![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
BOINC question
- Thread starter sl1982
- Start date
metallicafan
[H]ard|DCer of the Month - May 2010
- Joined
- Mar 30, 2005
- Messages
- 2,205
I have never tried the Linux client myself but I know nomad mentioned one time that his Suse box was faster than his Windows box. I'm sure he will chime in at some point.
sirmonkey1985
[H]ard|DCer of the Month - July 2010
- Joined
- Sep 13, 2008
- Messages
- 22,414
might be a slight difference.. but for me atleast its not worth switching to ubuntu.. but then again i was a die hard linux person a few years ago and cant stand using ubuntu.. but ive gone back to winblows because the gaming support is still lacking badly in linux.. maybe another year or so it will get better and i might switch back..
Heh, do some of the video encoding I'm doing right now. Rough figures put the encoding time at around 54 hours on the rig in my sig. Talk about using up cycles.
As for WCG, I don't see any performance benefit by running Linux instead of Windows. I've actually noticed a weird anomaly. My [email protected] running openSUSE 10.3 has about the same total runtime and number of results returned as the [email protected] running XP and yet the XP machine has about 200k more points according to the WCG stats. It doesn't make much sense but I deal with it. Both machines don't do much other than crunch WCG so they are comparable.
In general, any 64 bit OS is going to do better provided the individual science apps are coded for 64 bit. Many of them are.
For WCG there's a couple of things at play here. In general, Windows scores better "overall" than Linux on WCG by a small margin from a lot of reading I did. The exception there (and I've confirmed by directly comparing my results with the same system on both Win32 and Linux x64) is HCC. It scores closer to "claimed credit" on every unit compared to Windows. It also runs about 20% faster on the same type of WU for HCC and HPF2. HPF2 units score lower for me than the same system on Win32 BUT it runs the units faster so the overall result there is about a 10-15% increase in PPD just by completing more. Rice is a fixed 10+ hour run time and scores slightly lower. FAAH runs a little bit faster but scores slightly lower. Almost a wash there. I didn't run any CEP units on Linux due to them having many issues with the Linux units.
I have 1 quad on x64 openSUSE and 2 quads still on Win32. I based the decision to go with x64 Linux on a couple of things. One was that I just wanted to learn and try out Linux and at the same time go 64 bit without paying for Vista or XP x64. The other thing that swayed me is raw numbers. If you're like me at all (and yes.. I like numbers) take a look at this awesome benchmark spreadsheet courtesy of "D_A" over at XS. He's done a great job compiling a large list of results from just about every OS imaginable and various Boinc versions. Here is the last page of the thread. I threw my 3 quads in the mix there as well.
Two things became apparent to me looking over the numbers. 1. 64 bit OS's scored higher and Linux seemed to have a bit of an advantage as noted in my comments above. Overall, the same system "scores" better on x64 Linux than Win32.
For WCG there's a couple of things at play here. In general, Windows scores better "overall" than Linux on WCG by a small margin from a lot of reading I did. The exception there (and I've confirmed by directly comparing my results with the same system on both Win32 and Linux x64) is HCC. It scores closer to "claimed credit" on every unit compared to Windows. It also runs about 20% faster on the same type of WU for HCC and HPF2. HPF2 units score lower for me than the same system on Win32 BUT it runs the units faster so the overall result there is about a 10-15% increase in PPD just by completing more. Rice is a fixed 10+ hour run time and scores slightly lower. FAAH runs a little bit faster but scores slightly lower. Almost a wash there. I didn't run any CEP units on Linux due to them having many issues with the Linux units.
I have 1 quad on x64 openSUSE and 2 quads still on Win32. I based the decision to go with x64 Linux on a couple of things. One was that I just wanted to learn and try out Linux and at the same time go 64 bit without paying for Vista or XP x64. The other thing that swayed me is raw numbers. If you're like me at all (and yes.. I like numbers) take a look at this awesome benchmark spreadsheet courtesy of "D_A" over at XS. He's done a great job compiling a large list of results from just about every OS imaginable and various Boinc versions. Here is the last page of the thread. I threw my 3 quads in the mix there as well.
Two things became apparent to me looking over the numbers. 1. 64 bit OS's scored higher and Linux seemed to have a bit of an advantage as noted in my comments above. Overall, the same system "scores" better on x64 Linux than Win32.
Right here dude..
Unfortunately the site appears to be down. Whether it's permanent or temporary I don't know as Amanheis quit development on this several years ago and has been MIA since. Still works great with Boinc though as the Boinc developers have taken care to keep Boinc functional with it. I'm using it with 6.2.19 win client, 6.2.15 linux client and 6.4.2 win client for GPU. Someday maybe all the data BV has will be incorporated into Boinc Manager.
If the site is down permanently, I think I know where you can find the last available version..
Based on what I've read (and others can confirm/deny this for me), is that Claimed Credit is what you would get if your machine was EXACTLY like their benchmark. Granted Credit is based on a bunch of different variables (CPU speed, amount of available RAM, time to complete and others) so the algorithm to actually calculate Granted Credit is pretty complicated (and frankly convoluted). It is NOT like F@H where each WU is the same number of points regardless of the machine that it was completed on.
Based on what I've read (and others can confirm/deny this for me), is that Claimed Credit is what you would get if your machine was EXACTLY like their benchmark. Granted Credit is based on a bunch of different variables (CPU speed, amount of available RAM, time to complete and others) so the algorithm to actually calculate Granted Credit is pretty complicated (and frankly convoluted). It is NOT like F@H where each WU is the same number of points regardless of the machine that it was completed on.
Pretty much spot on except "Granted Credit" is based on a quorum that tries to equalize points granted for amount (total CPU time) of work done. This is only on units with a quorum >1. So if you run a unit with a 15 of 19 quorum and you have a fast CPU but get paired up with all slow clients, you're gonna get boned. If you're lucky enough to get paired up with similar speed/horsepower clients then you win.
Right here dude..
Unfortunately the site appears to be down. Whether it's permanent or temporary I don't know as Amanheis quit development on this several years ago and has been MIA since. Still works great with Boinc though as the Boinc developers have taken care to keep Boinc functional with it. I'm using it with 6.2.19 win client, 6.2.15 linux client and 6.4.2 win client for GPU. Someday maybe all the data BV has will be incorporated into Boinc Manager.
If the site is down permanently, I think I know where you can find the last available version..
Awesome!
We clearly need a BOINC central with things like this, account managers, project URL's for the BOINC client (which is NOT always the same as the home page), and all the little crap we've already stumbled over along the way.
I could add this info into the second post of the "software thread". We'll figure something out.