Benchmark thread

[Scoofy12]

Since the whetstone/dhrystone benchmarks affect your score so much, here's a thread to collect benchmarks... to find out what we could/should be getting. There are a lot of factors, so please make sure you specify which version you use (and who compiled it/where it came from)

I have some dual Xeon 2.8GHz (1M L2) machines (HT enabled, but the benchmark ignores virtual CPUS) running the linux SSE3 4.72 client from stats.free-dc.org that are benching thusly:
2005-12-13 13:39:14 [---] Number of CPUs: 2
2005-12-13 13:39:14 [---] 2406 double precision MIPS (Whetstone) per CPU
2005-12-13 13:39:14 [---] 1727 integer MIPS (Dhrystone) per CPU

i seem to recall seeing evidence that that's a little low. can someone prove it to me?

here are other benchmarks:
P4 3.4G (1M L2, HT on) running same linux SSE3 4.72 client:
2005-12-10 12:35:12 [---] Number of CPUs: 1
2005-12-10 12:35:12 [---] 2914 double precision MIPS (Whetstone) per CPU
2005-12-10 12:35:12 [---] 4690 integer MIPS (Dhrystone) per CPU

seems not quite right compared to the other one, unless memory bandwidth significantly affects the dhrystone score (which it very well may... on 2 procs it runs 2 benchmarks concurrently)

here's an Athlon XP 2100+ (1.73GHz, 512k L2) running linux SSE 4.72 client from stats.free-dc.org
2005-12-13 23:19:48 [---] Number of CPUs: 1
2005-12-13 23:19:48 [---] 1782 double precision MIPS (Whetstone) per CPU
2005-12-13 23:19:48 [---] 3138 integer MIPS (Dhrystone) per CPU

Here's an Athlon XP2800+ running (i believe) the windows SSE 4.72 client from stats.free-dc.org
1778 whetstone (FP)
4418 dhrystone (int)

I have a 5.3 client at school that i thought was performing a bit low, but i'll have to post its stats later.

[Bok]

Scoofy,

get the 5.2.14 clients from here (http://www.guntec.de/Crunch3r/) and try them again...

Bok

I'll add in some benchmarks tomorrow.

[Scoofy12]

I installed it on the windows machine (Athlon XP 2800+) and the benchmarks are now up to 2191 FP and 6038 int. not bad.

new score for the P4 3.4 is
2005-12-15 12:00:48 [---] 3320 double precision MIPS (Whetstone) per CPU
2005-12-15 12:00:48 [---] 5848 integer MIPS (Dhrystone) per CPU

new score for the dual Xeon 2.8 is
2005-12-15 11:55:24 [---] 1841 double precision MIPS (Whetstone) per CPU
2005-12-15 11:55:24 [---] 2305 integer MIPS (Dhrystone) per CPU

so for the duallie, got a lower FPU score and a higher integer score. i wonder which one matters more. rosetta isnt to FPU intensive, is it?

anyone else have number for duallies (especially intel ones that share an FSB as opposed to opterons, which don't)?

[Scoofy12]

and last, the P4 1.4G with the windows crunc3r 5.2.13 client:
928 FP, 1674 int

[PY 222]

I highly recommend the 5.2.14 for P4/Linux combo as it speeds up the benchmark considerably.

[Scoofy12]

here's interesting results from a P4 3.8G
benchmark v4.72:
2005-12-19 15:12:50 [---] Benchmark results:
2005-12-19 15:12:50 [---] Number of CPUs: 1
2005-12-19 15:12:50 [---] 3216 double precision MIPS (Whetstone) per CPU
2005-12-19 15:12:50 [---] 5497 integer MIPS (Dhrystone) per CPU

then benchmark on the new 5.2.14:
2005-12-19 15:37:06 [---] Number of CPUs: 1
2005-12-19 15:37:06 [---] 3704 double precision MIPS (Whetstone) per CPU
2005-12-19 15:37:06 [---] 6532 integer MIPS (Dhrystone) per CPU

nice

[PY 222]

Woh! That are some awesome numbers.

As advertised, the 5.2.14 is really a butt kicker. :woot:

[Chuck]

Sorry, but I didn't grab the gui output, I opted for the XML instead.
I've removed the xml'isms that could cause you all any problems.

I also ask you all to forgive my n00b status again... I've been wrapped up
in some deep stuff in the lab.

As for my machines.. (which technically would be KILLER servers)

The results are:

p_ncpus 2
p_vendor AuthenticAMD
p_model AMD Opteron(tm) Processor 250
p_fpops 2367647285.942213
p_iops 8460828841.759756
p_membw 1000000000.000000 <--- right where I have it set.
p_calculated 1135931402.668503


This is a Tyan Mobo with L3 cache, full Numa and all the memory interleaving,
interlacing matrix-switch capabilities you could think of for mem, I/O, cpu-cpu, just like the numa is (as I have always thought of) supposed to be **AND** I have cpu affinity control in this version of BOINC.

