Since switching to BOINC/SETI I noticed that three of my crunchers were giving much lower benchmark results (less than 70%) than those of other comparable rigs . they are all ABIT boards with Intel P965 chipsets .(I’m not sure whether the problem is confined to ABIT , the Chipset or if its generic.)
POST showed everything OK , overclocking in one case had identical values , memory and CPU to a much faster rig (NVIDIA 650 ). It turns out that if you go to far with your overclocking of the memory , a simple cmos reset is not enough. In all three cases the POST showed dual channel memory active , but removing one of the sticks of memory made no difference to the benchmarks . in all three cases removing the battery - shorting the battery connectors and cmos connectors for a couple of minutes solved the problem .
All benchmarks now give broadly comparable results and removing a stick of memory returns the benchmark results to former levels, and I’ve regained enough crunching power to equal another good C2D rig 
That’s interesting. I have one machine that’s performing below expectations. Unfortunately: 1) it’s my newest, so the average hasn’t settled down yet 2) it’s the only one running Linux - SUSE 10.2 and 3) SETI@Home has been an absolute mess for at least a week. Still, this thing is crunching half the WUs a Dell 2950 with identical CPUs running Vista produces. If this continues I’m going to be forced to do something drastic.
Is there any known issue with SUSE 10.2?
[QUOTE=Andy Williams;406118]That’s interesting. I have one machine that’s performing below expectations. Unfortunately: 1) it’s my newest, so the average hasn’t settled down yet 2) it’s the only one running Linux - SUSE 10.2 and 3) SETI@Home has been an absolute mess for at least a week. Still, this thing is crunching half the WUs a Dell 2950 with identical CPUs running Vista produces. If this continues I’m going to be forced to do something drastic.
Is there any known issue with SUSE 10.2?[/QUOTE]
Not that I know of , I use SUSE 10.3 on one of my machines , and I use the Linux 64 bit BOINC and optimized SETI clients on all.
Compare the benchmarks , they should be similar , but it could just be the very short WUs being dished out at the moment, I’ve got a C2D pushing them out every 17 minutes for 16 points apiece and a similar performance machine doing 1 hr 42 min WUs for 63 points each .
some good advice - the benchmarks are reasonably comparable now between windows and linux, or at least so the mails from the boinc group tell me. I can’t offer much more advice as my rigs are mainly OSX or Gentoo, with the odd occasional Microsoft crunch.
I find that BoincStats rac tends to pick up quicker than the official one, see if you can check the rigs performance on there :shrug:
DT.
[QUOTE=DoubleTop;406124]
I find that BoincStats rac tends to pick up quicker than the official one, see if you can check the rigs performance on there :shrug:
DT.[/QUOTE]
I find the opposite as the BOINC stats RAC is based over the last 60 days and the BOINC is based on I don’t know what. certainly it took about 60 days for my BOINC stats RAC on Riesel Sieve to reach the level of my actually RAC.
Is there something else on the SUSE machine eating the CPU time up?
The only thing taking CPU time other than S@H according to GNOME System Monitor is beagled-helper. It is taking a lot though. It seems to be maxing one whole core. I’m not sure if that is just because GNOME System Monitor is running. The machine is used for nothing else. It’s a dedicated cruncher. I wanted to get away from paying for another copy of Windows every time I set up a new cruncher. Also to get away from the M$ license hassles/restrictions, which have become intolerable.
The really weird thing though is the number of S@H instances that are running. According to my way of thinking there should be eight - two quad-core Xeons. When I first set up the machine there were sixteen. Eight running and eight sleeping. Now there are 32, 8 running and 24 sleeping. But all are consuming 35.3 to 36.4 MiB of memory. This can’t be good.
I’m about to put another one into motion with the same motherboard - ASUS DSBV-DX, slightly faster dual Xeon quad cores. I’ll definitely put SUSE 10.3 on that one.
The average on the existing one is heading north again, and it has only been running two weeks. It’s about 1400 right now. In contrast, my Dell 2950 with identical dual Xeon 5320s approaches 3000 regularly.
None of my clients are optimized, by the way. And none of them are running 64-bit OS’s. I was under the impression that neither factor affected throughput much. Correct me if I’m wrong.
Now that you mention it Beagled helper caused me problems when I was folding on Suse 10.2 . I just used to kill it and everything was OK, the problem went away with 10.3 .
As for 64 bit , the BOINC part makes no difference , but you need it to run the 64 bit clients which are a lot faster - 10 to 25% from what I’ve read , but I wouldn’t swear on that being gospel . the optimised is just an attempt to guarantee compatibility for the new 64 bit stuff
edit : my Fastest (single) Quad has a current RAC of 3570 so there might be something in it .
I rebooted the machine (for the second time ever, I think), killed beagled, and restarted S@H. Seems OK. It still shows 32 instances. I’ll keep an eye on it for a couple days to see whether there’s any dramatic difference. I think eventually I’ll finish the WUs on the machine and install SUSE 10.3 - 64 bit.
Are the 32 instances different times and dates?
Also you can set the number of cores to be used in the preferences overide file. Might fancy adding dont leave applications in memory if it’s using to much RAM by using 32 instances.
[QUOTE=Andy Williams;406188]
None of my clients are optimized, by the way. And none of them are running 64-bit OS’s. I was under the impression that neither factor affected throughput much. Correct me if I’m wrong.[/QUOTE]
I’ve been looking through my results for an illustration of 64 bit Linux vs. 32 bit Windows (read 32 bit client- Windows is easier to differentiate ).
This may show it a bit - Enthusiast - probably not using an optimised client - same processor - Kentsfield Q6600 - OK mine is overclocked (3.25Ghz) will explain some of it - but nearly half the time 
This is closer,This time a major player probably also using optimised clients , again same processor but this time the other is overclocked higher than mine - 3.6 Ghz at a guess ( check the benchmark figures ) but I still beat it for time by nearly 10%.