LHC@home

What I have noticed alreade long time ago, and now again when I redistributed the LHC subprojects between my computers: Sixtrack seems to challenge the CPUs much more than the VM tasks, thus increasing the CPU temperature considerabely.
Any explanation why this is the case?

I have SixTrack running on a Linux VirtuaMachine hosted on a Windows 10 PC which is running WorldCommunityGrid on the same CPU. CPU temperature is 64 C.
Tullio

---

What I have noticed alreade long time ago, and now again when I redistributed the LHC subprojects between my computers: Sixtrack seems to challenge the CPUs much more than the VM tasks, thus increasing the CPU temperature considerabely.
Any explanation why this is the case?

Yes, this is the observation I have made also. I don't know the actual reason for this.

---

I have SixTrack running on a Linux VirtuaMachine hosted on a Windows 10 PC which is running WorldCommunityGrid on the same CPU. CPU temperature is 64 C.
Tullio

to give an example from here:
Notebook with AMD Ryzen 5 4500U (6-core), Windows10 -
5 Theory tasks running simultaneously: temperature ~ 74°C.
When replacing 1 Theory task with 1 Sixtrack task: ~ 82°C

When running only SixtTrack on the Linux Virtual Machine temperature drops to 52°C. CPU is a Ryzen 5 1400 with 4 cores.
Tullio

---

Could be as it uses more AVX? Anyway, the code for sixtrack is super well optimised as it's from 1985 when computers were not powerful so it was a heavy focus back in the day.

Could be as it uses more AVX? Anyway, the code for sixtrack is super well optimised as it's from 1985 when computers were not powerful so it was a heavy focus back in the day.

I think that it is the general optimization of the code that is better for sixtrack and that it does not need to communicate constantly with the servers at CERN when running. The runtime differences between AVX and SSE programs are very small.

---

The runtime differences between AVX and SSE programs are very small.

and what is interesting to observe: the same machine sometimes downloads the AVX version of a Sixtrack task, sometimes the SSE version.
from what I remember, once it has turned out which version is the better one for a given machine, only this one would be downloaded. In fact, the same observation I made on another of my computers as well.

The runtime differences between AVX and SSE programs are very small.

and what is interesting to observe: the same machine sometimes downloads the AVX version of a Sixtrack task, sometimes the SSE version.
from what I remember, once it has turned out which version is the better one for a given machine, only this one would be downloaded. In fact, the same observation I made on another of my computers as well.

Boinc is supposed to try once in a while different versions of applications to detect possible changes in hardware performance.

I don't know how Boinc handles those tasks that run only for a few seconds but are still valid tasks. their amount can vary a lot. Are they also taken in the consideration when APR is calculated? If they are, then that could explain why the calculated flops sometimes favors AVX and sometimes SSE.

---

The runtime differences between AVX and SSE programs are very small.

what is also interesting: the list of the downloaded and also finished Sixtrack tasks also shows tasks without any indication as to AVX or SSE2. What kind of tasks are those?

The runtime differences between AVX and SSE programs are very small.what is also interesting: the list of the downloaded and also finished Sixtrack tasks also shows tasks without any indication as to AVX or SSE2. What kind of tasks are those?

I get those on my X86 host *SixTrack v502.05 windows_intelx86*

But when I check the finished Valid I see this https://lhcathome.cern.ch/lhcathome/workunit.php?wuid=161022530

That would be the sixtrack application for CPUs without any SSE or AVX support. That probably also gets tested once in a while in other hosts as well.

---

Yes I know my one X86 can not run SSE but every Valid task it has is also run by the wingman in either SSE or AVX
https://lhcathome.cern.ch/lhcathome/results.php?hostid=10447575

That would be the sixtrack application for CPUs without any SSE or AVX support. That probably also gets tested once in a while in other hosts as well.

yes, seems to be so.
These tasks then show up with "SixTrack v502.05 windows_intelx86"

I see that some of my sixtrack task quorum partners were running the same application as I did (Win 64 bit AVX), so validation does not require separate platforms to succeed. Mostly the valid tasks do come from different applications. It maybe that computers running a plain sixtrack application are so rare that they usually partner with other applications.

---