PDA

View Full Version : New Client AMD Fans roar!!!



IronBits
12-11-2004, 10:59 PM
Latest News
v2.2 Client -- Algorithmic Upgrade for All
(posted by louis helm)
Saturday, 11 Dec 2004

George Woltman recently finished work on a new gwmult core. The enhanced IBDWT algorithm that Pentium4 users have been enjoying is now available to all.

AthlonXP's get a 90-110% speed improvement (2x faster!!) Preliminary benchmarks show smaller, but meaningful improvements for older P2/3 and celeron processors.

Everyone should upgrade because there are also fixes for certain large tests (n > 9M) and we want everyone using v2.2 or higher by the time we start testing there.
---------------------------------------------------------------------------------
My AMD XP2600 does 1,000,000+ cEMs/sec
as fast as my new Xeon 3.6s logical cpu.

:|party|: We should be able to really crank em out now! :smoking:

Mustard
12-12-2004, 01:46 AM
In the grand geek world, we have the amd64. What does it do better using, the SSE2 stuff, or the Athlon 2.20 stuff? Anyone done a timing run on this? Just wondering cause I'm not overly impressed..... (or even close to it) with the performance of the amd64 running the SSE2 stuff when compared to a 2.8 533 Celeron. The celeron just walks over the amd64. :(

heh..... guess I should have read a little closer and I would have answered my own question..... :blush:

Mustard
12-12-2004, 02:03 AM
Well, for haha's, I put 1 xp2000, 1 xp2400 and 1 amd64-3200 on the new v2.20 sob client to see how the timing goes. All on linux of course. :)

PY 222
12-12-2004, 04:14 AM
Just threw a Xeon box onto this project.

Lets see how it boils out.

I really like the real time stats in this project. :cheers: :cheers:

em99010pepe
12-12-2004, 04:18 AM
Updating my AMD 1100Mhz. Let's see how fast is it.

And welcome PY 222!

Carlos

****EDITED****

The speed just double! Now is half the speed of my AMD 64 3000+.

IronBits
12-12-2004, 05:49 AM
Originally posted by PY 222
Just threw a Xeon box onto this project.

Lets see how it boils out.

I really like the real time stats in this project. :cheers: :cheers: Welcome aboard! :D
I just put my Dual Xeon on Suse 9.1, let's see how that runs :)

1,102,000 cEMs/sec on my AMD64 3200+ running on XP64 Beta 1st release.

Mustard
12-12-2004, 12:51 PM
Just a little informational post on the v2.20 client on various systems. After running overnight, there are 37 blocks difference between my amd64-3200 and my xp2400, both running the 2.20 client. Not much difference really. We seriously need some 64-bit clients for the DC world. Big price spread between the amd64 and the xp2400. For 37 blocks once could pick up multiple xp2400's and be bucks (and points) ahead of the game with the present state of affairs.

54 blocks, or one percentage point separate the xp2000 and the amd64-3200 running over roughly the same time frame. Meanwhile the celeron is crunching along putting out a block about every 2min 18 sec.

So the intel stuff runs super great, the regular old XP's run great, and the amd64's are a little disappointing. But the 2.20 client seems pretty good! :)

em99010pepe
12-13-2004, 02:40 AM
Just to say that for the next week my SOB production will decrease.

Thanks,

Carlos

CaptainMooseInc
12-13-2004, 02:52 AM
Damn you people! Now I want to run -another- DC project and I have no spare computers! Damn you all!

I already run FaD and have D2OL installed to run whenever FaD isn't!

No fair! I like SPEED increases. Fast things make me want to jump on board!

Damn you all! :moon:

:thumbs:

-Jeff

IronBits
12-13-2004, 11:07 AM
I finally got Windows 2003 SP1 XP64 Professional running on the network.
The Dual Xeon 3.6 puts up almost 5 million cEMs/sec
all four clients are showing 123#### cEMs/sec
Like getting an extra processor for free. :D
:cheers:

magicfan241
12-13-2004, 01:05 PM
My sempron runs at 666k when I'm using it.