With all that, running Rosetta... their 4.81 application runs for 2+ hours/cpu,
generating ~1800 pf/sec per cpu. Is this typical for everyone else?


What is interesting is boinc 5.2.13 behaves completely different as far as cpu WU download queue is concerned on an ASUS board with similar CPU performance (but with 100% ioslated memory busses and very limited NUMA features).

I have BOINC 5.3.2 here as well and can get even higher performance numbers, but does NOT have cpu priority or affinity control. But the application (4.81) runs wild and spends a great deal of time on the inter-cpu private numa channels swapping, resulting in a very limited throughput.


I have this problem whether running win/2k, win/xp, FC3, FC4... and I suspect Gentoo (about to load it). I've been away from things too long.


Should I be happy with this or am I missing a totally n00b obvious thing? I cannot imagine how you all are getting such high performance (like Bok) when I am lucky to scrape up a fraction of said throughput.

Tnx,
C.

[Chuck]

As a followup to my last post, I figured I would give you a 1:1 comparison between the two clients on the same hardware/software config.

This is the Bok-recommended client:

2005-12-30 03:33:13 [---] General prefs: no separate prefs for home; using your defaults
2005-12-30 03:33:13 [---] Remote control allowed
2005-12-30 03:33:15 [---] Running CPU benchmarks
2005-12-30 03:34:14 [---] Benchmark results:
2005-12-30 03:34:14 [---] Number of CPUs: 2
2005-12-30 03:34:14 [---] 3326 double precision MIPS (Whetstone) per CPU
2005-12-30 03:34:14 [---] 10109 integer MIPS (Dhrystone) per CPU
2005-12-30 03:34:14 [---] Finished CPU benchmarks


I am watching the cpus and seeing it do processor swaps at a fairly high rate.
I'm beginning to think there is too much memory in my machines (since they are linux oriented but currently running win/xx) for things to work properly.... I truly have no clue why I am not 'hauling buns'. Does anyone?

Tnx,
C.

PS: where's that little windbloze trick go for having the client '-return_results_immediately' ?? Is that a registry edit? (off-topic ... please pm me the answer)

[Bok]

I would say those benchmarks are spot on, if a little high actually for a 2.4Ghz processor, for comparison, I had just over 11000 when running at 2.77Ghz on an opteron 170.

I'm not sure you can do return_results_immediately with the windows client except for using the truxoft client where he added the code for it and you just put the line in the remote_hosts.cfg

Bok

[Angus]

The return_results_immediately function has been removed from the official BOINC core clients since about 4.3 versions.

The trux optimized core clients have it included, but the method of turning it on has changed.

With his older 5.3 versions, it needed a remote_hosts.cfg file with #return_results_immediately in it.

His current version 5.3.6.tx8 implements the function by adding <return_results_immediately/> to the global_prefs.xml file.

HOWEVER, I have NOT been able to get it to work on the new clients.
I restarted Rosetta yesterday on one box, with the 5.3.6.tx8 SSE client, and I'm watching it to see what happens.


On the benchmanrks, my dual Xeon 2.8 HT (Dell 2650 rackmounts) boxes got 1240/2600 on the older trux 5.3 core client. Don't have any numbers from the standard clients.

[Chuck]

I would say those benchmarks are spot on, if a little high actually for a 2.4Ghz processor, for comparison, I had just over 11000 when running at 2.77Ghz on an opteron 170.

I'm not sure you can do return_results_immediately with the windows client except for using the truxoft client where he added the code for it and you just put the line in the remote_hosts.cfg

Bok


Bok & Angus,
Thank you both VERY VERY MUCH ... I was wondering/worried... One machine is running @ 2.56 ghz (Asus). The Tyan mobo has variable FSB which is set at 232 Mhz [max rate *I* permit] and I counted 8, full speed NUMA channels in the cross-bar numa switch... so it should haul. I also found a BIOS update (talk about a surprise!) which is rare from tyan. I will update to bios 1.02 a bit later.
I was really concerned about has happened to BOINC while gone.. My machines (both of which you know were not cheap to build) were almost at the 'toss to scrap' stage in my mind. They do math (prime number proofs and huge prime number memory sieves) like crazy though.. hence the confusion. I guess Rosetta is simply that poorly written... must be C++ with a lot of 'new' and 'delete' statements in it. :(

I will let the work queue run out and put these babies to work on DPAD to check it out there under extended test. I quick-tested the client. Running 1 instance of it will use the whole machine... it is apparantly cpu/core smart. VERY memory efficient too (less than 100 pf/sec) and started writing the results.txt file in a few minutes.

