which gives better performance [overall speed vs higher fsb]

[Wigwam]

running a system at 12x166 [ram in synch at 6-2-2-2]
or running 9.5x200 [ram in synch at 11-2-2-2]

the first is 50MHz faster overall speed but in the latter case the whole system is running on a faster bus so which would give me better overall performace

 

[myocardia]

With only a 50mhz difference in processor speed, 9.5x200 would be noticeably faster.

 

[WA261]

FSB is where your performance is at

 

[saxophonoia]

Originally posted by: WA261
FSB is where your performance is at

Yup.

 

[suklee]

How bout at the same overspeed for a Barton 2500+

200 x 11 = 2.2 GHz @ 2-3-3-11

or

220 x 10 = 2.2 GHz @ 2.5-3-3-6

Which should be faster? Will the CL2 make a lot of difference?

 

[myocardia]

Originally posted by: Kai920
How bout at the same overspeed for a Barton 2500+

200 x 11 = 2.2 GHz @ 2-3-3-11

or

220 x 10 = 2.2 GHz @ 2.5-3-3-6

Which should be faster? Will the CL2 make a lot of difference?

Well, I can tell you that 2.5-3-3-11 will be faster than 2.5-3-3-6. But, those two choices aren't as clearly evident as
Wigwam's were. A higher fsb is almost always better, but it can depend on the application being used. Tell you what, why don't you test both, and let us know?

 

[TerryMathews]

The lower latency will speed up multiple small memory transfers. Most of what you do on a computer would be considered small transfers (IE, OE, small gaming, MP3s, etc). If you could get that 200MHz stable at 2-2-2-11, you'd be better off. But if we're really only comparing 2.5-3-3 to 2-3-3, take the extra 20MHz, it'll probably be faster overall. That .5 CAS cycle won't really make a difference.

 

[suklee]

I run 220 x 10 stable @1.7V, but sometimes my Shuttle doesn't boot up at that FSB... takes one or two resets to do the job.

Last night I switched it down to 200x11 @2-3-3-11 (couldn't boot @2-2-2-11) , and it's strange but loading explorer windows felt faster maybe it's all psychological.

What utils should I be using to benchmark in my case?

 

[myocardia]

There are quite a few benchmarks out there. If you don't game much, try using SuperPi 1m and 2m tests, the benchmarking utility in Prime95, and PCMock 2002. That should be enough to let you see which is faster in your case.

 

[Shimmishim]

in general higher fsb is better

but then again, it all depends on the application

some programs are about pure clockspeed while others are about bandwidth...

 

[suklee]

FSB200 2-3-2-11 (best I could do)

SuperPI -> 1M
47 seconds

Prime95
AMD Athlon(tm) XP 3200+
CPU speed: 2204.99 MHz
CPU features: RDTSC, CMOV, PREFETCH, 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 version 23.8, RdtscTiming=1
Best time for 384K FFT length: 26.695 ms.
Best time for 448K FFT length: 30.530 ms.
Best time for 512K FFT length: 32.884 ms.
Best time for 640K FFT length: 44.140 ms.
Best time for 768K FFT length: 52.499 ms.
Best time for 896K FFT length: 63.500 ms.
Best time for 1024K FFT length: 70.101 ms.
Best time for 1280K FFT length: 91.941 ms.
Best time for 1536K FFT length: 109.933 ms.
Best time for 1792K FFT length: 133.199 ms.
Best time for 2048K FFT length: 148.034 ms.

 

[suklee]

FSB220 2.5-3-3-11

SuperPI -> 1M
47 seconds

Prime95:
AMD Athlon(tm)
CPU speed: 2204.82 MHz
CPU features: RDTSC, CMOV, PREFETCH, 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 version 23.8, RdtscTiming=1
Best time for 384K FFT length: 26.334 ms.
Best time for 448K FFT length: 30.139 ms.
Best time for 512K FFT length: 32.444 ms.
Best time for 640K FFT length: 43.623 ms.
Best time for 768K FFT length: 51.881 ms.
Best time for 896K FFT length: 62.841 ms.
Best time for 1024K FFT length: 69.287 ms.
Best time for 1280K FFT length: 90.773 ms.
Best time for 1536K FFT length: 108.631 ms.
Best time for 1792K FFT length: 131.454 ms.
Best time for 2048K FFT length: 146.281 ms.

 

[suklee]

Here are some more results using PCMARK 04

200FSB @2-3-3-11
3519 PCMARKS
CPU 3619
Mem 2692

220FSB @2.5-3-3-11
3588 PCMARKS
CPU 3639
Mem 2826


Guess I'll stick to 220FSB for now

 

[DAPUNISHER]

Originally posted by: Shimmishim
in general higher fsb is better

but then again, it all depends on the application

some programs are about pure clockspeed while others are about bandwidth...

succinct yet encompassing :beer:

 

[myocardia]

Originally posted by: DAPUNISHER

Originally posted by: Shimmishim
in general higher fsb is better

but then again, it all depends on the application

some programs are about pure clockspeed while others are about bandwidth...

succinct yet encompassing :beer:

Punisher, if you keep using words like those, you'll have to change your icon.

 

[DAPUNISHER]

Originally posted by: myocardia

Originally posted by: DAPUNISHER

Originally posted by: Shimmishim
in general higher fsb is better

but then again, it all depends on the application

some programs are about pure clockspeed while others are about bandwidth...

succinct yet encompassing :beer:

Punisher, if you keep using words like those, you'll have to change your icon.

Why? Gorillas are smarter than I am!

 

[Jhatfie]

With my A64 3000+ I noticed almost no difference between running 245x10 or 272x9. PCMark04, PCMARK02, 3DMARk01 and 03, Sandra, Super Pi all were within 1%. So in this respect I'd go for outright clock speed rather than a ultra high FSB (or HTT as A64's go). For that matter, running the memory 1:1 or 5:6 made no difference as well. Now if I remember right, with my XP 2100+, the differences were a bit greater with 3DMark03 liking the higher FSB especially. Generally the higher FSB and memory speeds when running 1:1 would overpower any small clock increase.

 

[Wigwam]

hhmmm.
i guess 1900mhz at 9.5x200 it is then...[which suits me as i have been itching to justify running it at that fsb!]

now maybe if i can get the piece of crud to run at 10x200 this might have been a moot discussion...oh well!:disgust:

 

[Shimmishim]

Originally posted by: Jhatfie
With my A64 3000+ I noticed almost no difference between running 245x10 or 272x9. PCMark04, PCMARK02, 3DMARk01 and 03, Sandra, Super Pi all were within 1%. So in this respect I'd go for outright clock speed rather than a ultra high FSB (or HTT as A64's go). For that matter, running the memory 1:1 or 5:6 made no difference as well. Now if I remember right, with my XP 2100+, the differences were a bit greater with 3DMark03 liking the higher FSB especially. Generally the higher FSB and memory speeds when running 1:1 would overpower any small clock increase.

that's interesting....

running a superpi 1M test running at 2500 mhz (10 x 250 with 1:1 ratio with 2-2-2-6 timings) yields 35 seconds.

running a superpi 1M test running at 2550 mhz (10 x 255 with 5:6 ratio with 2-2-2-5 timings) yields 36 seconds.

50 mhz gain = 1 second loss in super pi.

I don't think that running 1:1 "overpowered" the small clock increase.

I'll run a superpi 1M test at 2600 mhz (10 x 260 with 5:6 with 2-2-2-6 timings) to see what sort of result I get.