mersenneforum.org  

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

Reply
 
Thread Tools
Old 2006-03-30, 12:48   #1
forcemaker
 
Mar 2006

5 Posts
Default no speed gain in Prime95's benchmark when overclocking

Hello!

I got a Athlon XP/MP/Mobile machine (a highly modd'ed and stable running CPU) which I change the CPU's multiplier with CrystalCPUID (http://crystalmark.info/download/) between x5 up to x15. (at a system bus speed of 133MHz, AMD's FSB266)

The problem about this is, that most of the CPU benchmarks I ran, do not vary their output. So, if multiplicator is x5 I get the same benchmark results (+/- some fluctuations) as with multiplicator of x15.

I am sure that the frequency has changed after using the Crystal software. You can tell by this:
- Motherboard Monitor tells other frequency
- CrystalCPUID tells changed freq.
- AMD's own CPU info program does tell changed freq.
- some other programs...

To get alway the same results is true for benchmarking with these software products:
- Prime95 (which resuls I can post, FullBench=1)
- SiSoft Sandra
- some others which names I would have to look up.

Different measured speeds are given with this software:
- Crystal mark (same author as CrystalCPUID)
The speeds are as asumed, e.g. x5 makes 500 points, x15 makes 1500points

So, has anybody got ideas? I am a little upset with my SMP system because of this. Some software just won't profit of higher CPU frequencies, even "if they should".

Hope, I could make You understand my little problem...
Just ask me, I will give as much information as possible ;-)

CU - forcemaker
forcemaker is offline   Reply With Quote
Old 2006-03-30, 13:48   #2
Cruelty
 
Cruelty's Avatar
 
May 2005

22·11·37 Posts
Default

For starters: if you have expected a linear increase of your benchmark scores, you were wrong. Apart from increasing the frequency of your CPU the other parts of your system stay at the default values (specifically RAM and FSB), because you are only changing the CPU multiplier.
Anyways, could you post some actual results?
Cruelty is offline   Reply With Quote
Old 2006-03-30, 14:42   #3
forcemaker
 
Mar 2006

1012 Posts
Default

Hi! I understand well, that increasing multiplier by 2 (x5 to x10) will not double the computational power. There are alway other bottle necks and flaws. All clear ;)

The subject is that I do not get ANY advance with a few programs. E.g. in Prime95 (That's actually the reason I am writing in this forum/at this site).

OK, I will post here ONE results.txt from benchmark mode of prime95 as the people in the "Perpetual benchmark thread..." did. I will post only ONE because they all have the same content. (I do a backup each time I do a new bench... just for my convenience.)

The CPU was configured to following values before running each benchmark:
133 MHz FSB * 5 = 666 MHz
133 MHz FSB * 10 = 1330 MHz
133 MHz FSB * 15 = ~2000MHz
Don't get irritated, the displayed CPU speed of 1747.56 MHz is the speed I get ON BOOTING, without modifying anything. And even with this speed, I get alway the same results.
Code:
AMD Athlon(TM) MP 3000+
CPU speed: 1747.56 MHz
CPU features: RDTSC, CMOV, Prefetch, 3DNow!, MMX, SSE
L1 cache size: 64 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
L1 TLBS: 32
L2 TLBS: 256
Prime95 32-bit version 24.14, RdtscTiming=1
Best time for 4K FFT length: 0.160 ms.
Best time for 5K FFT length: 0.240 ms.
Best time for 6K FFT length: 0.291 ms.
Best time for 7K FFT length: 0.354 ms.
Best time for 8K FFT length: 0.389 ms.
Best time for 10K FFT length: 0.529 ms.
Best time for 12K FFT length: 0.639 ms.
Best time for 14K FFT length: 0.779 ms.
Best time for 16K FFT length: 0.851 ms.
Best time for 20K FFT length: 1.118 ms.
Best time for 24K FFT length: 1.346 ms.
Best time for 28K FFT length: 1.652 ms.
Best time for 32K FFT length: 1.783 ms.
Best time for 40K FFT length: 2.411 ms.
Best time for 48K FFT length: 2.870 ms.
Best time for 56K FFT length: 3.551 ms.
Best time for 64K FFT length: 3.919 ms.
Best time for 80K FFT length: 5.138 ms.
Best time for 96K FFT length: 6.114 ms.
Best time for 112K FFT length: 7.393 ms.
Best time for 128K FFT length: 8.108 ms.
Best time for 160K FFT length: 10.744 ms.
Best time for 192K FFT length: 12.933 ms.
Best time for 224K FFT length: 15.708 ms.
Best time for 256K FFT length: 17.149 ms.
Best time for 320K FFT length: 22.264 ms.
Best time for 384K FFT length: 26.759 ms.
Best time for 448K FFT length: 32.775 ms.
Best time for 512K FFT length: 36.100 ms.
Best time for 640K FFT length: 49.692 ms.
Best time for 768K FFT length: 60.521 ms.
Best time for 896K FFT length: 73.988 ms.
Best time for 1024K FFT length: 82.366 ms.
Best time for 1280K FFT length: 105.721 ms.
Best time for 1536K FFT length: 127.027 ms.
Best time for 1792K FFT length: 152.981 ms.
Best time for 2048K FFT length: 171.143 ms.
Best time for 2560K FFT length: 232.959 ms.
Best time for 3072K FFT length: 285.055 ms.
Best time for 3584K FFT length: 345.443 ms.
Best time for 4096K FFT length: 384.863 ms.
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Cannot initialize FFT code, errcode=1002
Best time for 58 bit trial factors: 6.966 ms.
Best time for 59 bit trial factors: 6.973 ms.
Best time for 60 bit trial factors: 6.943 ms.
Best time for 61 bit trial factors: 6.969 ms.
Best time for 62 bit trial factors: 12.416 ms.
Best time for 63 bit trial factors: 12.401 ms.
Best time for 64 bit trial factors: 29.814 ms.
Best time for 65 bit trial factors: 30.349 ms.
Best time for 66 bit trial factors: 31.091 ms.
Best time for 67 bit trial factors: 31.130 ms.
forcemaker is offline   Reply With Quote
Old 2006-03-30, 15:10   #4
Cruelty
 
