Posts by Skip Da Shu

1) Questions and Answers : Wish list : Merge computers in view list... (Message 15931)
Posted 11 May 2006 by Profile Skip Da Shu
Post:
Currently the merge function is turned off. It is expected it will be turned on again in a few weeks. Merging was causing some problems with server performance. It is a BOINC issue and there is a fix coming.

Regards
Phil


Please "Announce" when both Merge and Delete are turned back on...I've got some clean-up to get done also.
2) Message boards : Number crunching : Times for Celeron CPU's (Message 955)
Posted 4 Oct 2005 by Profile Skip Da Shu
Post:
Sorry, Just realized we've hijacked the Celeron thread with building cheap AMD crunchers... lol

I'll cease and desist. -- Skip
3) Message boards : Number crunching : Times for Celeron CPU's (Message 954)
Posted 4 Oct 2005 by Profile Skip Da Shu
Post:
http://www.marisan.nl/seti/reference.htm

Notice the Sempron 3100+ (Paris core) with the SSE2 optimized app is faster than the fastest XP (my 2800 DTR).


I haven't run the SETI reference unit, with optimized app, on my XP 3000+ (Barton) but expect it to be quicker (yes, it is overclocked to approx. 2.4GHz) But since the Sempron is readily available while the XP isn't, I guess that's academic.


ZipZoomFly has the XP 3K and 3.2K but I was looking for the Desk Top Replacement(DTR) version (OPN AXDL instead of AXDA with the ability to lower the multiplier).

Are you running a 185MHz FSB with the 3000+ that has the 13x multiplier or a 218FSB on the 11x version? Is this on air cooling?

Probably would be faster if it's on an nForce mobo, probably not if it's on a VIA. At least for SETI. You can see on his list my 2500+ Barton @2.18GHz, nForce IGP is faster than my 2500+ Barton @2.19 on a KT600 (which I've since got running a bit faster with a bit lower memory timings).

U can download this to run the SETI benchmarks... http://home.austin.rr.com/skipsjunk/files/amd_ref_unit_bnchmrk.zip

Should take you about 6.5 hours to run the stock, r7 and r8 benchmarks or under 2 hours for the R8 only.
4) Message boards : Number crunching : Increased to 512MB as recommended memory requirement (Message 949)
Posted 4 Oct 2005 by Profile Skip Da Shu
Post:
I agree I only have 256Mb and am now 91% through 3-4 hrs to go and I will be uploading the WU


Do you have your settings to leave the app in memory or to swap it out?

-- Skip


I had Rosetta running on my 3 'normal' machines that have 512MB or better for a couple days w/o problems so I added rosetta to the slowest of my dedicated crunchers (XP 2000 OC'd to 1.99GHz). This 'basket cruncher' has a single 256MB stick of PC3200 and let it run for 48hours (WinXP Pro SP2). For the first few hours I suspend everything except Rosetta and the 1st two W/Us averaged out at about 2 hours to finish. Haven't seen any problems and so have just added Rosetta to the other 256MB XP based 'basket crunchers'.

If I get bored later this week I might try running it on one of the old 128MB boxes just to see what it does.

[SOAP BOX MODE ON]
256MB is overkill for the majority of the production BOINC based projects. I have two old slow machines w/ 128MB and they run 4 or 5 of the major projects (no CPDN on them, no E@H on the other but not due to memory). The PIII 550 w/ 128MB runs 5 BOINC project and DIMES! Not fast, but runs fine. So these 1G and 2G recommendations have nothing to do with BOINC (and yes, I have a GIG of sweet OCZ platinum 2-3-2-10 dual channel in my desktop).

The point I'm driving at here is IF / WHEN Rosetta really requires 512MB they will lose some hosts. Would it be a significant number? I think so. I'm sure I'm not the only one who will not go buy 'X' sticks of memory for 'the farm' just to run Rosetta on the crunchers. Roughly, you have to get past the 1st page of ranked participant to find one that's running less than 6 machines. It'd be interesting to run some sort of query to see how many hosts have under 499MB despite this projects recommendation of 512MB and what their credits total to for an idea of what the drop off for the project would be.

If you run BOINC as a service on a service/application stripped cruncher w/o graphics I'm sure I don't understand the need for more than 256MB of physical memory on a windows box and probably less on a Linux box.
[SOAP BOX MODE OFF]

Oh well, I started this out just to say, at least today, all is running fine on 256MB boxes.

Comments welcome, flames ignored. -- Skip

5) Message boards : Number crunching : Increased to 512MB as recommended memory requirement (Message 917)
Posted 3 Oct 2005 by Profile Skip Da Shu
Post:
I agree I only have 256Mb and am now 91% through 3-4 hrs to go and I will be uploading the WU


Do you have your settings to leave the app in memory or to swap it out?

-- Skip
6) Message boards : Number crunching : Times for Celeron CPU's (Message 899)
Posted 2 Oct 2005 by Profile Skip Da Shu
Post:
You can build a faster machine for less money if you'd go with a Sempron socket 754 CPU. The latest version has all the bell en whistles of the AMD64, but with a smaller cache.

The socket754/939 CPU's are very cool CPU's, especialy when compared to P4/Celeron's based on the Prescott core.

As I was planning/pricing out my next "basket" cruncher I started down my usual path of finding prices on Athlon XPs. Was hoping to find another 2800 DTR version but couldn't locate one. Then, after reading some articles on the newer Semprons and their ability to overclock, I decided to price out a 754 socket set up instead of my usual socket A. Yup, looks like the price / performance line has finally shifted to the Semprons. Next cruncher will be based on the later revision of a Palermo (OPN with the BO in it, I think) core. Looks like around $240 for a reasonable mobo with some OC ability, memory(CL2), S64 2800+, HSF and a 40GB HDD.

BTW, here's a good place to compare CPUs running the SETI reference benchmark work unit... http://www.marisan.nl/seti/reference.htm

Notice the Sempron 3100+ (Paris core) with the SSE2 optimized app is faster than the fastest XP (my 2800 DTR). Not as fast using stock application but faster once optimized and I don't think that's an OC'd clock speed. Lemme get my hands on my new Palermo core soon. ;-)
7) Message boards : Cafe Rosetta : Welcome to Boinc Synergy Team (Message 830)
Posted 30 Sep 2005 by Profile Skip Da Shu
Post:
Sorry, sig test... hmmm, sig ok, avatar sucks






©2024 University of Washington
https://www.bakerlab.org