Now that I know what the scoop is. I will "reacquaint" myself with what happened to the DC/BOINC world.

Thank you both for taking the time... I was afraid I'ld have to make a trip down to the Austin office (about 3 hrs drive) and pick up some of the new triple-core cpus just to keep up. I am glad I can belay spending the $4000+ on cpus until after my new car arrives. :)

Cheers,
Chuck

[Scooby Doo]

XP2400 (standard client)
1776 double precision MIPS (Whetstone) per CPU
2922 integer MIPS (Dhrystone) per CPU

(optimised client)
2077 double precision MIPS (Whetstone) per CPU
5689 integer MIPS (Dhrystone) per CPU

[LAURENU2]

I am getting this after the upgrad
12/29/2005 10:54:00 PM||Starting BOINC client version 5.2.13 for windows_intelx86
12/29/2005 10:54:00 PM||libcurl/7.14.0 OpenSSL/0.9.8 zlib/1.2.3
12/29/2005 10:54:00 PM||Data directory: C:\Program Files\BOINC
12/29/2005 10:54:00 PM||Processor: 1 AuthenticAMD AMD Athlon(tm) XP 2600+
12/29/2005 10:54:00 PM||Memory: 511.48 MB physical, 1.22 GB virtual
12/29/2005 10:54:00 PM||Disk: 57.25 GB total, 53.14 GB free
12/29/2005 10:54:00 PM||Version change detected (5.2.7 -> 5.2.13); running CPU benchmarks
12/29/2005 10:54:00 PM|rosetta@home|Computer ID: Peekaboo; location: home; project prefs: default
12/29/2005 10:54:00 PM||General prefs: from rosetta@home (last modified 2005-12-28 02:55:45)
12/29/2005 10:54:00 PM||General prefs: no separate prefs for home; using your defaults
12/29/2005 10:54:01 PM||Remote control not allowed; using loopback address
12/29/2005 10:54:03 PM||Running CPU benchmarks
12/29/2005 10:55:02 PM||Benchmark results:
12/29/2005 10:55:02 PM|| Number of CPUs: 1
12/29/2005 10:55:02 PM|| 2337 double precision MIPS (Whetstone) per CPU
12/29/2005 10:55:02 PM|| 6460 integer MIPS (Dhrystone) per CPU
12/29/2005 10:55:02 PM||Finished CPU benchmarks

[Angus]

And here (http://setiathome.berkeley.edu/forum_thread.php?id=25541#223448) is a message about the fix for the non-functional return_results_immediately feature.
Trux's clients are here (http://boinc.truxoft.com/)

[Bok]

Bok & Angus,
Thank you both VERY VERY MUCH ... I was wondering/worried... One machine is running @ 2.56 ghz (Asus). The Tyan mobo has variable FSB which is set at 232 Mhz [max rate *I* permit] and I counted 8, full speed NUMA channels in the cross-bar numa switch...

*snip*


I used to play with NUMA a lot, in fact I sysadmin'd Sequent NUMA-Q machines for a while (they eventually were bought out by IBM for $800M). Back in '99 we had a 4-way 450Mhz NUMA-Q with 4Gb Ram and about 300Gb storage over 10 disks or so. Cost about $250K at the time, though it was on a lease. Excellent piece of machinary and I really liked the dynix pts O/S. Went to Sequent headquarters for a week, just outside Portland, OR and they had 18 of these linked together :)

Bok

[Evilmoose]

Hows this for X2 3800 @ 2518MHz

12/30/2005 7:30:52 PM||Running CPU benchmarks
12/30/2005 7:31:52 PM||Benchmark results:
12/30/2005 7:31:52 PM|| Number of CPUs: 2
12/30/2005 7:31:52 PM|| 2484 double precision MIPS (Whetstone) per CPU
12/30/2005 7:31:52 PM|| 8534 integer MIPS (Dhrystone) per CPU
12/30/2005 7:31:52 PM||Finished CPU benchmarks

Seems allitle low?

[Bok]

Seems just a touch low.. which boinc clients are you using ?

these (http://boinc.truxoft.com/) or these (http://www.guntec.de/Crunch3r/) ...

second ones are faster historically, though I actually just tried trux's new one on a 4200+@2.52 and got just over 10000 for the integer test..

