mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware

Reply
 
Thread Tools
Old 2004-01-19, 22:12   #1
Matt_G
 
Jan 2004
Dacono, CO.

23 Posts
Default Prime 95 and Software OC'ing

Just built a new box with a P4 3.0D and a Gigabyte 8KNXP rev. 2 mobo. This is a 875 chipset board.
Due to the way Gigabyte has done their BIOS, if you go past 210FSB in the BIOS, PAT is disabled. (210 FSB=PAT on, 211 FSB= PAT off)

The only way around this is to overclock from Windows using software such as ClockGen or SetFSB. By using software to overclock this board PAT stays enabled.
The problem with this is Prime doesn't recognize a Windows based OC. It must take it's readings directly from the BIOS.

I have got this box running stable at 250MHz FSB (3.0@3.75) using ClockGen but my iteration times from Prime do not change from stock. I get .041ms at 200FSB and .041 with this OC using software.

Is there a way to get Prime to realize the CPU is overclocked?
Matt_G is offline   Reply With Quote
Old 2004-01-19, 22:24   #2
PrimeCruncher
 
PrimeCruncher's Avatar
 
Sep 2003
Borg HQ, Delta Quadrant

10101111102 Posts
Default

I would expect some increase but I suppose this could theoretically be a memory bandwith limit. Prime95 goes out of its way to use up every last CPU cycle. Can you check to see if CPU utilization is at 100% while Prime95 is running?
PrimeCruncher is offline   Reply With Quote
Old 2004-01-19, 23:04   #3
Matt_G
 
Jan 2004
Dacono, CO.

23 Posts
Default

Quote:
Originally posted by PrimeCruncher
I would expect some increase but I suppose this could theoretically be a memory bandwith limit. Prime95 goes out of its way to use up every last CPU cycle. Can you check to see if CPU utilization is at 100% while Prime95 is running?
Task Manager shows CPU0 at 100%. (Hyperthreading ya know....)
I don't think it's memory bandwidth either. With the comp at stock settings I get roughly 4700MB/s buffered per Sandra 2004.

With the OC I get roughly 5250MB/s. It's Corsair TWINX1024-3200LL RAM. I'm using the 1.66 multiplier so this DDR400 is running at 415MHz. It is running in dual channel. Why Gigabyte used multipliers instead of dividers is beyond me.

If I overclock thru the BIOS Prime shows the OC'd speed of the processor and iteration times go down accordingly. It just won't see this software based OC. CPU settings show it as a 3.0GHz while OC'd with software. It acts like it won't use more than it thinks it has!

Oh and thanks for the reply.
Matt_G is offline   Reply With Quote
Old 2004-01-19, 23:46   #4
dsouza123
 
dsouza123's Avatar
 
Sep 2002

2·331 Posts
Default

Maybe the software OC also disables/negates PAT.
Is there any overhead using the soft OC ?

Is PAT always enabled upto BIOS 210 fsb ?

Have you tried other programs that do large amounts of calculations like SuperPi with the different combinations ?

test with
BIOS stock, no soft OC
BIOS stock, soft OC

BIOS OC'd 210, no soft OC
BIOS OC'd 210, soft OC

BIOS OC'd 211, no soft OC
BIOS OC'd 211, soft OC

What program tells you that you are getting 3.75 Ghz with soft OC ?

If it is done by timing some code then if it is really 3.75 Ghz then it will be reflected.

If it is done by checking some register/setting for FSB speed and multiplier then doing a simple calculation, it could be tricked into saying a false higher clock speed.

It is very unlikely that Prime95 would throttle itself down to the BIOS clockspeed.
dsouza123 is offline   Reply With Quote
Old 2004-01-20, 00:02   #5
geoff
 
geoff's Avatar
 
Mar 2003
New Zealand

13·89 Posts
Default

I don't think Prime95 can or needs to know whether the CPU is overclocked. Have you tried doing the timing with a wristwatch for comparison? If 1000 iterations take 41 seconds then either the software overclock has not taken effect or maybe the CPU is throttling.

Try doing some trial factoring at stock and overclocked, the 'iteration' time should change linearly with clock frequency regardless of any memory settings. If it doesn't change then the software overclock has not taken effect.
geoff is offline   Reply With Quote
Old 2004-01-20, 01:44   #6
Matt_G
 
Jan 2004
Dacono, CO.

23 Posts
Default

dsouza123 and geoff, you both bring up some good tests to do.

I will run them all and get back to you with the answers, but it might be the weekend before I find the time to do them all.

One test I did run though, shows that the soft overclock is 'taking effect'. Used SuperPi to do this. Here's the results:

Stock setup:
All BIOS settings at default (200FSB and memory at 2.0 multiplier) CPU-Z shows the processor speed as 3014, Memory at 402MHz
Calculate to 1M digits = 47 sec.
Calculate to 2M digits = 108 sec.

Soft OC:
BIOS is set to 200FSB but I change the memory multiplier to 1.66 so my DDR400 will run....
Use ClockGen to set the FSB to 250 which puts my memory at 400MHz. CPU-Z shows the processor speed to be 3750.4 and memory at 400.2MHz
Calculate to 1M digits = 38 sec.
Calculate to 2M digits = 89 sec.

Prime95 runs iterations of 22594xxx at 41ms under both setups.
Matt_G is offline   Reply With Quote
Old 2004-01-22, 17:42   #7
delta_t
 
delta_t's Avatar
 
Nov 2002
Anchorage, AK

3×7×17 Posts
Default

Try running the built in benchmarking on Prime95 with and without the software overclock and see if it's working. You could compare the timings to similar computers posted here: http://mersenneforum.org/showthread.php?t=59
It does seem strange, but if the CPU is truly being overclocked to higher frequencies then the benchmarking results should show this out.

