[mikkyo]

Wait you haven't tried 4.11 yet?
It has been up for a while

[Lateralus]

Results are not as high.

2004-10-15 19:51:31 [---] Benchmark results:
2004-10-15 19:51:31 [---] Number of CPUs: 1
2004-10-15 19:51:31 [---] 932 double precision MIPS (Whetstone) per CPU
2004-10-15 19:51:31 [---] 1735 integer MIPS (Dhrystone) per CPU
2004-10-15 19:51:31 [---] Finished CPU benchmarks

[wiraqocha]

1.25 MDD G4, 1.28G ram
4.09 client:
Benchmark results:
Number of CPUs: 1
589 double precision MIPS (Whetstone) per CPU
1816 integer MIPS (Dhrystone) per CPU

4.11 client:
Benchmark results:
Number of CPUs: 1
871 double precision MIPS (Whetstone) per CPU
2051 integer MIPS (Dhrystone) per CPU

4.13 client:
Benchmark results:
Number of CPUs: 1
839 double precision MIPS (Whetstone) per CPU
1975 integer MIPS (Dhrystone) per CPU

[Lateralus]

1.33GHz PowerPC 7447A G4/768MBs/10.3.6 using Super-Optimized G4 4.11 + SETI Worker:

2004-10-24 21:59:08 [---] Benchmark results:
2004-10-24 21:59:08 [---] Number of CPUs: 1
2004-10-24 21:59:08 [---] 1830 double precision MIPS (Whetstone) per CPU
2004-10-24 21:59:08 [---] 2071 integer MIPS (Dhrystone) per CPU
2004-10-24 21:59:08 [---] Finished CPU benchmarks

[Lateralus]

1.47GHz PowerPC 7455B G4/1.5GBs/10.3.6 using G4 4.13:

2004-11-21 22:25:38 [---] Benchmark results:
2004-11-21 22:25:38 [---] Number of CPUs: 1
2004-11-21 22:25:38 [---] 979 double precision MIPS (Whetstone) per CPU
2004-11-21 22:25:38 [---] 2295 integer MIPS (Dhrystone) per CPU
2004-11-21 22:25:38 [---] Finished CPU benchmarks

Why the drop in performance with every client since 4.06?

[Lateralus]

Originally posted by Lateralus:
Why the drop in performance with every client since 4.06?

Anybody?

[Shaktai]

Originally posted by Lateralus:
Anybody?

The BOINC code is still very much in flux. It is changing on a regular basis, and this includes not only the BOINC program, but the project applications (workers) as well.

Expect to see frequent changes in performance as the code continues to evolve and mature. Everyone is impacted equally by this, so there is no disadvantage to anyone. I believe version 4.56 is the current version in alpha testing, so you can see by the version numbers they have been very busy working on the code. If you monitor the BOINC-SETI forums, you can learn some of what is happening. The future holds faster optimized clients for all platforms, and even a GUI interface of sorts.

[Shaktai]

Anyone have benchmarks with the optimized client for a G5 1.8ghz single? I just ordered one (my early Christmas present to myself, but couldn't afford the dual). It was a decision, trying to decide between the G5 single PowerMac or the G5 iMac. Not much difference performance wise, but decided on the PowerMac for "expandibility" reasons, and because I didn't need another monitor.

If you don't have the latest single, even the older single, a G5 iMac or or 1 cpu of a dual would be close enough for guestimation purposes. Just curious what to expect.

[mikkyo]

Originally posted by Lateralus:
[b[Why the drop in performance with every client since 4.06?[/b]

Originally posted by Lateralus:
Anybody?

I'm pretty sure the benchmarking code itself changed at some point to better reflect the performance of the machine vs the performance of the seti worker.

[mikkyo]

Originally posted by Shaktai:
Anyone have benchmarks with the optimized client for a G5 1.8ghz single? I just ordered one (my early Christmas present to myself, but couldn't afford the dual). It was a decision, trying to decide between the G5 single PowerMac or the G5 iMac. Not much difference performance wise, but decided on the PowerMac for "expandibility" reasons, and because I didn't need another monitor.

If you don't have the latest single, even the older single, a G5 iMac or or 1 cpu of a dual would be close enough for guestimation purposes. Just curious what to expect.

See the first post of the thread.

Here is a current one:
G5 1.8GHz 1Gb Ram OS X 10.3.6
2004-11-29 11:49:06 [---] Benchmark results:
2004-11-29 11:49:06 [---] Number of CPUs: 1
2004-11-29 11:49:06 [---] 1626 double precision MIPS (Whetstone) per CPU
2004-11-29 11:49:06 [---] 3263 integer MIPS (Dhrystone) per CPU

[Shaktai]

Originally posted by mikkyo:
See the first post of the thread.

