• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Some data about high/low FSB comaprison...

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

dreIU

Member
Joined
Nov 11, 2003
Location
Bloomington, IN
I have conducted a few tests with the same system at the same clock of about 2.5gigs, only with different multiplier and fsb in orde to show the effects of higher FSB.

AMD Mobile Barton 2600+ IQYHA 0351
DFI Lanparty Ultra B
OCZ PC3700EB @ stock timings -->8-2-3-3
ATI Radeon PRO->XT @ 425.25/375.25


166x15 = 2490Mhz (2500 according to windows) 15.5 multi unavailable

Sandra CPU Arithmetic Benchmark: 10,363
3DMark '01 SE: 17,874
SuperPI: 45 seconds
Aquamark03: 43,085 with a Graphics score of 6,090 and a CPU score of 7,368


200x13 = 2600Mhz 12.5 multi unavailable

Sandra CPU Arithmetic Benchmark: 10,782
3DMark '01 SE: 19,517
SuperPI: 40 seconds
Aquamark03: 45,657 with a Graphics score of 6,240 and a CPU score of 8,509


210x12 = 2520Mhz

Sandra CPU Arithmetic Benchmark: 10,200
3DMark '01 SE: 19,500
SuperPI: 41 seconds
Aquamark03: 45,742 with a Graphics score of 6,222 and a CPU score of 8,635

245x10.5 = 2572.5Mhz

Sandra CPU Arithmetic Benchmark: 10,600
3DMark '01 SE: 20,500
SuperPI: 39 seconds
Aquamark03: 47,028 with a Graphics score of 6,279 and a CPU score of 9,371

Conclusion: It appears as if the higher FSB has had very minimal effect on those tests, thus cauusing me to question why on earth I shelled out $345 on PC3700 memory when I could have been satisfied with some PC3200. I would have expected a greater jump in performance with the higher FSB, but it does not appear that way.

Please post your comments.
 
Last edited:
OP
dreIU

dreIU

Member
Joined
Nov 11, 2003
Location
Bloomington, IN
Yuriman said:
It appears to me it helped a lot. Try 166x15.5 and show the results plz. Thats 2573mhz.

Give me a minute and I will. What do you mean helped a lot? The difference is nearly irrelevant. Also, I will post 200x12.5, something that a lot of Mobile users have.
 

jonspd

Super Speedy Senior
Joined
Jan 18, 2003
1K in 3dmark
2 sec in superpi
1,300 in aquamark

IMO thats good enough for me plus you have to count in the fact you lost 50mhz on your clock speed so really it's not as much as it look's so maybe it is just alittle better but still we all live for that 1k more or 1sec faster remember.


Jonspd
 
OP
dreIU

dreIU

Member
Joined
Nov 11, 2003
Location
Bloomington, IN
Updated.... the difference still does not seem great enough to justify having such high fsb. Any comments? Thanks for the input.
 

Yuriman

Member
Joined
Aug 27, 2002
Location
The OCFORUMS
Between 166x15 and 210x12, you gained only 30mhz on the cpu but got nearly 2,000 more 3dmarks and aquamarks. You also dropped 4sec in superpi. That not bad IMO. I would think that a similar percentage increase(20%) would give a similar boost in performance. 20% more than 210mhz is about 250mhz.
 

Docta_Z

Member
Joined
Aug 6, 2002
Location
Canada
since you're not really changing the clock speed ( or in some cases decreasing it) obviously Sandra CPU Arithmetic Benchmark wouldn't show much change since its affected very minimally by memory bandwidth.

The point of raising fsb is to raise this and the overall board bandwidth, so other operations like drawing frames in a game (or 3dmark) can be done more quickly and efficently.

I think this test obviously points out high fsb should be a target one wants to hit.

Keep in mind as well you don't have very much control in your experiment since all the tests have different clock speeds...
 

Zeke968

Member
Joined
Apr 8, 2003
The only real reason I OC is for gaming anyway, so I think your 3dmark increases are pretty good!
 
OP
dreIU

dreIU

Member
Joined
Nov 11, 2003
Location
Bloomington, IN
Well, I tried to keep the clock speeds as close as I could with the multipliers provided by my motherboard. I do not think that anyone who is a half serious overclocker would keep their FSB @ 166, so I am not much looking at that test.

