PDA

View Full Version : New pre-release clients



alpha
03-12-2003, 02:09 AM
Some new pre-release clients (http://www.distributed.net/download/prerelease.html) are available - sporting some nice new RC5-72 cores :)

new: x86: rc5-72 core by Gnatiuc Ianos (Athlon optimized)
new: ppc: rc5-72 core by Malcolm Howell (603e optimized)
new: ppc: two additional rc5-72 cores for ppc by Didier Levet
new: ppc: two rc5-72 cores by Didier Levet (AltiVec optimized)
new: arm: rc5-72 assembly cores by Peter Teichmann

All this amongst other updates which you can peruse at your leisure :D

the-mk
03-12-2003, 11:10 AM
I did some benchmarking to compare the old win32-client (ver 477) and the "new" (prereleased) one (ver 481)

RC5-72-Benchmark of client ver 477 (fastest core):

[Mar 12 15:14:40 UTC] RC5-72: using core #5 (DG 2-pipe).
[Mar 12 15:15:00 UTC] RC5-72: Benchmark for core #5 (DG 2-pipe)
0.00:00:16.32 [5,093,906 keys/sec]


RC5-72-Benchmark of client ver 481 (fastest core):

[Mar 12 15:21:38 UTC] RC5-72: using core #8 (SS 2-pipe).
[Mar 12 15:21:58 UTC] RC5-72: Benchmark for core #8 (SS 2-pipe)
0.00:00:16.57 [5,529,030 keys/sec]


Benchmark-situation: Windows 2000 SP 3, AMD XP2000+ (I don't know if it's a t-bred or a palomino; if it makes a difference...), 512 MB DDR333

Could someone with an Intel P4 (or so) please do a benchmark to see how much they are faster?

wirthi
03-12-2003, 12:06 PM
Originally posted by the-mk
Could someone with an Intel P4 (or so) please do a benchmark to see how much they are faster?
Of course :)

Old Client (V 2.9001-477)


[Mar 12 16:59:45 UTC] RC5-72: Benchmark for core #6 (DG 3-pipe)
0.00:00:17.37 [2,151,160 keys/sec]

New Client (V 2.9003-481)


[Mar 12 16:26:01 UTC] RC5-72: Benchmark for core #6 (DG 3-pipe)
0.00:00:17.51 [2,133,530 keys/sec]

That's a minor drop in my eyes. Both times core #6 was the fastest, but this time even slower than last time. It was mentioned the new core was optimized for Athlons, but they didn't mention making Intels even slower ... Using Intels for RC5 seems more and more useless (ok, I don't have a Nothwood, and I only have SDRam. Still ...)

CPU: 1.5 GHz P4, SDRam

Greets,
Christian

the-mk
03-13-2003, 09:45 AM
OK, there is not much performance-increase... so I will stay at OGR to push team Free-DC into the top 100 :D

@ wirthi: did you run winamp, icq, any programming-editor or so while you did your benchmarks? (could be a reason for the difference; I benchmarked it with no task running behind...)

TheOtherPhil
03-13-2003, 10:58 AM
Hmm, interesting. The new client works well on my dually AMD box running linux giving 13.94Mk/s. On my dually AMD under 2k and XP, I get 0kk/s and 50% CPU usage :confused:. Going back to the full release client solves the problem.

Edit: I also selected core 8 and set 2 threads manually but no joy.

wirthi
03-13-2003, 11:05 AM
Originally posted by the-mk
@ wirthi: did you run winamp, icq, any programming-editor or so while you did your benchmarks? (could be a reason for the difference; I benchmarked it with no task running behind...)
You think I'm nuts? I'm trying to squeeze every bit of power out of my P4 (still, honestly, there's not too much of it inside), so there were no other tasks running.

No increase? The new Athlon core is almost 10% faster than the last one according to your tests. The core the P4 is using feels slower, though.

TheOtherPhil
03-13-2003, 11:17 AM
I've submitted a bug report (3203) so hopefully this will get fixed.

TheOtherPhil
03-16-2003, 04:14 AM
Interesting stuff re: Client (V 2.9003-481)


------- Additional Comment #8 From Décio Luiz Gazzoni "acidblood" Filho 2003-03-16 00:26 GMT -------
Fixing the core has slowed it down considerably. Probably the best choice is to run one
instance of the faster, non-reentrant version, while running the slower version on the
other instances.

------- Additional Comment #9 From Décio Luiz Gazzoni "acidblood" Filho 2003-03-16 03:30 GMT -------
Aside from the reentrancy issue, this core has a problem whereby signal handlers on
Unix systems can overwrite the core's local variables, rendering the results incorrect. A
fix brings its performance down to ca. 94% of DG 2-pipe, and thus the core is being
removed from the client until the author provides a reentrant and signal handler safe
version with improved performance.


Sounds like a big no to the SS 2-pipe core.

TheOtherPhil
03-18-2003, 08:06 AM
New 482 clients with the ANSI and buggy Athlon cores removed, plus some minor enhancements: http://www.distributed.net/download/prerelease.html

IronBits
03-25-2003, 01:35 AM
Where did you go Phil? :cry:
Last seen 5 hours!!! :scared:

PY 222
03-25-2003, 04:11 AM
Nah... don't you worry there IB!

I am sure he is saving up enough workunits to bring you down with a huge flush :notworthy

TheOtherPhil
03-25-2003, 12:58 PM
:bang: :bang: :bang: My router crashed....funny thing is, it's a hardware router :eek:. I have now learnt that a local proxy is needed as my clients didn't have enough work cached locally :bang:. I am down in production today because of this :o

alpha
04-11-2003, 04:00 PM
More new pre-release clients, featuring:

fix: x86: re-added SS-2 core - new version fixes bug #3187
fix: linux/ppc: no longer coredumps while resolving (#3213)
new: ppc: support for altivec under linux (#1349)
new: sparc/solaris: optimized rc5-72 core by Didier Levet
new: sparc/solaris: cpu detection