Cruelty's Avatar
 
May 2005

22×11×37 Posts
Default

I have compared your results to those presented here, and they are near Athlon XP 2200+ (1800MHz).

My conclusion is: you are always running at exact same speed and other applications simply deliver you false information. Maybe you should try changing your multiplier in BIOS? AFAIR Athlon MP CPUs should have an unlocked multiplier.

Try using CPU-Z to verify if the CPU clock and multiplier is actually changing.
Cruelty is offline   Reply With Quote
Old 2006-03-30, 15:23   #5
forcemaker
 
Mar 2006

5 Posts
Default

Hi and thx for answer.

I found an explanation to this phenomena...
http://www.ocforums.com/showpost.php...%3Bpostcount=4

I know about this "alway running at same speed" idea... I had it myself, too. But I even checked the CPU speeds with AMD's own program, with different other tools.

As I imagined, the problem is that Windows creates a timer (or measures or calibrates it) at startup - and there the CPUs clock timer is at 1733MHz.

OK, I will seek for solutions :)

CU all
forcemaker is offline   Reply With Quote
Old 2006-03-31, 19:05   #6
VJS
 
VJS's Avatar
 
Dec 2004

13×23 Posts
Default

Your multiplier looks like it's stuck at 13 which is the highest value you get without cutting some bridges.

Either that or your not changing anything at all using the bios. The speed reported in the client is the speed of the processor not anything else.

If your not convinced use sisoftsandra and do a CPU details. It will show you your multiplier and fsb. That's always correct if you set the multiplier to 15 using software, wiremods, bios, or devine intervention, if sandra says it's a 13x it's 13x no questions.

Quote:
Windows creates a timer (or measures or calibrates it) at startup - and there the CPUs clock timer is at 1733MHz.
Nonsence, it doesn't work that way.

By MP your talking about moblie correct... it sounds like your only doing half the steps b.c you fear cutting the L5 bridge. CUT IT!!! and do the wiremod instead of the software.
VJS is offline   Reply With Quote
Old 2006-04-02, 11:07   #7
forcemaker
 
Mar 2006

5 Posts
Default

Quote:
Originally Posted by VJS
Your multiplier looks like it's stuck at 13 which is the highest value you get without cutting some bridges.
Sorry, no, the start-up multiplier is x13. The CPU has been modd'ed to x13 at start-up, but being able to change by software. Changing is done, in my case, with CrystalCPUID.
Quote:
Either that or your not changing anything at all using the bios. The speed reported in the client is the speed of the processor not anything else.
My BIOS "only" lets me change the FSB speed in 1 MHz steps... actually nice, but it makes my system a little unstable (needed to do more tests).
Quote:
If your not convinced use sisoftsandra and do a CPU details. It will show you your multiplier and fsb. That's always correct if you set the multiplier to 15 using software, wiremods, bios, or devine intervention, if sandra says it's a 13x it's 13x no questions.
SiSoft Sandra tells the "wrong" CPU speed but the "right" multiplier and FSB (the one set by software), but it doesn't benchmark right... as I posted the link before (the overclockers forum), it is (most likely) a software timer problem.
I did measuring of CPU speeds also with MBM (Motherboard Monitor)

The thing is, that also prime95 does read the CPU speed told by Windows. The above link explains the details.

