Prime95 benchmark conundrum with OC

[mooo]

:QPrime95 says overclocking the FSB is bad!:Q

My system:
XP 1600+ AGOIA F week 12
Epox 8K5A2+
Alpha PAL8045 + Artic Silver 3 + YS Tech adjustable speed fan (turned down low to reduce noise)
Antec Plus660AMG (TruePower330)
Corsair XMS 512MB PC2700 C2

Here are the Prime95 benchmarks for different states:


Raw system (everything at default)
FSB: 133; Memory: by SPD

Best time for 256K FFT length: 27.568 ms.
Best time for 320K FFT length: 34.909 ms.
Best time for 384K FFT length: 44.872 ms.
Best time for 448K FFT length: 50.415 ms.
Best time for 512K FFT length: 56.017 ms.
Best time for 640K FFT length: 72.762 ms.
Best time for 768K FFT length: 89.384 ms.
Best time for 892K FFT length: 104.491 ms.
Best time for 1024K FFT length: 119.342 ms.
Best time for 1280K FFT length: 155.087 ms.
Best time for 1536K FFT length: 186.884 ms.
Best time for 1792K FFT length: 225.368 ms.


Memory tuned
FSB: 133; Memory: 2-3-6-3-1T, 4-way bank interleave (these are spec values for the Corsair module)

Best time for 256K FFT length: 25.784 ms.
Best time for 320K FFT length: 32.946 ms.
Best time for 384K FFT length: 42.573 ms.
Best time for 448K FFT length: 47.968 ms.
Best time for 512K FFT length: 52.483 ms.
Best time for 640K FFT length: 68.617 ms.
Best time for 768K FFT length: 83.359 ms.
Best time for 892K FFT length: 100.806 ms.
Best time for 1024K FFT length: 111.456 ms.
Best time for 1280K FFT length: 147.635 ms.
Best time for 1536K FFT length: 177.864 ms.
Best time for 1792K FFT length: 217.432 ms.

Good these values show an improvement due to tuning the memory timings.


FSB Overclocked
FSB: 166; Memory: 2-3-6-3-1T, 4-way bank interleave (these are spec values for the Corsair module)
166FSB x 10.5 locked.
Vcore = 1.775V (1.79 actual); this is default + 0.025V.
Idle temp = 38~39C; Load temp = 46~47C

Best time for 256K FFT length: 29.900 ms.
Best time for 320K FFT length: 37.955 ms.
Best time for 384K FFT length: 48.344 ms.
Best time for 448K FFT length: 54.902 ms.
Best time for 512K FFT length: 60.634 ms.
Best time for 640K FFT length: 77.519 ms.
Best time for 768K FFT length: 94.854 ms.
Best time for 892K FFT length: 111.338 ms.
Best time for 1024K FFT length: 125.864 ms.
Best time for 1280K FFT length: 168.356 ms.
Best time for 1536K FFT length: 203.553 ms.
Best time for 1792K FFT length: 243.139 ms.

These values are worse than with FSB at 133, with the same memory timings!



What is happening here? Can somebody please explain?

 

[zsouthboy]

Same thing happens to me on my AGOIA Y wk 13...... running a gigabyte 7VRXP rev 1.1

its like the 166 fsb isn't "tuned" if thats the right word

 

[mooo]

^ what?! nobody finds this interesting or comment worthy? :Q

 

[Duvie]

Contact the creator of the program!!! I don't really benchmark it with this program...Just stressing....

I benchmark real world apps, and thay have shown me my ocing has increased my performance...

Sorry no answer!!!

 

[mooo]

my post on the AROIA/AGOIA thread:

Okay, just completed 24hrs of Prime95.

XP 1600+ AGOIA F week 12

166FSB x 10.5 locked.
Vcore = 1.775V (1.79 actual); this is default + 0.025V.
Idle temp = 38~39C; Load temp = 46~47C

Epox 8K5A2+
Alpha PAL8045 + Artic Silver 3 + YS Tech adjustable speed fan (turned down low to reduce noise)
Antec Plus660AMG (TruePower330)
Corsair XMS 512MB PC2700 at 2-3-6-3-1T; VDimm = 2.576 (default)

Lets put a stop to the rumour that the F steppings are duds!

