Benchmarks - different OSs

[IronBits]

Ok - you guys that do random benchmarks and watch your clients alot more than I do, I have a couple of questions for you.

Which is the better OS, and how bad is Win98se compared to the other two in DF only, dedicated crunching, with bare installs?
And by how much better?

No p*ssing contests, just give me some raw data and facts to look at.

All AMD single processor boxen only please.
Any optimal client settings you have found for each is appreciated :)

Win98se vs Mandrake 9.1

Win98se vs Windows 2000

Mandrake 9.1 vs Windows 2000 (is there really *any* difference to speak of?

Thanks! :cheers:

[jonnyw]

can i semi hijack this thread asking for intel benchmarks as well?

:p

tell me to pee off if ya want and i'll edit this out :spank: :haddock: :bonk: :Pokes:

:machgun: :sniper:

[rsbriggs]

Well, sounds like you are really getting serious about doing the gauntlet. It would (almost) be funny to find out that Win-95 clients are fastest. I can just picture IB downgrading dozens of boxen to Win95.....

:confused: ......Nahh....... Guess not.... :D

[Chinasaur]

IB,

Speaking for AMD boxes (dual and single XP1800's) Mandrake 8.1 always gave me the highest throughput on Phase 1 compared to Win98, later versions of Mandrake, Suse 7.3 to 8.0 and Debian. It's the cleanest, leanest Linux I've ever run and didn't cause problems with the DF build. Even custom compiled kernels under Debian never gave me the structure count that 8.1 did.

[Fozzie]

running NT bar 3 and TBH I think NT kicks XP's arse.

Sorry to say I haven't got figures for you as the NT boxes are different specs to the XP ones.

My work box is NT and my test is XP. The XP box just sits there most of the day crunching DF, no other processes 866 and 256mb RAM. My work box runs Outlook User manager, Server manger SQL based helpdesk software and anything else I need during the day. It's a 733 with 256mb RAm and it keeps up with the other machine.

As far as MicroShaft OSes go I like NT. No whistles and bells, just stable(ish) business like computing.

[rsbriggs]

I've got a stock Barton 2500+ M7NCD Pro MOBO box that I can get you a number of folding benchmarks from this evening. I can set all BIOS settings to normal and use it as a reference box for benchmarking - same hardware - different OSes.

Just one question about it - it is running PC2100 ram. Would it make sense to upgrade to PC2700 or PC3200 before doing benchmarks, or would that affect them at all? Or, would you like benchmarks before AND after upgrading RAM to see what effect that has?

I can get you immediate XP-Pro, Knoppix, Red Hat 9.0, and clusterKnoppix benchmarks tonight. I could get you benchmarks run for nearly any other OS by this weekend. Want me to include mandrake 9.1? Mandrake 8.x? Red hat 8.x ? W2k? Server 2003? Windows 98?

Want any benchmarks other than just several runs of the DF bench mark on each OS ??

No - it's not a problem. It's the least I can do....

[IronBits]

I thank you for that!! Remember, Dyyryath is doing an extensive benchmarking suite, so I'm just looking for a "close enough for me" kinda thing.

I do not want to run win98se if mandrake is going to yield a significant increase in a days output for instance. If there is almost no difference, then I won't bother converting them over. ;)

You should check with Dyyryath and co-ordinate your efforts with his, might help get the BMs done sooner ;)

[bwkaz]

Single-processor Barton (so 640KB cache total) 2500+, running at stock speeds (166FSB, 11x multiplier to put the CPU at 1833MHz). 768MB of PC333, CL2 memory (at least, that's what the mfr. claims, but when I set my BIOS to use SPD, it runs the memory at CL2.5? ... whatever; it's running at 2.5 for this test).


$ ./foldtrajlite -bench
One moment, opening rotamer library...
Predicting secondary structure and generating trajectory distribution...
Folding protein...
Benchmark complete.

Summary
-------
Usr time Sys time
-------- --------
Maketrj 4.420 0.440
Foldtraj 59.690 10.100 This is Linux From Scratch, approximately version 3.3 (though it's had TONS of updates since then). gcc 3.2, glibc 2.2.5, kernel 2.4.20 (ALL of it self-compiled, hence Linux From Scratch ;)).

Interestingly enough, the first time I switched from Mandrake (8.0, I think) to LFS (3.0, with gcc 2.95.3 still), I saw about a 20% improvement in my glxgears speed. I'm thinking that this is because I had compiled my glibc and XFree86 for my specific processor (-march=i686 -O3), rather than for i586 and with -O2 or -O1 (whatever Mandrake uses).

I don't know how much different DF ran, because I didn't run it before the switchover. About the only piece of my upgrade that it uses is glibc, so I have no idea if it would see a huge improvement or not.

BTW -- rsbriggs, my BIOS won't let me run PC2100 memory with my Barton 2500+; it says that the processor FSB cannot be faster than the memory bus (and PC2100 is DDR266, which runs at 133MHz). Maybe yours is different, though, so I don't know. Just a heads-up, something you might want to check on. ;)

[rsbriggs]

Barton 2500+ M7NCD Pro MOBO CPU clock settings at [default]
ram settings aggressive - [turbo] (RAS-5, RCD-2, RP-2, Latency-2.0)

All tests are 10 benchmarks averaged together......



Windows XP-Pro

MAKE 7.734 .332
FOLD 41.375 7.522 Added all up 56.954 (lower is better)

Gentoo (optimized *slightly* for AMD)

MAKE 4.296 .712
FOLD 41.248 9.434 Added all up 55.69


Debian
MAKE 4.40 0.682
FOLD 43.37 11.77 sum 60.226

Mandrake 9.1
(somewhat painful - doesn't recognize nforce2 hardware)

MAKE 3.34 0.60
FOLD 42.94 10.39 sum 57.27





Windows XP-Pro right out of the box is looking pretty good here....

[Angus]

Athlon XP 2600+, A7N8X, 1GB PC3200, aggressive memory settings
dual boot W2K Pro SP3 and Mandrake 9.1
Nothing optimized - straight as it came.

W2k PRO command line version of DF

Make 7.656 .344
Fold 37.500 8.047


Mandrake 9.1 linux-i386-icc version of DF

Make 3.750 .680
Fold 39.200 11.170