Hey,

I'm relatively new to DC, and am trying to optimize my computers. Right now my P4 Mobile 1.4GHz 512MB is out-benchmarking my P4 Xeon 2.4 Ghz HT 4GB computer in floating point, and nearly 3:1 in integer. Is there something I'm doing wrong with my Xeon, or is that just how things are?

Link to computers

Thanks in advance, and sorry if this is a common question. I could not find any answers when searching.

-Eric

Hey,

I'm relatively new to DC, and am trying to optimize my computers. Right now my P4 Mobile 1.4GHz 512MB is out-benchmarking my P4 Xeon 2.4 Ghz HT 4GB computer in floating point, and nearly 3:1 in integer. Is there something I'm doing wrong with my Xeon, or is that just how things are?

-Eric

The Pentium M is quite a processor. Yes, that's just the way things are! Sorry to hear about your Xeon.

---

Hey,

I'm relatively new to DC, and am trying to optimize my computers. Right now my P4 Mobile 1.4GHz 512MB is out-benchmarking my P4 Xeon 2.4 Ghz HT 4GB computer in floating point, and nearly 3:1 in integer. Is there something I'm doing wrong with my Xeon, or is that just how things are?

-Eric

The Pentium M is quite a processor. Yes, that's just the way things are! Sorry to hear about your Xeon.

The p4 mobile isn't the same chip as a pentium M. My guess is the xeon has hyperthreading enabled and the 1.4 doesn't. The benchmarks are per cpu, and since a xeon looks like 2 cpus (even though this version isn't), the benchmarks are lower. If HT is enabled, try limiting your preferences to 1 cpu and see if it affects anything.

---

Eric,

Aside from affecting the credit claims the benchmarks are not a good indicator of actual performance.

In this case, you can see a "slower" processor benchmark higher, yet perform worse in actual use. One of the things that affects this is that the benchmarks for multiple-cpu systems effectively divides the registered score by the number of CPUs. So, there is a 64-processor system on SETI@Home that claims about 0.1 CS per work unit. Solely due to the abysmal benchmark numbers.

Second, the Xeon may take longer to process a work unit than you might expect, but, you will be doing 2 work units in the time the other computer is doing only one. You can rummage my account and look at typical numbers if you like. I have an AMD 3500 single CPU system and two dual Xeons at 3.4 GHz. What you will find is that the Xeons take, for example, 12 hours to do a work unit that the AMD takes 5 hours for (using some Einstein@Home numbers now as I don't have Rosetta@Home running at the moment).

So, in 15 hours, the AMD will have done 3 work units. In the same time the Xeons have done 4 complete work units and are 25% done with 4 more. Effectively the Xeon has done 5 work units in the time the AMD has done 3. So, slower processing time but higher THROUGHPUT.

Disabling HT will improve processing speed at the expense of throughput. Basically the second work unit is done using CPU time that the single threaded processor would have "wasted".

---

Thank you all for your feedback. I have just been a little confused, since when browsing other people's computers, I saw very high benchmarks for Xeon procs, andnot on my own. Perhaps they are doing something special, but I'll settle for what I have.

Again, thanks!

-Eric

Thank you all for your feedback. I have just been a little confused, since when browsing other people's computers, I saw very high benchmarks for Xeon procs, andnot on my own. Perhaps they are doing something special, but I'll settle for what I have.

If you run an "optimized" version of the BOINC client software, one of the main, if not the ONLY effect is to raise the benchmark scores. In some cases to absurd levels. I found a guy that had a G5 that had benchmarks 10x mine. But, he took longer to do the work.

Here, because there is no averaging to correct for inaccurate claims if you have inflated benchmark scores you will get much higher awards.

Again, none of this is improper, imoral, illegal, sinful, or fattening ... but it is a little childish and pointless. The real point is to do the science.

Disclaimer: I do use optimized applications for both SETI@Home and Einstein@Home and in BOTH cases, even though I use the standard BOINC Client SOftware I get a "bonus" in that I get higher awards, in some cases significanlty higher awards because I am almost always the lowest claimant.

Said disclaimer is one of the reasons I think it is pointless to up the benchmarks to make them report what they "should" and to "fix" claims.

---

Also, on multi-CPU XEON boxes (which hyperthreading is, sort of), the benchmarks are run simultaneously. Memory-bandwidth seems to become a problem there. To see if that is your problem, set your preferences to (temporarily) use only one cpu on multi-cpu machines, then rerun the benchmark. If it turns out significantly higher, memory may be constraining it. Why this does not hold true for everybody, dunno. Older XEONs? Worse memory? Worse motherboards? Could all very well be.

---

The p4 mobile isn't the same chip as a pentium M. --snip!--

Check his active computers. He has one 1400Mhz processor and it's a Pentium M.

---

You know, I've had that laptop for a little over 2 years now, and I think the whole time I was under the impression that it was a P4 Mobile, not the M chip.

Thanks for pointing that out :-)

-Eric

Why this does not hold true for everybody, dunno. Older XEONs? Worse memory? Worse motherboards? Could all very well be.

it's a motherboard problem, i've got the "older" generation, and it's much slower than the newer versions even if they used the same processors
also the newer versions allow for faster memory :(

Disclaimer: I do use optimized applications for both SETI@Home and Einstein@Home...

Paul, there's an optimised app for einstein?

Paul, there's an optimised app for einstein?

Since Paul may not pass through again tonight - yes, the project themselves released an Altivec-optimized app for Mac OS X. Cut run times in half.

---

Paul, there's an optimised app for einstein?

Since Paul may not pass through again tonight - yes, the project themselves released an Altivec-optimized app for Mac OS X. Cut run times in half.

oh right, know about that, was wondering if they'd released the source or something and someone had recompiled it for various instruction sets, like in the SETI world

Lee,

Not yet. I am not sure why. But, the Altivec is sure sweet. I am running into the daily quota if they give me certain work unit as the G5 does them in 52 minutes. With 16 (dual so I get 32) a day I am 8+ short if I do only Einstein@Home and I get too many of these. Most off the rest seem to be running in about 3 hours. Though I see one here that took 1:11; so, I may not be out of the woods yet..

The source for Einstein@Home is not going to be released, I forget if it is a library thing, or if it is that part of the source is licenced and cannot be open to the public.

I am sure they would love to have an SSE3 version for the Windows world, but, to this point we have not seen it yet. It could also be they have not been able to figure out how to issue the various compiles to the Intel/AMD CPUs so that the proper version was issued.

We could put up a question in NC and ask.

==== edit

Done! See this

---