I just moved from the oldest optimized client on http://www.marisan.nl/seti/ to the newest(w-r8.1) for my Athlon 64+ 3000. Now it takes about 20-30 minutes longer to finish a work unit. Should i try 7? or just go back to the old one since it seemed to work a lot better....or do i need to wait til i start working on work units that i download with this client and then see how it does?

---

Join the team, SETI.USA We are growing and could use your help to overcome SETI.Germany...www.setiusa.net

To answer that question you really need to run each client with the reference work unit. It's impossible to get an accurate answer just comparing a few WUs as all WUs are different and will take a different amount of time to process. Running the reference unit with each client will give you an accurate result as to which is quickest.

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

but when im getting times that are about 20 minutes longer than anything i've done in the last month, i'm pretty sure its the optimized client. I'm switching to 7 for the next WU i do, and will see if it makes any difference. If it still doesnt do better than the oldest one i will go back to that.

---

Join the team, SETI.USA We are growing and could use your help to overcome SETI.Germany...www.setiusa.net

The newer optimized apps are compiled using IPP and the older one uses FFT. There have been reports from some users that the newer ones don't run as well on AMD cpu's. But it is still your choice as to what to use.

---

98SE XP2500+ @ 2.1 GHz Boinc v5.8.8

And God said"Let there be light."But then the program crashed because he was trying to access the 'light' property of a NULL universe pointer.

The newer optimized apps are compiled using IPP and the older one uses FFT. There have been reports from some users that the newer ones don't run as well on AMD cpu's. But it is still your choice as to what to use.

Are those newly compiled clients clobbered by Intel's "don't use competitors CPU features" dongled code (in IPP)?

Regards,
Martin

---

See new freedom: Mageia Linux
Take a look for yourself: Linux Format
The Future is what We all make IT (GPLv3)

The newer optimized apps are compiled using IPP and the older one uses FFT. There have been reports from some users that the newer ones don't run as well on AMD cpu's. But it is still your choice as to what to use.

Are those newly compiled clients clobbered by Intel's "don't use competitors CPU features" dongled code (in IPP)?

Regards,
Martin

It's the ICC compiler that's nobled, not the IPP fft libs.

I _think_ so, but they're still fast. They _may_ not actually be using the SSE features, but the benefits of the IPP fft library still make them way faster than the stock client. I haven't benchmarked them on AMD processors but all three (the fftw and 2 IPP) clients seem in the same ballpark.

I'll say it again - someone NEEDS to acurately benchmark all three clients against the reference WU on an AMD processor, otherwise we're just hypothesizing at which appears to be faster. We've done this for linux, it's just the Windows guys who are slow on the uptake ;)

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

I'll say it again - someone NEEDS to acurately benchmark all three clients against the reference WU on an AMD processor, otherwise we're just hypothesizing at which appears to be faster. We've done this for linux, it's just the Windows guys who are slow on the uptake ;)

I'll take up this challenge. :)
As soon as I have run the reference-unit on my system and know how to read the cpu-time from that I will invite others in a new thread to send results to me and I will make a listing of those on my website.

Greetings,
Speedy67

---

I'll say it again - someone NEEDS to acurately benchmark all three clients against the reference WU on an AMD processor, otherwise we're just hypothesizing at which appears to be faster. We've done this for linux, it's just the Windows guys who are slow on the uptake ;)

I'll take up this challenge. :)
As soon as I have run the reference-unit on my system and know how to read the cpu-time from that I will invite others in a new thread to send results to me and I will make a listing of those on my website.

Greetings,
Speedy67

Great!!

Details on how to benchtest against the reference WU together with where to get the timings are included on my site towards the bottom of this page:

http://www.pperry.f2s.com/boinc-compile-seti.htm

The instructions are for linux, but the process is not really any different for Windows. Place the optimized seti client and the reference WU in a separate directory and run the seti client from the command line. The time taken to process the reference work unit (in seconds) is stored in the init_data.xml file under the wu_cpu_time field.

