SETI MB CUDA for Linux

[sunu]

Well, just play with that AR value to keep both CPU and GPU busy till new WUs come out to refill your cache.

[riofl]

hmmm don't remember if i asked this before, but is this -9 result overflow message anything to worry about. i  have seen it a few times. the wu was validated and credit given though.

Work Unit Info:
...............
WU true angle range is :  0.508292
SETI@Home Informational message -9 result_overflow
NOTE: The number of results detected exceeds the storage space allocated.

Flopcounter: 27434172839395.402344

Spike count:    28
Pulse count:    2
Triplet count:  0
Gaussian count: 0
called boinc_finish

</stderr_txt>
]]>

[sunu]

These result overflows are common. You should pay attention though if you generate a lot of these while your wingmen return "good" results. In this case it could be a hardware problem in your part and these results will be invalid.

[riofl]

These result overflows are common. You should pay attention though if you generate a lot of these while your wingmen return "good" results. In this case it could be a hardware problem in your part and these results will be invalid.

hmm. i got two of these from the 285 since i started cuda back up with the vhar units.  i ran a memory tester i got off the net which showed good on all available memory..  and i am not geting any strange things happening to the machine or the desktop or movies when i push the gtx a little bit. might have been luck-o-the-draw. ill just have to keep an eye on this.

[riofl]

ok my mistake. those were from yesterday when i ran out of units... no way to sort by date i assumed newest were first then i actually read the date   all of the 2 sept ones were fine.

[lordvader]

After a small hiatus I'm back 
... snip ...

@riofl and lordvader about kernel versions
Have you compiled these kernels yourselves or have you got them from elsewhere? Maybe some performance/optimization options you left out? Do you have any nvidia related errors in your syslog when running cuda? Any other observations with these newer kernels?

Hi !

About the kernel versions. I compiled the 2.6.30 kernel using the same config used in Ubunutu's 2.6.28 kernel. I used the same config on the 2.6.27 kernel, which currently gives me equivalent CUDA performance, as well as stable Astropulse results.

I've been using the latest nvidia drivers, and the latest CUDA library (oh ! and my CPU is a Phenom II 955, not overclocked, and I'm running 64bit Kubuntu).

An interesting observation. While crunching CUDA units (using the fast kernels), my GPU heats up quite a bit, as expected. On the slow kernel, it barely goes above idle, so it's definitely being under-utilised.

I've yet to try kernel 2.6.29, and probably won't have time to test it until sometime next week.

Any more info you may need ?

[sunu]

About the kernel versions. I compiled the 2.6.30 kernel using the same config used in Ubunutu's 2.6.28 kernel. I used the same config on the 2.6.27 kernel, which currently gives me equivalent CUDA performance, as well as stable Astropulse results.

So let me get this straight:
2.6.27: custom compiled, fast
2.6.28: stock, fast
2.6.30: custom compiled, slow

All three supposedly with the same pre-compile config. Am I right?

In http://kernel.ubuntu.com/~kernel-ppa/mainline/ there are pre-compiled mainline kernels along with build options from Ubuntu's kernel team ready for install. Have you tried those?

[lordvader]

That's right, same config for all of them.

I've downloaded the precompiled kernels, but won't get a chance to try them till the end of the weekend, at the earliest.

[riofl]

credit granted is the same for a given workunit whether it is processed by cpu or gpu correct?

my thinking is that since the vhar are short, only taking my gpu about 12 min to process what would be the harm in assigning all vhar to gpu to speed up things? unless they would be granted more credit being done by the slower cpu then i will  go back to assigning them to cpu.

[sunu]

credit granted is the same for a given workunit whether it is processed by cpu or gpu correct?

Yes.

my thinking is that since the vhar are short, only taking my gpu about 12 min to process what would be the harm in assigning all vhar to gpu to speed up things?

Some people say that CPU is more efficient for VHARs than GPU but can't see their reasoning.

[pp]

That's right, same config for all of them.

I've downloaded the precompiled kernels, but won't get a chance to try them till the end of the weekend, at the earliest.

Remember that newer kernels usually have some new configurable options so an old config won't apply cleanly.  In the transition from 2.6.28 to 2.6.30 these comes to mind:

Code: [Select]

Processor type and features  --->
    [*] Supported processor vendors  --->
        [*]   Support Intel processors
        [ ]   Support Cyrix processors
        [ ]   Support AMD processors
        [ ]   Support Centaur processors
        [ ]   Support Transmeta processors
        [ ]   Support UMC processors

Really bad things seems to be possible if you don't enable this option and configure it correctly for your cpu. I can't say that this option is the cause of your problem (there are several new ones) but I always check every new option during kernel upgrades and computation times are the same for me regardless of kernel version.

[pp]

credit granted is the same for a given workunit whether it is processed by cpu or gpu correct?

Yes.

As noted by people over at S@H forums, the GPU seems to claim credit that is about 30% higher than what the CPU does. If your wingman use the CPU you will be granted the lower value but if your wingman use GPU like you, you will receive the higher value it seems.

[sunu]

As noted by people over at S@H forums, the GPU seems to claim credit that is about 30% higher than what the CPU does. If your wingman use the CPU you will be granted the lower value but if your wingman use GPU like you, you will receive the higher value it seems.

The cuda app usually overclaims (other times underclaims) because it can't calculate precisely the flops needed for a certain workunit.

[sunu]

Shameless plug:

I've managed to climb to the #3 spot in the Top hosts list. This is probably the highest I'll ever be so I'll savour the moment. 

PDF attached for future proof.

[attachment deleted by admin]

[riofl]

Shameless plug:

I've managed to climb to the #3 spot in the Top hosts list. This is probably the highest I'll ever be so I'll savour the moment. 

PDF attached for future proof.

congrats! when i get my 2nd 285 to replace this tesla maybe ill be able to reach around 15k.

do you process ap too or just mb? i blocked ap and am only doing mb.

either i'm doing something wrong to only break into the 10ks and now dropped to 9.9 since the weekend or something.. i see so many people with considerably higher rac with basically the same equipment.. unless its because i keep this machine so busy being a workstation... my load averages are consistantly betw 5.0 and 8.5