To the developers,
My i7 is working hard for Einstein@home with S5 jobs and the ABP CUDA-app. However I got a GTX 285 graphics card with one processor, so it runs only one CUDA-app. at a time.
However I got 33 WU's to crunch yesterday late.
This is too many to handle in time I think. The additional work buffer is set on 0.30 days, and this is not change in the last months.
Can there something be done with the load of CUDA-apps. as I can run only one at a time. With this huge load I will not get any S5 jobs, som only one cpu-core is running then.
Thanks.
Greetings from
TJ
Copyright © 2025 Einstein@Home. All rights reserved.
Too many CUDA-WU's
)
A problem with to many units is mostly related to a buggy boinc version.
That's a BOINC problem, not
)
That's a BOINC problem, not an Einstein one. You should try an other BOINC version (older or newer, can't tell since your computers are hidden). See the BOINC dev boards for more info.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)

RE: However I got a GTX 285
)
[nitpicking-mode]
Your one GPU has 240 unified shaders/stream processors, which means that the one CUDA task is attacked by those 240 processors in parallel. So it doesn't have just one processor, but 240 of them. It only has one Graphics Processing Unit, but quite a powerful one at that.
[/nitpicking-mode]
I am using BOINC version
)
I am using BOINC version 6.4.7 for a long time now and it is very stable. I run it on quads as well, but with now CUDA. The i7 is running other projects as well and there now overload of WU’s have even been occurred. With Einstein@home, this is not the first time this happened. My concern is that at this time I have to wait for all the WU’s are done (approx. 166 hours) before I get new and hopefully S5 WU’s.
Because as I run for months now with version 6.4.7 on several pc’s I do not think it is a “buggy†version.
Greetings from
TJ
RE: Because as I run for
)
That version is probably doing just fine with CPU tasks, but the complication is that if you are using the beta test app, you are running a mix of CPU and GPU tasks from the same project. And that complicates things a lot. The BOINC client must now try to keep both CPU and GPU busy AND make sure that the resource shares of the different projects are taken into account as well, which is not trivial. What worked in a CPU-only task environment may fail in a mixed task environment.
CU
Bikeman
RE: That version is
)
Hello Bikeman,
You are probably right but I still have some remarks.
I run this version 6.4.7 from after two months as it came out. It ran Einstein@home CPU tasks and later GPUGRID CUDA tasks. It took only 3 of these CUDA tasks at a time when updating. When the Einstain@home CUDA app. came out I stopped with the GPUGRID and ran only Einstein@home, CPU and GPU tasks. My i7 is set to run 7 tasks. At times I got 3 CUDA tasks from Einstein@home and sometimes 7 of them, but now more and with a mix with CPU tasks all ran smoothly with a lot jobs done per day. And now since 24 September I got 33 CUDA jobs from Einstein@home at once. That means only one CPU is running and approx 5 jobs are done per day. So despite the good ideas I get from the forum I still believe something happened at Einstein@home during 24 September as before that all went very very smoothly since the beginning of the CUDA app.
Greetings from
TJ
The problem is that your
)
The problem is that your BOINC v6.4.7 only keeps a single queue of work - it just thinks that you have 'enough work', it doesn't know about 'enough CUDA, not enough CPU'. You will have been getting hierarchical (S5R5) tasks, which run on the CPU, and ABP1 tasks which run on CUDA: the problem must have arisen because, at the instant you needed work, only ABP1 tasks were available.
If you install BOINC v6.6.38 (now the recommended download, for Windows at least - you don't mention what OS you use), it will maintain separate queues for CPU and CUDA: it will 'see' that your i7 is idle, and ask (and keep asking) for CPU work separately until it gets some.
I am trying to download BOINC
)
I am trying to download BOINC 6.6.40, the recommended version for Linux, but all I get is a 404 error: file not found.
Tullio
OK, found it. It works.
RE: I am trying to download
)
Hi Tullio.
Same here, but you can try this alternative link
http://boincdl3.ssl.berkeley.edu/mirror/boinc_6.6.40_x86_64-pc-linux-gnu.sh for the 64 bit version.
CU
Bikeman
Thanks Bikeman, but I only
)
Thanks Bikeman, but I only have a 32-bit SuSE 10.3 Linux. I am waiting for 11.2 in November to switch to 64 bit. The BOINC manager starts the client, contrarily to 6.6.36. Cheers.
Tullio