Here is a current one:
G5 1.8GHz 1Gb Ram OS X 10.3.6
2004-11-29 11:49:06 [---] Benchmark results:
2004-11-29 11:49:06 [---] Number of CPUs: 1
2004-11-29 11:49:06 [---] 1626 double precision MIPS (Whetstone) per CPU
2004-11-29 11:49:06 [---] 3263 integer MIPS (Dhrystone) per CPU

Thanks mikkyo. That first post was soooo long ago.

The current client sure improves the floating point, thanks to your hard work. Of course the new Mac is intended to be my "photography" machine. (It just isn't the same on windows). Naturally it will serve second duty as a cruncher. I don't do any high end work, so the 1.8 single will be more then adequate, and I can always update the graphic card and expand storage and memory if I need to, since it is a powermac. It will be a huge improvement over my iBook or my former PowerMac 800. Extra points per day for DC will just be icing on the cake.

FYI: Einstein is aggresively working to develop native apps (workers) for the different platforms. Their "faster" Windows and Linux clients are already out, and they are testing a Mac optimized application in-house now, claiming up to 2.5 times faster (sounds like altivec), but it hasn't yet been released to the alpha team yet. Still another month before they go public it looks like. So far, it is smooth, but their servers haven't really been tested yet. Current work units take 30 minutes on a fast box. They are talking larger work units on the order of SETI size or 6-12 hours. Team MacNN is #4 with 5 members, out of 103 teams and 741 participants. Of course because it is closed alpha, a lot of the big teams aren't involved yet. This project draws a lot of big "university farms" though. All those physics departments now have a project they can sink their teeth into.

[Shaktai]

BOINC Bencharks with Mikkyo's optimized G5 client.
iMac G5 1.6 ghz
768mb RAM

2004-12-18 13:53:10 [---] Benchmark results:
2004-12-18 13:53:10 [---] Number of CPUs: 1
2004-12-18 13:53:10 [---] 1465 double precision MIPS (Whetstone) per CPU
2004-12-18 13:53:10 [---] 2914 integer MIPS (Dhrystone) per CPU
2004-12-18 13:53:10 [---] Finished CPU benchmarks

Not a super speed demon, but I can live with that.

[fillmeup]

I don't know if this has been discussed yet, but after watching my Seti stats I have found another positive attribute of the optimized client. Not only does it have some speed increase, but it claims a HIGHER credit amount.

I used both the standard stock client and mikkyo's optimized client with javalizards worker for weeks at a time. After comparing the results, I can clearly see that the optimized client has a higher claimed credit for work units. This has translated to more credit per CPU time overall.

Thanks again

[virex]

the only reason to optimize your benchmarks is to get higher claimed credit. credit is a factor of compute time, and benchmark. higher benchmark=higher claimed credit

optimizing the seti worker however, will help you get it done faster. which would technically make you claim less credit than using a non-optimized seti worker, but the time savings outways the credit difference. and...the most important part...it allows you to get more units done. which means more units for the seti team to look at. which is great for the science, which is the whole reason for this right?

[Lateralus]

Dual 1.80GHz PowerPC 7447A G4s/1.5GBs/10.4 (8A420) using G4 4.19 GCC 3.5:

2005-03-30 21:03:28 [---] Benchmark results:
2005-03-30 21:03:28 [---] Number of CPUs: 2
2005-03-30 21:03:28 [---] 1339 double precision MIPS (Whetstone) per CPU
2005-03-30 21:03:28 [---] 5341 integer MIPS (Dhrystone) per CPU
2005-03-30 21:03:28 [---] Finished CPU benchmarks

[mikkyo]

That doesn't look too bad, but I think with recompiles we can get better numbers.

BTW the benchmarks code has changed in boinc client 4.20 and up, including the current version, 4.27.
It looks like there are several additional tests.
So when you see newer boinc benchmarks, keep that in mind.
It will be a while before 10k dhrystones numbers show up.

[Lateralus]

Dual 1.80GHz PowerPC 7447A G4s/1.5GBs/10.4.2 - BOINC Menubar w/ 4.72:

2005-09-23 10:48:40 [---] Benchmark results:
2005-09-23 10:48:40 [---] Number of CPUs: 2
2005-09-23 10:48:40 [---] 2606 double precision MIPS (Whetstone) per CPU
2005-09-23 10:48:40 [---] 7302 integer MIPS (Dhrystone) per CPU
2005-09-23 10:48:40 [---] Finished CPU benchmarks

[beadman]

Hi, Lateralus.
Now switch out the 4.43 version in menubar for the 4.44 superbench version. The combination of 4.44 Superbench with the new highly-optimized SETI client from alexkan will give you a claimed credit of just about the same as what you'll eventually receive as granted credit.

beadman