Last fiddled with by delta_t on 2004-01-22 at 17:43
delta_t is offline   Reply With Quote
Old 2004-01-22, 19:50   #8
E_tron
 
E_tron's Avatar
 
Sep 2002
Austin, TX

23116 Posts
Default

Quote:
Originally Posted by Matt_G
I have got this box running stable at 250MHz FSB (3.0@3.75) using ClockGen but my iteration times from Prime do not change from stock. I get .041ms at 200FSB and .041 with this OC using software.

Is there a way to get Prime to realize the CPU is overclocked?
I remember the old versions of the software (v20 v21). You had to tell it what kind of processor you had and how fast it was. One time i told it that my Athlon was a 5.5ghz processor and it showed iteration timings of a 5.5ghz Athlon, however it only performed the speed of a 1.3ghz athlon.

I don't like overclocking through an OS, because most software will not detect the new cpu speed. Could you flash the BIOS on your system? Usually a bios flash to the latest ROM will remove most problems.
E_tron is offline   Reply With Quote
Old 2004-01-24, 00:58   #9
markhl
 
Apr 2003
California

5C16 Posts
Default

[QUOTE=E_tron]I remember the old versions of the software (v20 v21). You had to tell it what kind of processor you had and how fast it was.QUOTE]
Good point. Matt_G, what version of Prime95 are you using? Can you upgrade?
markhl is offline   Reply With Quote
Old 2004-01-24, 03:02   #10
Matt_G
 
Jan 2004
Dacono, CO.

23 Posts
Default

Quote:
Originally Posted by markhl
Good point. Matt_G, what version of Prime95 are you using? Can you upgrade?
I'm using version 23.7.1.
I still haven't had time to run the other tests. Maybe tomorrow.

What I don't understand is the findings I posted earlier and have reposted and added too below. Any comments on this?

Stock BIOS settings, No soft overclock.

SuperPi results:
Calculate to 1M digits = 47 sec.
Calculate to 2M digits = 108 sec.

Prime95 results running my current exponent. (22594xxx)
.041 seconds per iteration.

Benchmark results from Prime95:
Best time for 384K FFT length: 11.817 ms.
Best time for 448K FFT length: 14.046 ms.
Best time for 512K FFT length: 16.030 ms.
Best time for 640K FFT length: 19.196 ms.
Best time for 768K FFT length: 23.308 ms.
Best time for 896K FFT length: 27.664 ms.
Best time for 1024K FFT length: 31.038 ms.
Best time for 1280K FFT length: 40.866 ms.
Best time for 1536K FFT length: 50.019 ms.
Best time for 1792K FFT length: 59.420 ms.
Best time for 2048K FFT length: 67.298 ms.

Stock BIOS except using 1.66 memory divider instead of 2.0. Soft overclocked to 250FSB. Processor runs at 3750.4 GHz and memory is at 400.2 MHz.
SuperPi results:
Calculate to 1M digits = 38 sec.
Calculate to 2M digits = 89 sec.

Prime95 results running my current exponent. (22594xxx)
.041 seconds per iteration.

Benchmark results from Prime95:
Best time for 384K FFT length: 11.914 ms.
Best time for 448K FFT length: 14.187 ms.
Best time for 512K FFT length: 16.143 ms.
Best time for 640K FFT length: 19.294 ms.
Best time for 768K FFT length: 23.475 ms.
Best time for 896K FFT length: 27.780 ms.
Best time for 1024K FFT length: 31.200 ms.
Best time for 1280K FFT length: 41.118 ms.
Best time for 1536K FFT length: 50.301 ms.
Best time for 1792K FFT length: 59.767 ms.
Best time for 2048K FFT length: 67.639 ms.

This shows (The SuperPi results) that the softoverclock is indeed in effect, yet Prime doesn't use it.
In fact it was a little slower under the soft overclock.
This just doesn't make any sense to me at all.

Last fiddled with by Matt_G on 2004-01-24 at 03:06
Matt_G is offline   Reply With Quote
Old 2004-01-24, 04:16   #11
sdbardwick
 
sdbardwick's Avatar
 
Aug 2002
North San Diego County

13·53 Posts
Default

Here's my WAG on the situation:
Software overclocking also speeds up whatever reference Prime95 uses for determining iteration time. So, Prime95 report the same per iteration time, while "real world" time is in fact reduced. SuperPI might use another clock for reference, so it accurately shows the increase in speed.

Perhaps some manual timing with a stopwatch (of say 5K or 10K iterations) would reveal the answer, since the discrepancy between 3.0 and 3.7 GHz should be detectable even with such a crude method.
sdbardwick is online now   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Prime testing software suggestions please. ishkibibble Conjectures 'R Us 15 2013-03-14 08:41
Prime Factoring/Calculating Software antimath Software 20 2012-01-05 08:24
Best Software? Unregistered Homework Help 0 2008-10-07 01:00
[SunOS 5.10] Software for prime search pacionet Programming 3 2008-02-12 12:36
Network Administration software for Prime ? fuzzfuzz Software 6 2002-09-10 08:46

All times are UTC. The time now is 18:48.


Fri Oct 22 18:48:19 UTC 2021 up 91 days, 13:17, 0 users, load averages: 2.04, 1.74, 1.56

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.