Invalid (81) : OK but ~9% ( I think < 5% is more typical )
Validation in BRP4 is still bad. 20% invalids per app version is no exception. Overall this also applies to the new Intel GPU version;, however as 'Beta' version we disallow comparison with results from the same app version. I'd really kike to see how comparison goes with CPU resutls.
The UHD 620 is currently running the last BRP4, before it switches fully over to BRP4G. Once that one's gathered in, I'll eyeball every one of the invalid tasks (currently 88), to see what it was compared with.
But what I'm genuinely more concerned about is the 'Validation inconclusive (97)', which we don't normally notice, don't care about, and can't investigate in detail.
Some new tasks of BRP7 with new version 0.05 of the Meerkat BRP7-cuda55 application just went thru my hosts. All failed with error 1015. The other hosts (GTX-1070) run them one at a time and other run two at a time (GTX-970). Both failed.
shows a new set of four applications for BRP7, ringing the changes on Windows and Linux, cuda55 and opencl_ati.
All list creation dates shorting after 14:45 UTC August 22, 2022
One of my Windows machines with an AMD GPU got six tasks.
1. Although the deadline is very short again (24 hours from issue?), mine did not start immediately. I've reduced my task queue substantially, so boincmgr did not immediately leap into panic mode.
3. application name for me is: 0.05 Binary Radio Pulsar Search (MeerKAT) (BRP7-opencl-ati)
4. initially running 1X, this keeps my GPU around 80% utilization, and CPU around 15%
5. with my system in equilibrium running GRP GPU tasks at 3X, the initial time completion estimate in my task queue for these was 21 minutes, which is not wildly different from the actual observed time of 12:31.
6. indicated progress seemed smooth, and the task finished promptly on reaching 100%.
7. it seems likely that running 2X and 3X will give modest, but not major productivity increase on my particular machine.
Got into a quorum with a Linux/Nvidia task, and my Windows/AMD compared poorly enough with it to generate an inconclusive. The tie-breaker is as yet unsent.
Some new tasks of BRP7 with new version 0.05 of the Meerkat BRP7-cuda55 application just went thru my hosts. All failed with error 1015. The other hosts (GTX-1070) run them one at a time and other run two at a time (GTX-970). Both failed.
Same errors here with rtx 2080 super, all 7 failed.
My second task of 0.05 Binary Radio Pulsar Search (MeerKAT) (BRP7-opencl-ati) ended in computation error after seemingly normal progress to the end.
While it was running I unsuspended GRP work in hopes of getting another BRP tasks download so that my multiplicity could update from 1X to 2X.
A consequence is that this task ran briefly with a GRP GPU task, and then ran briefly with the third BRP7 task. It is possibly that one of these two conditions contributed to the computation error.
I seem to be getting mixed results on ATI MeerKAT. Randomly on my cards (10 Tahiti chips and one Fury chip) some tasks run a lot faster and some run a lot slower. The GPU usage and heat generated seem to indicate it's working hard on something. Should these tasks all be the same size?
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
I also got some "new" BRP7-cuda55-WUs. On my machine with the RTX 3090, it seems it doesn't use the GPU, but only runs on the CPU. GPU-Utilisation is between 1% and 10%, runtime until now about 23.5 minutes and about 15% progress.
Mike Hewson wrote: Invalid
)
Validation in BRP4 is still bad. 20% invalids per app version is no exception. Overall this also applies to the new Intel GPU version;, however as 'Beta' version we disallow comparison with results from the same app version. I'd really kike to see how comparison goes with CPU resutls.
BM
The UHD 620 is currently
)
The UHD 620 is currently running the last BRP4, before it switches fully over to BRP4G. Once that one's gathered in, I'll eyeball every one of the invalid tasks (currently 88), to see what it was compared with.
But what I'm genuinely more concerned about is the 'Validation inconclusive (97)', which we don't normally notice, don't care about, and can't investigate in detail.
Some new tasks of BRP7 with
)
Some new tasks of BRP7 with new version 0.05 of the Meerkat BRP7-cuda55 application just went thru my hosts. All failed with error 1015. The other hosts (GTX-1070) run them one at a time and other run two at a time (GTX-970). Both failed.
Quite recently the
)
Quite recently the applications page:
https://einsteinathome.org/apps.php
shows a new set of four applications for BRP7, ringing the changes on Windows and Linux, cuda55 and opencl_ati.
All list creation dates shorting after 14:45 UTC August 22, 2022
One of my Windows machines with an AMD GPU got six tasks.
1. Although the deadline is very short again (24 hours from issue?), mine did not start immediately. I've reduced my task queue substantially, so boincmgr did not immediately leap into panic mode.
2. task names look like: M22_1_cfbf00002_segment_3_dms_200_85_1350000_1
3. application name for me is: 0.05 Binary Radio Pulsar Search (MeerKAT) (BRP7-opencl-ati)
4. initially running 1X, this keeps my GPU around 80% utilization, and CPU around 15%
5. with my system in equilibrium running GRP GPU tasks at 3X, the initial time completion estimate in my task queue for these was 21 minutes, which is not wildly different from the actual observed time of 12:31.
6. indicated progress seemed smooth, and the task finished promptly on reaching 100%.
7. it seems likely that running 2X and 3X will give modest, but not major productivity increase on my particular machine.
My lead task:
https://einsteinathome.org/workunit/666186426
Got into a quorum with a Linux/Nvidia task, and my Windows/AMD compared poorly enough with it to generate an inconclusive. The tie-breaker is as yet unsent.
I've got a few on different
)
I've got a few on different Windows machines with NVidia cards. Ran task 1341611765 by pausing the ones waiting in line.
Exit status:1015 (0x000003F7) Unknown error code
and
Harri Liljeroos skrev:Some
)
Same errors here with rtx 2080 super, all 7 failed.
https://einsteinathome.org/da/host/12870120/tasks/0/57
Edit: Got a few more, all failed. Have deselected the app for now.
My second task of 0.05 Binary
)
My second task of 0.05 Binary Radio Pulsar Search (MeerKAT) (BRP7-opencl-ati) ended in computation error after seemingly normal progress to the end.
While it was running I unsuspended GRP work in hopes of getting another BRP tasks download so that my multiplicity could update from 1X to 2X.
A consequence is that this task ran briefly with a GRP GPU task, and then ran briefly with the third BRP7 task. It is possibly that one of these two conditions contributed to the computation error.
The task stderr:
https://einsteinathome.org/task/1341568780
mentions:
(unknown error) - exit code -1073740940 (0xc0000374)
3 BRP7 failures after 19
)
3 BRP7 failures after 19 seconds.
Tullio
I seem to be getting mixed
)
I seem to be getting mixed results on ATI MeerKAT. Randomly on my cards (10 Tahiti chips and one Fury chip) some tasks run a lot faster and some run a lot slower. The GPU usage and heat generated seem to indicate it's working hard on something. Should these tasks all be the same size?
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
I also got some "new"
)
I also got some "new" BRP7-cuda55-WUs. On my machine with the RTX 3090, it seems it doesn't use the GPU, but only runs on the CPU. GPU-Utilisation is between 1% and 10%, runtime until now about 23.5 minutes and about 15% progress.