I am confident that I can take the FSB higher, but is there any point in trying to do this? From what I have gathered, there is no need to increase the FSB bandwidth unless the cpu itself is working faster (multiplier). I have also gathered from Anandtech articles that at current multipliers, Athlon XPs do not have any use for the 333MHz FSB bandwidth. Didn't know about this before I installed. If I did, I might have considered unlocking right at the start.

If it is true that higher FSB bandwidth is not of any use, would some please shed light on the point of increasing FSB and computing a value that is really not representative of performance? Yet people seem to be increasing the FSB as far as it will go to get a high MHz/GHz clock speed value.

Also, shouldn't people be comparing benchmark values rather than clock speed? Please see the separate thread that I have created about a Prime95 benchmark conundrum.

OC gurus, please post your thoughts and comments!

 

[chmike]

Also, shouldn't people be comparing benchmark values rather than clock speed?
Yes, you've got it! Obviously Prime95 isn't affected (or, is affected negatively) by increasing the FSB. I benchmark my applications and I see an increase when I up the FSB. That's the golden rule for benchmarks: Use your application, other benchmarks may vary.

From what little I've used P95, it seems to stress both memory and cpu. While a faster FSB should indicate faster memory access, something else may be slowing you down.

 

[mooo]

chmike, i can understand if Prime95 was not affected by increasing the FSB. But affected adversely?! Doesn't make sense, esp, when the FSB is synchronized to the DRAM clock at 166.

agree with using a real world application to benchmark. but isnt the very definition and purpose of a benchmark to provide a measurement standard? it would make no sense to share with you the size of my case in number of palm lengths (my palm lenghts).

perhaps they should stop calling them benchmarks.

 

[Duvie]

I can't give exact number comparison but I know that last one when I was running at default was 100+...so it appears on the p4 front may not be affected....

Here is my score at 1.6@2.53 (158fsb) w/ 1.52vcore w/ 420mhz ddr cas2.5,6,3,3

Best time for 256K FFT length: 9.536 ms.
Best time for 320K FFT length: 12.623 ms.
Best time for 384K FFT length: 15.282 ms.
Best time for 448K FFT length: 18.391 ms.
Best time for 512K FFT length: 20.554 ms.
Best time for 640K FFT length: 26.931 ms.
Best time for 768K FFT length: 32.627 ms.
Best time for 892K FFT length: 40.008 ms.
Best time for 1024K FFT length: 43.270 ms.
Best time for 1280K FFT length: 61.458 ms.
Best time for 1536K FFT length: 77.016 ms.
Best time for 1792K FFT length: 93.882 ms.


I will post 1.6 at 1.6 soon but I definitely remember the last 3 at least being in the 100's like you 1600+...

Maybe you are not stable at that point in either cpu or mem...have you rpime95 tested it thoroughly??? I just got done testing this for 12 hours...no hiccups!!!

1.6a@1.6 w/ 266mhz ddr cas 2,3,3,6 w/ 1.52 vcore

Best time for 256K FFT length: 15.226 ms.
Best time for 320K FFT length: 20.160 ms.
Best time for 384K FFT length: 24.287 ms.
Best time for 448K FFT length: 29.253 ms.
Best time for 512K FFT length: 32.729 ms.
Best time for 640K FFT length: 42.948 ms.
Best time for 768K FFT length: 52.034 ms.
Best time for 892K FFT length: 63.789 ms.
Best time for 1024K FFT length: 69.034 ms.
Best time for 1280K FFT length: 97.588 ms.
Best time for 1536K FFT length: 121.544 ms.
Best time for 1792K FFT length: 149.722 ms.

