Hi all,

Just a quick question related to CPU cache. Recently I have searching the forums and trying to learn how to get my computers to run a little faster. I figure the more my computers can do the better. After looking at many threads and comparing my computers to other similar computers, I have found my computers are probably not running as efficiently as they could. I guess I have several questions related to this:

1. I read that CPU cache is pretty important for running SETI but since the three computers in question are running CUDA WUs and wont use the CPU, how important is this?

2. The following computers are the ones I wonder about:

http://setiathome.berkeley.edu/show_host_detail.php?hostid=4941154

in cpuz the cache is reported as 1024 KBytes but 488.28 KB for SETI

-----

http://setiathome.berkeley.edu/show_host_detail.php?hostid=4911647

in cpuz the cache is reported as 2x6144 KBytes but 244.14 KB for SETI

-----

http://setiathome.berkeley.edu/show_host_detail.php?hostid=4907209

in cpuz the cache is reported as 2x4096 KBytes but 244.14 KB for SETI


Why is SETI not reporting the cache sizes as reflected in cpuz? Does this impact the amount of work these computers are doing even though they are doing CUDA WUs? All three of these computers are using 32bit Windows OS, does that make a difference?


I have seen other computers with similar configurations using a Q6600 CPUs and see they are producing RACS of 10k+ while mine (host 4907209) is around 1,780. I figure even if those hosts are overclocking and do other tweaks, is it really possible they are getting 4x performance? Or is there something I can do to make these systems perform better?

I started am using the optimized applications recently thinking this may help increase the performance a bit.

If there is any information I need to provide, please let me know. I am using the the 6.6.36 BOINC client on all these systems.

As a side note, I saw in the readme about CUDA that a host using CUDA would not run both CPU and CUDA WUs. My one host (4941154) does. Don't know why, just thought I'd mention it.

Thanks.

---

The CPU Cache numbers reported by BOINC are usually wrong, ignore them. Trust what CPUz is telling you. I'm guessing that getting the CPU cache reporting correctly is not their highest priority as it doesn't really affect any operational things, it's info only. As it's allways wrong it should probably be turned off to avoid confusion. (no info is better than wrong info)

The optimised applications will be the biggest boost you can make for your system, especially when there are Astropulse units available again. You wont see an instant boost in RAC, it takes a couple of months to build up to a steady level, especially with the long turn around on AP.

Overclocking will also help, but not as much as the Opt App.

My Q6600 was getting 8,000 RAC stock standard, but thats dropped down to 6,000 due to the lack of AP workunits. So 10,000 would be possible with some significant overclocking.

Loooking at your Q6600 rig it appears to be processing only CUDA work units. Thats all it's returned anyway. So the CPU has 3 cores sitting idle, that will account for the low RAC. Get the optimised apps running on the spare cores and your RAC will start climbing.

The E2200 CPU is running both CUDA and CPU. But with only a 8400 card it may be better with CPU instead?

Ian

Ian,

Thanks for the information. I guess from what you are saying that things are OK then. My machine with the Q6600 in it is running other BOINC projects as it never downloads any SETI work other than the CUDA WUs. Same goes for the Q9550 machine. Both these machines have the optimized apps installed. From what I originally understood, the machine would only run CUDA or CPU WUs but not both (as indicated in Q 5 here http://setiathome.berkeley.edu/cuda_faq.php) so I'm not too sure what's going on there. Maybe there is a setting I may have missed. Anyway, I was thinking they would eventually start getting normal CPU units but nothing yet but as long as they are doing something useful for the project, I'm OK with that. I was just thinking maybe there was a way to get these machines to do more work but it seems all is well. Incidentally, that Q6600 machine only does 3700 RAC across all projects but this might not mean much as all the projects calculate RAC and credits differently.

As for the E2200 with the 8400 card, I just threw that in there so it could maybe do a little more work on the GPU. That is a backup server and I don't think I'm going to get a better graphics card for a machine I rarely log into. I will probably change the configuration of that machine (E2200) to the SETI project only though. I was thinking about doing that for the Q6600 but changed my mind when I saw it wasn't downloading any CPU SETI WUs. At that point, I attached it to other projects so that the CPU wouldn't be idle. If there is a way to change that so that it will crunch both CUDA and CPU WUs for SETI, I would make that a dedicated SETI box as well.