Quote:
Nonsence, it doesn't work that way.
By MP your talking about moblie correct... it sounds like your only doing half the steps b.c you fear cutting the L5 bridge. CUT IT!!! and do the wiremod instead of the software.
Well, in my case I modified two Athlon XP 3000+ (FSB333) Bartons (Model 10) to run on an ASUS A7M266-D mainboard. The result of modification is an MP capable and Mobile-like processor. I followed the instructions found on Fab51.com
  • MP capable is simple: It works by cutting some bridge. The result is that I can use two plain (carefully preselected) processors to run. Saves You more than 50% on buying a real Athlon MP
  • Mobile-like is simple, too: also cutting and/or closing brigdes. The result is that I can change multiplier by software.
forcemaker is offline   Reply With Quote
Old 2006-04-02, 12:28   #8
Cruelty
 
Cruelty's Avatar
 
May 2005

65C16 Posts
Default

There is no explanation to behaviour you describe other than: you are stuck with original multiplier and nothing actually changes.

Also read the manual of this crystalcpuid application:

1. "AMD K6/K7/K8 Multiplier for Power Users. It does not operate in all environments."
2. "Required PowerNow! or Cool'n'Quiet & BIOS & OS Support." - maybe this is a hint?
3. "Support CPU List... K7 : Mobile Athlon XP-M, Mobile Athlon 4, Athlon XP (LV), Mobile Duron (Morgan)" - seems to me that standard XP and MP Athlons are NOT supported.

Additionally, the ASUS board you use should have the ability to change multipliers either by DIP switches / jumpers or BIOS. Check out this review here.
Cruelty is offline   Reply With Quote
Old 2006-04-02, 15:18   #9
forcemaker
 
Mar 2006

5 Posts
Default

It's an ASUS A7M266-D ;-)

Well, just explain me how I get speed-up with certain benchmarks? Tell me why I can measure in some programs like MBM or even AMDs own tool for CPU frequency measurement? Or why is the amount of electricity used (Watt) rising when overclokcing or downclocking...

.... ok, sorry, I got new results now: I got myself VirtualDub, XviD-Codec and a movie. Just did a few experiments like re-encoding the movie. Results were nice: with multiplier x5 I got about 5-10 fps, with multiplier x15 I got around 15 up to 20 fps.

I don't want to stress anymore on this subject, really! It should be clear, where the problem is.

ATM, I am just looking for a possibility to a) recalibrate the Windows timer or b) set CPU frequency before Windows starts.

Sorry for bothering! >_<
forcemaker is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Odd Prime95 benchmark behaviour studeimus Software 1 2017-08-15 14:25
overclocking and prime95 test nik PrimeNet 19 2014-07-25 08:29
NVIDIA Quadro K4000 speed results benchmark sixblueboxes GPU Computing 3 2014-07-17 00:25
Prime95 benchmark question Builder Information & Answers 2 2009-10-25 20:43
Wrong CPU speed reported in mprime benchmark James Heinrich Software 5 2009-06-13 12:56

All times are UTC. The time now is 04:50.


Wed Aug 10 04:50:04 UTC 2022 up 33 days, 23:37, 1 user, load averages: 1.53, 1.12, 1.11

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2022, 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.

≠ ± ∓ ÷ × · − √ ‰ ⊗ ⊕ ⊖ ⊘ ⊙ ≤ ≥ ≦ ≧ ≨ ≩ ≺ ≻ ≼ ≽ ⊏ ⊐ ⊑ ⊒ ² ³ °
∠ ∟ ° ≅ ~ ‖ ⟂ ⫛
≡ ≜ ≈ ∝ ∞ ≪ ≫ ⌊⌋ ⌈⌉ ∘ ∏ ∐ ∑ ∧ ∨ ∩ ∪ ⨀ ⊕ ⊗ 𝖕 𝖖 𝖗 ⊲ ⊳
∅ ∖ ∁ ↦ ↣ ∩ ∪ ⊆ ⊂ ⊄ ⊊ ⊇ ⊃ ⊅ ⊋ ⊖ ∈ ∉ ∋ ∌ ℕ ℤ ℚ ℝ ℂ ℵ ℶ ℷ ℸ 𝓟
¬ ∨ ∧ ⊕ → ← ⇒ ⇐ ⇔ ∀ ∃ ∄ ∴ ∵ ⊤ ⊥ ⊢ ⊨ ⫤ ⊣ … ⋯ ⋮ ⋰ ⋱
∫ ∬ ∭ ∮ ∯ ∰ ∇ ∆ δ ∂ ℱ ℒ ℓ
𝛢𝛼 𝛣𝛽 𝛤𝛾 𝛥𝛿 𝛦𝜀𝜖 𝛧𝜁 𝛨𝜂 𝛩𝜃𝜗 𝛪𝜄 𝛫𝜅 𝛬𝜆 𝛭𝜇 𝛮𝜈 𝛯𝜉 𝛰𝜊 𝛱𝜋 𝛲𝜌 𝛴𝜎𝜍 𝛵𝜏 𝛶𝜐 𝛷𝜙𝜑 𝛸𝜒 𝛹𝜓 𝛺𝜔