BUT, since the main focus seems to be on 3dMark01, look at the jump from 200fsb and 245fsb (less than 30mhz overall clock difference--2600mhz and 2573Mhz). There is about 1000 3Dmarks difference, does that justify the extra 45mhz on the FSB? I personally would have thought that the increase would have been a little more noticeable. There is also only 1 second difference in superPI (which, from what I understand, is memory bandwith-related)... I mean, what is a second for the extra 45mhz on the fsb???

I just did not expect the increases to be that minimal, that is all.
 

DumpALump

Member
Joined
Jun 24, 2002
Location
California, US
whats the cost difference of pc3700 and pc3200 ram?

Some people here enjoy getting high fsb and clock speeds. Though spending that money on a new 6800GT would have had the best price/performance ratio in games/3dmark. (After selling the 9800 of course).
 

Psykoikonov

Member
Joined
Dec 16, 2003
Location
ON, CA
This exact same thing happened to me in 3DMark2k1 as well. See I get the same 3DMark2k1 @ 200 as I do @ 250fsb....and I think it's because the fsb is waiting for the agp. I get 11500 @ 166MHz or 12500 @ 200MHz or 12500 @ 250MHz.
The Sandra CPU arthimetic has no bearing on fsb only CPU speed, use RAM bandwidth bench instead.
The drop of 6 seconds is impressive (15.4% increase).
I notice the Aquamark CPU test is not very good as it is fsb related.
Also remember that the fsb increase cannot be used as a reference on it's own...a 45MHz gain from 100->145 (45% increaase) is ALOT different than 200-245 (22.5% increase).

Psyko
 

mikapc

Member
Joined
Apr 28, 2003
Location
Maryland
This finding might lead me to instead of buying a gig of 3500 or 3700 memory to just purchase another 512 of 3200 to augment my existing 512 megs of ram on two sticks.
 
OP
dreIU

dreIU

Member
Joined
Nov 11, 2003
Location
Bloomington, IN
autoMATTic said:
Were these test done at a 1:1 ratio?

of course.

mikapc, I am glad you are getting something out of this.

Psykoikonov, what exactly do you mean by "the fsb is waiting for the agp?" I understand what you mean by the percentage increase, BUT I still must argue against all the people that told me to lower the multiplier and up the FSB. Apparently, it doesnt make a valuable difference.
 

zxlr8

Member
Joined
Jan 14, 2004
Location
Powell, WY
Well, I would think you would want the best performance for your particular overclock that your chip can do. It is emphasized because for that speed, high FSB will give you the best performance in games. Thst is the only reason I overclock....I don't know about you.
 

Docta_Z

Member
Joined
Aug 6, 2002
Location
Canada
dreIU said:
\There is about 1000 3Dmarks difference, does that justify the extra 45mhz on the FSB?

Doesn't seem minimal to me. Sounds justified to me.


Besides - this test is run with a single board/cpu/ram combo. Your ram might not even be properly supported by your board...

You've proved that raising fsb improves performance in gaming benchmarks and therefore games as well... this is why a good deal of us raise fsb, to achieve the best performance in games...

I can't see to many other reasons to go pushing fsb...
 
OP
dreIU

dreIU

Member
Joined
Nov 11, 2003
Location
Bloomington, IN
Docta_Z said:
Doesn't seem minimal to me. Sounds justified to me.


Besides - this test is run with a single board/cpu/ram combo. Your ram might not even be properly supported by your board...

You've proved that raising fsb improves performance in gaming benchmarks and therefore games as well... this is why a good deal of us raise fsb, to achieve the best performance in games...

I can't see to many other reasons to go pushing fsb...

What exactly do you mean that my ram may not be supported by my mobo? As far as I know, the chips used in my ram are supported by NF2 boards. please explain this...

Of course the only reason to raise the FSB, or OC at all is to improve performance, all I am saying is that the difference between high fsb/low multi and mid fsb/high multi is not as great as I thought it would be.

EDIT: it appears as though I found a new high... with a LOWER FSB:

236x11 = 2596Mhz


Sandra CPU Arithmetic Benchmark: --
3DMark '01 SE: 20,570
SuperPI: 38 seconds
Aquamark03: 47,191 with a Graphics score of 6,291 and a CPU score of 9,443

If I keep the ram at that lower FSB, I could tighten the timings on that biatch and get even higher scores...(dare I say it?) with a lower fsb.
 
Last edited: