RAC: 1861769

Once upon a time one of the

Once upon a time one of the differences in speed between the Nvidia GPUs and the Radeon GPUs was what happened at 89%.

The Radon GPUs would pause and then "snap" to 100%.

The Nvidia GPUs would keep munching and then (eventually) get to 100%

Lately, it seems like the Radeons are pausing a couple of minutes at 89% and then jumping to 100%.

Both of my Nvidia GPUs are offline at the moment.

Are the Nvidias munching steadily or are they pausing at 89% and jumping to 100% after a couple of minutes?

Thank you.

Tom M

 

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)

Comment viewing options

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