1.6a@2.66ghz w/ 333mhz ddr cas 2,3,3,6 w/ 1.66v

Best time for 256K FFT length: 9.857 ms.
Best time for 320K FFT length: 12.579 ms.
Best time for 384K FFT length: 15.178 ms.
Best time for 448K FFT length: 18.184 ms.
Best time for 512K FFT length: 20.328 ms.
Best time for 640K FFT length: 26.550 ms.
Best time for 768K FFT length: 32.115 ms.
Best time for 892K FFT length: 39.518 ms.
Best time for 1024K FFT length: 42.682 ms.
Best time for 1280K FFT length: 59.990 ms.
Best time for 1536K FFT length: 75.120 ms.
Best time for 1792K FFT length: 93.043 ms.

So no problem here...It appears to be your system or a gltch with amd capability....

Make sure you set cpu to its correct size before testing...It should give you an error if it thinks you are not running the speed cpu set in there....

I ran at default vcore and oc'd vcore and no issues...I ran memory at diff speeds...I ran at diff cas ratings...

 

[masterxfob]

when running prime95, what should the available memory be set at? i've got 256mb pc2100 ddr and a p4 2.0a. the p4 can hit 2.66 (default voltage, stock hs/f) easy but the ram is holding me back, if i go with aggressive ram settings, it won't even boot.

when i had the available memory set at 8mb it ran fine, but when i upped it to 128 it started crashing within minutes

so what should the available memory be set at?

 

[SupermanCK]

1) make sure you change you CPU clock speed each time you change you FSB
2) available memory depends on how extensive you want to test it...maximum is 400mb i believe...

• 
  <LI>if you have 256mb...you should test 128mb max...most application won't be using 50% of your mem....besides...windows already use up 100mb of it
  <LI>if you have 512mb...you can choose anywhwere from 128mb to 384mb...i set mine to 256mb...50% of what i have for my system

 

[mooo]

Duvie, just looked at benchmarks at mersenne.org and it appears that even the Celerons beat the hell out of the AMD CPUs! :Q

Good thing I didn't look at this before finalizing on my AMD system, or I might have been struggling with second thoughts.

What is happening here? Why are AMD CPUs so bad with Prime95? Can people with knowledge about AMDs strengths please help me defend it?

My numbers are okay but not up to the mark. I did test with Prime95 sucessfully for 24 hrs with no problems.

 

[mooo]

oh i see. it is due to the SSE2 instruction set!

Duvie, how much memory do you have in your system and how much are you having Prime95 use? I used the max that it would allow - 459MB of the total of 512 that I have.

 

[Duvie]

I tested it at a known unstable speed but faster then the fastest I tested and the score still went down....So I don't know why your scores are suffering as you should be increasing performance....

I have 512mb of ram and whether I give 8mb or 256mb it makes no more then .5ms difference.....


It could be sse2 definitely that gives it an edge over celeron, but it doens't xplain that when you increase overall mhz (by ~400mhz) and fsb (+33mhz) you get ~10 percent worse instead....

I tested my 1600+xp work machine on a ecs sis735 mobo with memory at default and noraml timings and I got 232ms for the last one....so not to far off of yours.....

 

[mooo]

Originally posted by: Duvie

I tested my 1600+xp work machine on a ecs sis735 mobo with memory at default and noraml timings and I got 232ms for the last one....so not to far off of yours.....

ya Duvie, it does not explain reduction in performance. i am beginning to think that Prime95 stresses memory more than the CPU and increasing the FSB causes some kind of caching to allow the memory to keep up and this may be the cause of the reduction in performance.

did you happen to test default fsb (133) vs. overclocked (166) for the same (tuned) memory settings on your xp 1600+ system? any difference?

 

[Duvie]

Unfortunately I cannot do that as the 1600+ is on an ecs mobo with no overclocking bios flash.... I can't oc the fsb at all...

We need to get some others of the dozens and dozens of ppl here who have oc'd the 1600+ to test to see if they have similar results...go into the 1600+ official thread and ask there.....

If anything I am surprised as least you are running you memory in synch with the fsb of the chip....in 2 of my test I am not yet I don't have the adverse effects....