I just missed the screenshot of the computer running at 666,666.6 CEM/s

I soo wanted that screenshot.

magicfan241

IronBits
12-13-2004, 08:33 PM
:rotfl: /me waits for PIC ! :D

What are you two up to now?
:Pokes: em99010pepe
:Pokes: FDC_Lexx
:bonk:

PY 222
12-14-2004, 01:50 AM
A little offtopic here but I can't seem to be able the answer to my question.

Anyway here it goes. I've got a P4 2.8Ghz HT running RH9 and I've just installed the new 2.2 client in one folder. Question is that do I need to run two clients in two separate folders or can I install it like a Windows service with the "-o2" flag to fully utilize the HT?

Is there such a thing in Linux?

Mustard
12-14-2004, 02:06 AM
Install it in a separate folder for each instance you want to run, I put mine in sob1 and sob2 on a HT system.... :)

And if you want to let it run til hell freezes over and log out of the system, just

./sb sclient.conf &

and it will run and run and run..... like that bunny rabbit. :)



IB -- re what am I up to? I'm in a mad crunch-off trying to get some numbers up in another project. Athlons still aren't made for this in my book..... just too slow. But I'll get the tests crunched out so that the points aren't lost. :)

PY 222
12-14-2004, 02:09 AM
I like the way you put it Lexx.

Thanks, I'll try to let the Bunnies keep on goin... :D

em99010pepe
12-14-2004, 02:41 AM
PY 222,

If you have dedicated crunchers run SOB at normal priority. Extra boost.

IronBits,

I told my production would decrease.

Carlos

em99010pepe
12-15-2004, 02:35 AM
Last night there was a power failure and now I can't restart my AMD 1100Mhz. :cry:
The AMD 64 is running without a problem.

Carlos

IronBits
12-15-2004, 07:01 AM
Check the power supply... :bang:

em99010pepe
12-15-2004, 07:11 AM
Originally posted by IronBits
Check the power supply... :bang:

I was out all day but my father managed to fix the problem. He told me he had to shutdown SOB or as he said "the program 17". Don't ask me why but now the computer is OK running again SOB.:D

Carlos

fivemack
01-07-2005, 07:54 PM
Originally posted by IronBits
I finally got Windows 2003 SP1 XP64 Professional running on the network.
The Dual Xeon 3.6 puts up almost 5 million cEMs/sec
all four clients are showing 123#### cEMs/sec
Like getting an extra processor for free. :D
:cheers:

Not sure this is quite the right way to go; I'm getting 1.5M with one client on a Northwood 2400. So I think your four clients are each getting half a processor; try running only two, preferably with processor affinity to put one on each CPU, and I'd expect it to go even faster.

Chinasaur
01-07-2005, 11:46 PM
Actually sounds good. Multi-processor overhead is a tax we all (dually owners) have to pay :)

Cache coherency, the scheduler and application not coded for multi-threading contribute to this. There are other factors I'm sure i don't know about.

Still, I'd rather run a dual system than not. Once you felt the power of two (or more) processors at work on your toughest stains...you'll never go back to a single agitator ;)

:cheers:

fivemack
01-08-2005, 06:54 AM
Originally posted by Chinasaur
Actually sounds good. Multi-processor overhead is a tax we all (dually owners) have to pay :)

Cache coherency, the scheduler and application not coded for multi-threading contribute to this. There are other factors I'm sure i don't know about.

Still, I'd rather run a dual system than not. Once you felt the power of two (or more) processors at work on your toughest stains...you'll never go back to a single agitator ;)

:cheers:

I'm not talking about dual versus single CPU; I'm talking about using both halves of an HT processor, which I don't believe is the right way to achieve extra performance when the underlying code is as optimised as George Woltman's FFTs: there aren't the spare issue slots from thread 0 to be filled with instructions from thread 1.

eg

1 copy runs at speed N
2 copies run at total speed 1.8N because of memory contention
4 copies run at total speed 1.6N because of memory and HT contention

SoB 2.3.0 is not coded for multi-threading; otherwise you'd only have to run one copy on a dual-processor machine.