I have read a lot of information about the AP WUs and look forward to when they come online again. Nice to have steady WUs like those as well.

When time permits, I will keep up to date with possible ways I can change the configurations so that I can dedicate more machines to SETI only.

Again, thanks for your time and the good information.

---

...
2. The following computers are the ones I wonder about:

http://setiathome.berkeley.edu/show_host_detail.php?hostid=4941154

in cpuz the cache is reported as 1024 KBytes but 488.28 KB for SETI

-----

http://setiathome.berkeley.edu/show_host_detail.php?hostid=4911647

in cpuz the cache is reported as 2x6144 KBytes but 244.14 KB for SETI

-----

http://setiathome.berkeley.edu/show_host_detail.php?hostid=4907209

in cpuz the cache is reported as 2x4096 KBytes but 244.14 KB for SETI


Why is SETI not reporting the cache sizes as reflected in cpuz? Does this impact the amount of work these computers are doing even though they are doing CUDA WUs? All three of these computers are using 32bit Windows OS, does that make a difference?
...

For Windows, BOINC does not even attempt to measure cache size. Instead it uses 1000000 divided by the number of CPU cores. Then when it is reported the value is divided by 1024. All single CPU systems show as 976.56 KB, duals as 488.28, quads as 244.14...

Just like the BOINC benchmarks, the number doesn't have any effect on what a host can do. All those values were merely intended to help BOINC manage work, etc.

                                                                 Joe

Joe,

Thanks for that explanation about the way BOINC resolves the cache value on Windows machines. I now know that everything is good and well thanks to the good information provided here.

Again, thanks all for the information and assistance.

---

Ian,

As for the E2200 with the 8400 card, I just threw that in there so it could maybe do a little more work on the GPU. That is a backup server and I don't think I'm going to get a better graphics card for a machine I rarely log into. I will probably change the configuration of that machine (E2200) to the SETI project only though. I was thinking about doing that for the Q6600 but changed my mind when I saw it wasn't downloading any CPU SETI WUs. At that point, I attached it to other projects so that the CPU wouldn't be idle. If there is a way to change that so that it will crunch both CUDA and CPU WUs for SETI, I would make that a dedicated SETI box as well.

As to the Running CPU and GPU, which originally was a Boinc issue coupled with Server Configuration, which has been resolved. As you stated you are running Boinc 6.6.36 you should be able to run both. The easiest way is obtain Lunatics Unified Installer x32/x64 bit at this URL http://lunatics.kwsn.net/index.php?module=Downloads;catd=9 Windows Downloads area. As always with any optimized Application you need to monitor for when updates are provided.

As for the 8400 that is almost "effectively a CPU runtime" with an optimized application.

---

Please consider a Donation to the Seti Project.

My 8500GT runs about one and a half workunits in the time it takes my CPU to finish one optimized so I'm a little faster than the 8400. Even at only one workunit it is still like adding an extra core to his machine.

---

PROUD MEMBER OF Team Starfire World BOINC

Pappa,

The link you provided was exactly what I used to install the optimized app on all the machines. I used the "Win32 Lunatics' Unified Installer FINAL BETA" and it did all the calculations and configurations for the app_info.xml file that I previously had to do manually. Thats a nice nice install tool they made.

As for running both CPU and CUDA SETI WUs on the Q6600 machine, it never seems to download the CPU WUs as the E2200 machine does. I will be testing out some alternate configurations tonight to see if I can get it to work.

perryjay,

As you and Pappa have mentioned, that 8400 card doesn't seem to do to bad. It more or less seems to be like an extra core as you mentioned which is nice.


Once I can figure out how to get the Q6600 to crunch both CUDA and SETI WUs, I will probably get the Q9550 to the same config as well but the Q9550 lost a HDD last night so I'm gonna wait for the replacement before I do the final reconfiguration on it.

What started out as a cache question seems to be spawning into more of a BOINC Manager configuration discussion. I hope going a little off topic here is OK. Again, I really appreciate all the information and insight from everyone. Thanks.

---