Bok

[Evilmoose]

12/30/2005 7:56:14 PM||Version change detected (5.3.1 -> 5.3.6); running CPU benchmarks
12/30/2005 7:56:16 PM||Running CPU benchmarks
12/30/2005 7:57:19 PM||Benchmark results:
12/30/2005 7:57:19 PM|| Number of CPUs: 2
12/30/2005 7:57:19 PM|| 3171 double precision MIPS (Whetstone) per CPU
12/30/2005 7:57:19 PM|| 4245 integer MIPS (Dhrystone) per CPU
12/30/2005 7:57:19 PM||Finished CPU benchmarks

ouch lets try onther one


12/30/2005 8:06:53 PM||Version change detected (5.3.6 -> 5.2.7); running CPU benchmarks
12/30/2005 8:06:55 PM||Running CPU benchmarks
12/30/2005 8:07:54 PM||Benchmark results:
12/30/2005 8:07:54 PM|| Number of CPUs: 2
12/30/2005 8:07:54 PM|| 3162 double precision MIPS (Whetstone) per CPU
12/30/2005 8:07:54 PM|| 5528 integer MIPS (Dhrystone) per CPU
12/30/2005 8:07:54 PM||Finished CPU benchmarks

[Bok]

You are using the SSE2 version right ? That looks like the SSE1 to me...

[Evilmoose]

this is more like it

12/30/2005 8:17:54 PM||Version change detected (5.2.7 -> 5.2.13); running CPU benchmarks
12/30/2005 8:20:21 PM||Benchmark results:
12/30/2005 8:20:21 PM|| Number of CPUs: 2
12/30/2005 8:20:21 PM|| 3491 double precision MIPS (Whetstone) per CPU
12/30/2005 8:20:21 PM|| 10696 integer MIPS (Dhrystone) per CPU
12/30/2005 8:20:21 PM||Finished CPU benchmarks

I satisfied for now ;)

I was using the SSE2 version but the first benchmark after a version change always seem slow, so I reran the benchmark on 3 different clients. This one is just allitle faster then the trux ones.

[Bok]

yup, that looks right!

:cheers:
Bok

[Digital Parasite]

I used to play with NUMA a lot, in fact I sysadmin'd Sequent NUMA-Q machines for a while (they eventually were bought out by IBM for $800M).

Hmmm we might need to talk then. I have access to a 128CPU Itanium2 SGI Altix NUMA system that doesn't seem to be performing as well as possible. I wanted to run the GIMPS verification on it but it was going 2x slower than the other guy's NUMA box. They are just setting it up so its not fully configured yet. Any NUMA advice would be helpful.

[Bok]

What O/S ? Is it running Irix ?

Main thing on NUMA machines is to check and re-check the fibre connections. Make sure the cross-connections are all setup correctly.. Been a few years mind. What exactly are the problems ?

Bok

[Shish]

Ooh...you know about fibre as well Bok?

I was glad when I retired 2 of those bloody things.....sick of lying on my back inside them to get at the cards and wiggle them. Nearlt always burnt my fingers as well.:hiya:

[LAURENU2]

Anyone want to comment on what was posted in the Rosetta forum

http://boinc.bakerlab.org/rosetta/forum_thread.php?id=845

-----------------------------------------------------------------
How the @#$% are you claiming so much credit with those benchmarks?!?!

Excused for my English.
They are not mine account. I have only made of the controls I intercross to you on of the hosts doubts that perhaps cheat. The problem of Rosette is quorum 1/1 where the benchmark he is fundamental for the credit. There are many possibilities of credit, client officiali, client optimizes, different methods of benchmark, Ht on/off, overclocching, etc.... I have experimented various client on various PC, with various client, the result is always various but in the norms of 250 credit to the maximum of the performances. Over these values the thing becomes suspects, and then I have tried to understand as they make; enough to modify the result of the benchmark in the 2 files Client_state and Client_stat_prev.xml. Example: You suspend attivity boinc, Run benchmark, the result is 2000/4000, you exit from boinc, you open client_stat with editor xml modification the result of 2000/4000 in 12000/14000 I have only added a 1 in front of the values you exit and saves. You open client_stat_prev and you make the thing equal. Run Boinc you make update and you have credits x 100; that is 10=100 20=200 50=500 etc..... I have had to try on mine account on 15 WU in order to see if it is right or and I cannot say that it works perfectly. I assume myself all my responsibilities of forehead to all the persons who join to several the Boinc plans and to all the Staff that the plans direct several. Aspect the decisions of the leaders of the staff of Rosette in my comparisons hoping that my search is served in order to improve the plan and to avoid futures errors. In wait I augur good a 2006 for all the world. Ps: excuse to Stephan_t that I had raised the problem and to the beginning I had not understood..........
:bang: :bang: :bang: It Is A Shame People even try to do this:bang: :bang: :bang:

[Digital Parasite]

What O/S ? Is it running Irix ?

