I looked back on the thread and it looks like they were talking about an brp4 intel-gpu app that was in beta. The 530 iGpu is on windows.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Richard’s 530 is running on windows. But you can run Intel GPU on any OS if there’s an app. That’s why I asked. I’ve run my laptop iGPU under Linux occasionally and noticed the same validation problem. That’s why I asked if they could make a similar new beta app for Linux.
First one back, successful completion. But remember this is about validation, so we need to get a statistical summary over many validated jobs.
Fortunately, these things go like the clappers once they get going (if you know how) - 657 seconds elapsed - so we should have some results for Monday.
OK, it's about 10 hours since this test really got under way. The scoreboard so far reads:
All (91)
In progress (25)
Pending (37)
Valid (21)
Invalid (0)
When you compare that with the last report from this machine, message 189926, that's a great improvement. Most welcome of all is the last line, zero invalid.
But we're not out of the woods yet. Check the figures above - they don't add up. The separate categories add up to 83, when the total is 91. The missing 8 will be 'validation inconclusive', and that's the critical group. I'm just loading a new batch to run overnight, and we'll see how it looks in the morning.
How’s the speed difference? Is that slower than before without the optimizations?
another thing is why is this app for BRP4 and not BRP4G? They previously were making tasks for the Intel GPUs only the G variant of tasks to help alleviate stress on their server.
Also, in the same vein of the previous request for a new version of the Linux BRP4/G Intel GPU app, would be nice to get a Linux Intel GPU version of the FGRPB1G app, which also is currently only built for Windows.
All (242)
In progress (109)
Pending (60)
Valid (59)
Invalid (0)
Valid going up, invalid still zero: but the void is still there. I make it 14 inconclusives now.
On the speed question: my experience is that with the ongoing Intel driver/runtime development, speeds have been going up, and accuracy has been going down. So it's a long time since I ran this particular app, and I didn't keep exact records. The sample time I posted for the first task was an outlier - I couldn't get my local management tools in place until the app had downloaded and started running. But my impression is that the new app/driver combination is both faster and more accurate - win-win.
Bernd will have to speak to the BRP4G issue - but remember that the Arecibo data archive is a finite resource: since the telescope collapsed, no new data has been added.
OK, I'm a little over-fetched
)
OK, I'm a little over-fetched after those experiments. Running the cache down to minimum now, will check at intervals when I've got rid of the glut.
Linux Intel GPU BRP4 fix?
)
Linux Intel GPU BRP4 fix?
_________________________________________________________________________
The working week has already
)
The working week has already finished in Germany (where Bernd is based). Let's concentrate on one at a time over the weekend?
Ian&Steve C. wrote:Linux
)
I looked back on the thread and it looks like they were talking about an brp4 intel-gpu app that was in beta. The 530 iGpu is on windows.
Tom M
A Proud member of the O.F.A. (Old Farts Association). Be well, do good work, and keep in touch.® (Garrison Keillor) I want some more patience. RIGHT NOW!
Richard’s 530 is running on
)
Richard’s 530 is running on windows. But you can run Intel GPU on any OS if there’s an app. That’s why I asked. I’ve run my laptop iGPU under Linux occasionally and noticed the same validation problem. That’s why I asked if they could make a similar new beta app for Linux.
_________________________________________________________________________
I didn't expect Bernd to sort
)
I didn't expect Bernd to sort out the plan_class/driver issue yesterday, but looking today, I see v1.70 'opencl-intel_gpu-newer' is there already.
I've force-downloaded a batch, which should start running soon. Results will start to appear at https://einsteinathome.org/host/12496320/tasks/0/19.
(edit - at least the downloads prove that my driver is accepted now)
First one back, successful
)
First one back, successful completion. But remember this is about validation, so we need to get a statistical summary over many validated jobs.
Fortunately, these things go like the clappers once they get going (if you know how) - 657 seconds elapsed - so we should have some results for Monday.
OK, it's about 10 hours since
)
OK, it's about 10 hours since this test really got under way. The scoreboard so far reads:
All (91)
In progress (25)
Pending (37)
Valid (21)
Invalid (0)
When you compare that with the last report from this machine, message 189926, that's a great improvement. Most welcome of all is the last line, zero invalid.
But we're not out of the woods yet. Check the figures above - they don't add up. The separate categories add up to 83, when the total is 91. The missing 8 will be 'validation inconclusive', and that's the critical group. I'm just loading a new batch to run overnight, and we'll see how it looks in the morning.
How’s the speed difference?
)
How’s the speed difference? Is that slower than before without the optimizations?
another thing is why is this app for BRP4 and not BRP4G? They previously were making tasks for the Intel GPUs only the G variant of tasks to help alleviate stress on their server.
Also, in the same vein of the previous request for a new version of the Linux BRP4/G Intel GPU app, would be nice to get a Linux Intel GPU version of the FGRPB1G app, which also is currently only built for Windows.
_________________________________________________________________________
Another interim
)
Another interim report:
All (242)
In progress (109)
Pending (60)
Valid (59)
Invalid (0)
Valid going up, invalid still zero: but the void is still there. I make it 14 inconclusives now.
On the speed question: my experience is that with the ongoing Intel driver/runtime development, speeds have been going up, and accuracy has been going down. So it's a long time since I ran this particular app, and I didn't keep exact records. The sample time I posted for the first task was an outlier - I couldn't get my local management tools in place until the app had downloaded and started running. But my impression is that the new app/driver combination is both faster and more accurate - win-win.
Bernd will have to speak to the BRP4G issue - but remember that the Arecibo data archive is a finite resource: since the telescope collapsed, no new data has been added.