One or two of my computers, 7431929 for example, has been reporting 230 to 250 WK per day, mostly Cuda WU. I have noticed that the total number of WK In Progress never exceeds 200. I have the Minimum work buffer set to 10 days and the Max additional work buffer set to 10 days. On the Disk and Memory Usage tab I have it set to use at most 100Gb (actual is 180Mb), Leave 0.5Gb (actual is 150Gb), use at most 25% (actual is 0.04%). I would think these settings would allow more WU to be stored on the computer than 200 so that during the Weekly Outage and Initial Catch Up it will not be starved for WU. Does anyone know why there is such a low limit or how to overcome it?

---

100 work units for your CPU and 100 for each GPU that you have. Total for your machines is 200.

If you had 2 GPUs and the CPU then it would be 300..

3 GPUs and cpu then 400....

You get the idea.

First off : there is no way to overcome this. This limit was set by S@H a long time ago (100 tasks for the CPU and an additional 100 per GPU in your rig)

The theory behind this is the heavy duty crunchers were pounding the servers so hard they would collapse.

So you're not alone in this,everyone faces the same limits. The only workaround there is is to build a virtual machine, but this would only give you 100 more GPU tasks(and 100 more CPU of course, but that won't be a problem for anyone).

I hope I answered your question...

---

Thank you Etienne and Zalster for the explanation. Perhaps this limit was established in a time when processing power was much less capable than it is today, even for the servers which have been updated a few times since BOINC rolled out. I think I may look into building a VM or two.

---

Try to run few BOINc client instances instead. It will be cheaper from overhead point of vew than full-scale VM.

Thank you Etienne and Zalster for the explanation. Perhaps this limit was established in a time when processing power was much less capable than it is today, even for the servers which have been updated a few times since BOINC rolled out. I think I may look into building a VM or two.

Actually it is put in place to keep the servers from crashing and burning.

When the limits were not in place, several of the servers were crashing because their hard drives were full.

---

100 work units for your CPU and 100 for each GPU that you have. Total for your machines is 200.

If you had 2 GPUs and the CPU then it would be 300..

3 GPUs and cpu then 400....

You get the idea.

Which is, in a way, annoying... I have two machines with 20 cores each (and I've seen some similar with HyperThreading(TM) turned on so they run 40 WUs at once) and they are by no means the biggest CPU-count machines around. If I could work out how to get S@H running on my Xeon Phi co-processor... It has 60 cores and 240 threads, so it could never be filled under the current regime (but realistically it has only 4-5 GB of memory for all those threads to share). The next-generation Xeon Phi will actually have the processor as the onboard CPU rather than a co-processor -- which makes it look a much better prospect for BOINC/S@H. I hope I'm not breaking any NDAs when I say that it will have between 60 and 72 cores, again each with four hardware threads, but this time it will have access to system memory (as well as 16 GB of fast on-board RAM) so running that many jobs will be feasible.

---