Main thing on NUMA machines is to check and re-check the fibre connections. Make sure the cross-connections are all setup correctly.. Been a few years mind. What exactly are the problems ?

Its running Linux. I don't have physical access to the machine, I'm just a lowly luser. They just got it and don't have it fully configured/setup properly so things might be fine in the near future. Nice to have access to that kind of power though.

[Shish]

Duunno what that was all about Laurenu2, I couldn`t be bothered to wade through it all.
I`ve finally got contact with one of my wayward duallies (in a new bar at Oldham) and loaded the new Rosetta and Dimes on it.
Just a lowly dual 2000XP on an Asus A7M266-D but I`ve put the optimized latest files into Dimes and Boinc and it ain`t running anything else but a wifi access point service (like Starbucks, only free for customers) so it should do alright.
Got the confirmation of my order for the Opteron 165 and a "preference" pre order for another from a firm who makes servers so actually have some of the limited supplies of Opterons currently available and as I`ve put a fair bit of business their way in the past, I should manage another 2 from them and at a good price (bit less than current retail and less the vat) so I`ll have one up and running by the weekend and another 2 sometime towards the months end, depending on funds for decent memory and drives.
So, full steam ahead and I`ll post some benchies for the new rigs (nice to see how these Opties running on 4400 memory can do) as soon as I finish sorting them.

[DocWardo]

Thanks again to Laurenu2 (You are way to generous!) for this new Sempron 2200 rig. I finally had a chance to set it up and after upgrading the ram to 512 MB and installing Fedora Core 4 (it had linspire or whatever it was, but I have FC4 on my other server and wanted some consistancy. I upgraded to the optimized clients suggested by BOK (using the 5.2.14 client for P3/Athlon XP with SSE)

Here is my benchmarks:
3582 integer MIPS (Dhrystone) per CPU
1680 double recision MIPS (Whetstone) per CPU


Does this sound about right?
if not, any suggestions?

[Shish]

OK, got the new Opti 165 burning in gently but took some benches while it was doing Memtest etc and Rosetta.
Finally dropped it back to 2520Mhz on a little more than default volts (1.4ish) to let it settle in but I`m really pleased with it and the Thermalright SI 120 has kept it remarkably cool (even had the thermal fan stop as it went under 40C, scared the s#@t outa me) so the results so far bode well for the water cooled case it`s going in and it`s already very quiet and running at 45C ish.
So here goes:-

Started at default 1800Mhz, actually 1810Mhz.
1716 and 3186 @ 1809Mhz.
2137 and 4000 @ 2251Mhz
2232 and 4123 @2363Mhz
2345 and 4367 @2475Mhz
2382 and 4441 @2522Mhz
2474 and 4618 @2620Mhz.

So a fairly linear progression as you can see.
The memory (Geil Ultra 4400) still hasn`t peaked after getting to 300Mhz but I`ve left anymore clocking until it all settles down and I get it into the water cooled case though it hasn`t anywhere near peaked out and max temp has only been 55C at max 1.55v and 2650 reliable for 3 hours on Memtest and Rosetta plus looping Doom4 bots.
It eats video processing (Maya and a Japanese cartooning processor prog) and is just loping along at 2520Mhz.......well pleased and def going for some more, especially if the retail price drops again cos now I know where to get `em from.

[Bok]

That's the standard BOINC isn't it, not the optimized version ?

You would get > 10000 for the integer score at 2.5Ghz or so.

What is the stepping of this opty ? My 0530 170 will do 2.5@stock voltage and 2.6Ghz with only +0.25 volts running at 46C under load. I'm guessing your's is perhaps an 0546 ?

Bok

[DragonOrta]

I was about to say, those numbers don't look right at all. This optimized client seems to do wonders, then.

[Shish]

:blush: Sorry, meant to say that`s the OPTImised one from Truxsoft.
When I get a bit more time to play with the rather complicated overclocking on this board (MSI K8N Neo2) and the cpu has bedded in (most Optis seem to need this) I`ll get it up to 2700+ ,at least for a while, and see what it does then. I`ve had it, briefly, to 2780 and into Windoze, but it needs considerable tweaking and better cooling to stay there cos I don`t like 60C+ temps.

Edit:-
Just realised I used the files I downloaded for the Athlon sse and this cpu has sse2 as well....silly me.
I`ll redo it later and we`ll see if sse2 makes a difference.
Off to get some breakfast (:blush: :slap: been in bed for 21 hours) or at least a mixed grill so I`ll sort it out later cos I need to take a load of pills as catchup time and that needs food.