One tip, on linux at least, if you stop the client half way through the timing data gets reset so you need to do the complete benchtest run in one go. You'll need to test all three optimized clients separately under identical conditions and ideally the machine should not be used for anything else during the test.

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

I'll take up this challenge. :)
As soon as I have run the reference-unit on my system and know how to read the cpu-time from that I will invite others in a new thread to send results to me and I will make a listing of those on my website.

Great!!

Details on how to benchtest against the reference WU together with where to get the timings are included on my site towards the bottom of this page:

http://www.pperry.f2s.com/boinc-compile-seti.htm

Google showed me the way already. ;)
I will try it out tomorrow after work, and make a new thread inviting people to send me their init_data.xml files. Is that file created after the wu has finished? I have run the reference unit for a few minutes and aborted it (going to shut down this computer in a minute as it is not running 24/7) but it did not make an init_data.xml file.

One tip, on linux at least, if you stop the client half way through the timing data gets reset

And the init_data.xml is deleted at that time? (or not created?)

Greetings,
Speedy67

---

I assume init_data.xml is created once the run is finished. It's certainly present in the directory at the end of the benchmark run :)

When I stopped the client during the benchmark run and restarted it, the final timing data appeared to be only from the point I restarted, not the complete duration.

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

If you are going to go to the trouble to run the "reference_work_unit.sah" you may want to take a look at the results that you get for differences against the "reference_result_unit.sah.

From what I have heard and read at (www.pperry.f2s.com/boinc-compile-seti.htm) the optimized client should produce results that are within validation limits set up by SETI.

One way is to open both up side by side in notepad and compare back and forth. Another way is to get a program like (ExamDiff). You load the "reference_result_unit.sah" and the "result.sah" from your test and it will display the parts that are different.

It will have parts that vary. I think even if you run the standard client on different systems you will get things that vary. You just need to be sure that the results are within the limits.

We have a good track record with the optimized clients validating so we shouldn't have a problem. I just thought you may want to explore a little more.

---

I found that YAOSCW-W-r7 works best for me. Slightly better than the older one, but a lot better than the newest(8.1) for me. Just thought i'd update anyone who wondered.

---

Join the team, SETI.USA We are growing and could use your help to overcome SETI.Germany...www.setiusa.net

If you are going to go to the trouble to run the "reference_work_unit.sah" you may want to take a look at the results that you get for differences against the "reference_result_unit.sah.

From what I have heard and read at (www.pperry.f2s.com/boinc-compile-seti.htm) the optimized client should produce results that are within validation limits set up by SETI.

One way is to open both up side by side in notepad and compare back and forth. Another way is to get a program like (ExamDiff). You load the "reference_result_unit.sah" and the "result.sah" from your test and it will display the parts that are different.

It will have parts that vary. I think even if you run the standard client on different systems you will get things that vary. You just need to be sure that the results are within the limits.

We have a good track record with the optimized clients validating so we shouldn't have a problem. I just thought you may want to explore a little more.

Yes, indeed.

However, as you note in this case we have a excellent track record with these clients and we know results validate fine.

This methodology is however useful when developing new clients to determine if results are within validation limits before unleashing them on real live WUs.

Ned

---

*** My Guide to Compiling Optimised BOINC and SETI Clients ***
*** Download Optimised BOINC and SETI Clients for Linux Here ***

Just for info:

On a 1 GHz Pentium III-M with 512 KB level 2 cache YAOSCW-K-r7 is the fastest.
On a 1.8 GHz Pentium 4 Willamette with 256 KB level 2 cache YAOSCW-N-r7 is the fastest.
On a 1.4 GHz Athlon Thunderbird (no XP!) with 256 KB level 2 cache seti-plainvanilla is the fastest.

---

Aloha, Uli

I started the new thread I mentioned earlier, about trying to make a complete listing of cpu times for the reference unit on different cpu's.

Thanks Ned for the good idea, and thanks in advance to this community that is of course going to help realize it.. :)

Greetings,
Speedy67

---