[Shish]

OK, here`s the latest before I start optimising the memory etc and continuing my search for MORESPEED (OK, I`ve always been a Speedfreak, especially after a very hard to organise flight in a Lightning 2 seat fighter/bomber with an old friend, Wing Commander Perkins. Man, sitting on top of 2 engines, wrapped in some thin alloy and plastic,with a G-suit squashing the blood back up to your head, then going vertical on full afterburners to 50,000 ft from sealevel and then whooping across the North Sea at very low level above the speed of sound (have to watch out for fishing boats, tends to get them upset apparently) is the biggest buzz I`m likely to have ever unless they get this Virgin Galactic thing going or let me stowaway on the next Shuttle mission).

Anywy, enough of fond memories of youth......

TruXsoft client @2520Mhz=2296 and 4285 and that`s the SSE2 one.
Guntec.de client@2520=2326 and 4283 also with the SSE2 one

So Guntec.de gives a slight lead but it`s too close to call.

After I do some multiplier dropping and memory upping (fantastic range on this Geill 4400 3.5nsec memory) and see what voltage is appropriate I`ll continue the benchmarks as and when.

[Bok]

Those benchmarks are very very wrong Shish. You should be much closer to 10000 with those clients....

Bok

[Shish]

OK, after several attempts to finish a post, I`ve switched comps, unless you can put in an autosave feature Bok?
Got a heap of data here but won`t post most of it, including SiSoft benches unless anybody wants it. BUT....SiSoft says the benches are better than a new Intel board running a P4EE@3.46.
However, I think I`ve damaged Windoze and some of the filesystem so I may need a reinstall.

Not sure about what`s up with the benches in Boinc Manager cos I`m running the latest client (from the Boinc download page) with the Trux or other files installed as instructed.

Anyway, here`s WCPUID once I figgered that I had to downsize it for upload.

I`ve also figgered out what the MSI Smartfan does after a heart stopping climb to 60C with the cpu fan stopped and a VERY rapid shutdown.:scared: :swear: That`s what stops the fan if you set the bios to 40C/50C/60C plus or minus 1-5 where 1 is bigger (strange Chinese reasoning?) variation allowed.

What else...?
Sandra benches got it performing better than a new Intel board (955?) and an Intel P4 EE @3.46 with an enormous memory bandwidth which I can`t find a match for but forget looking at any of Intel`s DDR2 stuff, this hammers it (appropriate pun intended).
Only thing approaching it and that includes using CL2 Registered ECC for some reason, is the AMD chipset boards which I think are AMD 8131 chipset and that`s what I want next (when I get a mortgage on the new house....didn`t want one really..... or the daughter pays me for this one....may not happen in my lifetime.) as well as a couple of 265 Optis (need a bloody new mortgage for those as well....God knows what the next 4 core ones will be :|party|: :confused: :idea: mebbe sell both the daughters to some Arab sheik.....:scared: )

So, what`s up with the benches....any ideas?

And after a rapid search through 3 million bits of paper and sticky ups (have to ring my mate at 3M for another box or get a new lock on the stationery cabinet :D ) I`ve found where I wrote the cpu stepping (on the invoice of course stoopid:cry: ).
So itsa...
OSA 165 DAA6CD
CCBWE 0547 TPMW
1375764L90001

I`ve now figured out that the board volts wobbles a bit (common with MSI on all their boards I`ve had) and I can`t get the memory up to it`s max volts of 3.1v, so I`m stuck with 2.85v and that`s only good enough for 285FSB.
Still trying to figure out some of the other settings but it`s now running at 1.38v and seems OK.

[Shish]

Can`t figger how to put pics in the actual post...Bok???

[Shish]

OK, got the benches figgered.
I`d put the files into the wrong Bonk in an old copy of programme files.
New corrected benches:-
3505 and 10723 @ 2563.
Better...silly mistake as I`m using half the old server raid 5 array till I can upgrade to SataII and a new Highpoint 8 channel PCI-X card.
Phew :swear: :smoking: :lmao: :blush:

Edit:-
Still trying to figger the ocing on this board, defo need the max memory volts but I`m down to around 1.35v on the cpu atm...but need more than the max 2.85 for this memory to run optimally.

[Shish]

OK, gotta go and stop playing here....got dinner for 12 to do and roast leg of lamb been marinating in 4 Bells Rum for 2 days.

Just back to backed the 2 optimised clients and the Guntec.de one has beaten the Truxsoft one easy. That`s the second one linked on the 2nd optimised clients page.

I`ve defo fsked Windoze tho` as it keeps wanting to tell Micro$ that IE has got an error and very sorry but stuff you I`m closing even if you are in the middle of a post.:bang: :looney: :umm:
I`ve deleted the bmp file, sorry about the resizing....replaced with a jpeg of suitable size, just click it for full size cos I can`t figger out how to get it display properly with this new version.....just me probably.

903

[Digital Parasite]

Man, I really hate the BOINC benchmark system. Is there any way you can set how often the benchmark gets re-run? Every week or two I notice my main dual-rig starts getting 1/3 or 1/2 of the points in normally does. Since I have to actually do real work (yeah I know) I seem to have the bad luck of running something else when the benchmark re-runs. That means it sets the value much below the normal bench and I get penalized for all the points after that until I realize again. :cry:

[Shish]

You can always run them manually from inside Bonk Manager. I don`t leave it running after putting the connect every result thing in as it buggers the startup on most rigs and I`ve not got on very well with BonkView so I`ve junked that.
I just run Bonk Manager when I want to check things.

[Shish]

OK, just finished sorting through some bits and what nots.
Tempted to lash out on both some 9700Pro All In Wonders (new in box, very good price, very good make) I`ve just been offered or maybe the first of the Dell 24 widescreens ( I`m running Dell 2001s just now and they`re superb) I`ve been offered as refurbs ( I know they`ve put new panels in `em at least cos these are the returned ones with dead pixels mostly but the firm does a pixel check on `em for free anyway) also at a good price. The 2005s were OK but seem very small in their widescreen format after you`ve been used to the 2001. Anyway, I`ve dealt with the company concerned before and they know better than to give me hassle cos they get it back in spades:D
Back on topic, I`m gonna be setting up my old fleet of ex Xeon server boards again though I was going to Ebay them and I`ll bench them as I go and I`ve got several spare Athlon boards or may get some cheap NV2 ones for all the old XPs I`ve got lying around. Be nice if somebody could give me some comparitive benchmarks but I`m gonna set `em all up with truXoft`s send immediately which seems to be working well and Guntec.de for the optimised files as they seem to have a slight advantage on all the ones I`ve tried to date (now I`ve got `em in the right place :D ).
All the 5 I`ve got running just now have that config of the truXoft send immediate and the Guntec.de optimisations and all are doing fine.
Still got to sort out the final oc`ing for the new Opteron 165 but after further testing with some loaned 4800 (bloody expensive) memory, it seems to be topping out at around 2.626Ghz on air and reliable/solid but will go higher if I can get the 3.1v for the memory and some water cooling which is on my list cos this one`s gonna be my lounge lizard server if I can get this stupid Cisco to handle Ultra Wideband MIMO from the router frame bits they`ve given me to test. Might shell out on the big, look Ma, no fans, water tower jobbie cos it`s dropping in price tho` I can`t find the other similar one with the base fans built in.
Anyway, hopefully get a good fleet going in the next month or so for us on Rosetta and they`ll be living in a refrigerated ex-office space, environment in a big ol` pub cellar in the Midlands with, hopefully not too many remote admin problems.

[Digital Parasite]

You can always run them manually from inside Bonk Manager. I don`t leave it running after putting the connect every result thing in as it buggers the startup on most rigs and I`ve not got on very well with BonkView so I`ve junked that.

Do the benchmarks only get re-run periodically if you have the BOINC manager running?

That is what I do, manually re-run the benchmark to get them back where they should be, but its a pain to have to check all the time and re-run to make sure I'm not getting jipped for my credits.

[Bok]

Shish,

XP's have pretty good benchmarks too. For instance, this is an XP2700+ at stock speed

here (http://boinc.bakerlab.org/rosetta/show_host_detail.php?hostid=64264)

This one is running a Gentoo linux build. I don't have any XP's on windows to compare unfortunately..

Bok

[Shish]

I`ve got 2 more dual XP boards to get going and several single XP older ones when we get this new pub organised.
Just comparing benchmarks between a dual Barton running 2.4Ghz which came out at 1689/6183 and the Opti 165 which came out at 3505/10723 at 2.51Ghz seems these new large cache machines are deof worth their salt.
I`ve got 3 old dual Xeons from 550 to 900 and free electric makes them worthwhile but the dual PIII 933 on a Tyan Tiger has never been very good and should be Ebayed.
Dunno what to do for the best, flog it and the other old Xeons off or keep `em and use `em. Don`t really like getting involved with Ebay much.
Anyway, after all that oc-ing I`ve got to reinstall windoze cos I`ve foobarred a lot of the file system and even had to reinstall java and Dimes so I guess it`s a re-image and t`other Barton at the same time to swap vieo cards until these 9700 AIWs arrive early next week.
Back to the 48 hour days but that`s the way my body works and part of my condition along with a catchup sleep of up to thirty hours which means I wake up worse than I went to sleep cos I`m supposed to eat pills (be nicer if it was Pils) every 4 and 6 hours plus a twelve hourly special handful.
Such is life but it can have it`s advantages when you`re on a trouble shooting project.
So, 2 rigs to sort for now (have to hard reinstall I`ve just realised cos you can`t swap Radeons as Windoze flakes out when going from 8500DV to 9800 AIW and vice versa....... windows stores the wrong atimtag file and won`t let you replace without a reinstall) .
OK...onwards and upwards. see you on the flip side.

[Thor]

Hey Sish, here are the benchies for my XP running W2K SP4 with the newest client from guntec.de(?):

Benchmark results:
Number of CPUs: 1
2587 double precision MIPS (Whetstone) per CPU
7117 integer MIPS (Dhrystone) per CPU

This cpu is an XP 3200 running 11*220Mhz FSB, RAM running at the same speed with the "agressive" timings, which makes it a 240 Mhz overclock

Maybe tonight I'll do a a bench at sock speed and a bench on an Xp2400...

Thor

[Shish]

My Opti looks good then.
I`m running at 278 FSB just now with fairly tight timings and dual channel.
I already know the Opti`s 1meg cache kicks butt in video but it looks to be doing well at most Dc as well.
Have to get the other 2 up and running, maybe after sorting the Xmas.New Year overspend tho`.:rotfl:

[Digital Parasite]

The first phase in upgrades is now complete, I put together a new system thanks to the FreeDC Christmas party. I upgrade my wife's old P3 800 so one of my clients has gone from:

Measured floating point speed: 552.71 million ops/sec
Measured integer speed: 1641.48 million ops/sec

to

Measured floating point speed: 2222.02 million ops/sec
Measured integer speed: 6742.19 million ops/sec

Sweet!

[IronBits]

Nice bump! :clap:

[Bok]

Cool,

Glad you could use those mobos/procs Jeff!!

Bok

[DocWardo]

I just wanted to add that I too (as noted above) have been using the overly large generosity of Lauren to enhance my stats as well. I have the 2200+ system up and running with 1/2 gig of ram and Fedora core 4, and I have put the check (I told you to keep it, but hey I need parts ;) ) in the bank and have started purchasing the parts to complete the 2600+ system (I'll post it's stats as soon as I finish it, and post what happens when I up my main desktop to 1.5 GB).

That was a wonderful idea to have that party. I do apologize for not being in chat more, but it is the start of the spring semester and I have been outrageously busy, in a week or so I'll start poppin my rear in chat more often. I hope others have done the same. Those of smaller means (both physical and computer skills) have wonderful crunching colleagues that help them out in both ways and in the most polite and understanding ways possible. I'll never look for another team, just point me to a project worthy of our cycles and I'm there for ya (with my/our computers and my electric bill of course :D ).

-Doc

PS. Now move out of the way Angus :beep: , my new and improved farm is movin on up!

PPS. Thank you Google toolbar for hopefully preventing me from embarrassing myself with stupid spelling mistakes.

[Digital Parasite]

Glad you could use those mobos/procs Jeff!!

Yeah, my wife has a snappy fast "new" machine and I have a much faster borg so everyone is happy. :clap: I got a good deal on a Sonata II case + PSU which I really like the layout of and its very quiet. Running at full load the CPU is only at 27C!!!

I'm slowly going to work on getting the other one up and running as well (to upgrade my Celery 900 linux server).

PS: Free-DC ROCKS!!!!!

[GHOST]

1/24/2006 7:21:37 PM|| 4360 double precision MIPS (Whetstone) per CPU

1/24/2006 7:21:37 PM|| 12571 integer MIPS (Dhrystone) per CPU

xp2500 at 10 x 198 winxp pro guntec

these benchmarks are more than double normal [running benchmark with all apps closed]

got the higher benchmark while running wintv and a video on vlc

i then tried benchmark while running several media players, or while running sandra, but have not reached those marks again. i'm running several hundred points lower right now, but not the 4000/5000 Dhrystone i usually get. really i think this system sucks.

[Shish]

I`d take those figures any day.
Can`t figure this benchmark system at all though. Seems skewed all ways to Xmas.

[GHOST]

1/25/2006 4:45:15 AM|| 1949 double precision MIPS (Whetstone) per CPU

1/25/2006 4:45:15 AM|| 3164 integer MIPS (Dhrystone) per CPU

P4/Celeron 2.2ghz dell laptop. i have turned this off the last couple weeks but gave it a try to see if i could double my benchmarks.

no go. when i load up apps the benchmark drops by half. the above numbers are the best with all aps off and turning off everything i could in task manager.

[Shish]

Hmmm......
Just tried the latest Truxoft client ( 5.3.12.tx36) and it dropped my benchmarks considreably compared to the Guntec.de one.
Instead of 10283 mips, I was down to 83256 mips which ain`t very good.