Does decreasing the multi for higher FSB improve or decrease frame rates in games?

[EastCoasthandle]

For any who lowered you the multi of your CPU to increase the FSB to increase dram frequency /performance have you noticed an increase or decrease in the frame rate of games:
-COD4
-BF2
-Grid
-Assassins Creed
-Company of Heroes: Opposing Forces
-Medal of Honor:Airborne
-and other games



400x9 PL6 = 1793




400x9 PL7 = 1755



450x8 @ 1080 = 1812 using




450x8 @ 1128 = 1857




400x9 @ 1066
Min 26.33 FPS
Max 46.65 FPS
Avg 38.735 FPS



450x8 @ 1080
Min 26.87 FPS
Max 46.68 FPS
Avg 38.775



450x8 @ 1128
Min 27.79 FPS
Max 46.16 FPS
Avg 38.785
Sorry, I forgot to scroll the doc up but the score is found on the benchmark tool itself.


PL7
9x multi
1093 Dram Frequency

All settings are the same as before.
(summary from other posts in this thread)

[CrackerJack]

i got lower scores with benchmarks, when multi higher and fsb lower (stock)16x200

I'll test FEAR and COD4. (There my Favs) I'll post screenshots of FEAR, since it has in game benchmark.

16x200


I keep crashing, when try 13x240 and14x220

[DrPepper]

I've only saw improvements in memory benchmarks, I was doing this last night coincidentally and didn't see a change in games although I'm open to benchies

[erocker]

For games having a higher FSB does yield in higher FPS but it's negligible. It seems to improve memory benchmarks quite a bit however memory benchmark numbers don't translate to FPS in games very much. It is known that the FSB is the bottleneck of Intel systems. I suppose by how much all depends on the hardware.

* I read an article (no idea where) where increasing the FSB gave memory benchmarks around a 20% gain while FPS in games went up 1-3%.

[Mussels]

if you have the same CPU and ram clocks in the end, the scores will be the same.

You should only raise the FSB more if something will end up faster - such as 1:1 overclocking, 400x9 gets you 800MHz, while 450x8 gets you 900MHz for ram. In that case your ram would be faster and have an impact on speed.

If you want FPS increased, OC the video card. thats where the gain is.

[PVTCaboose1337]

I find that I like the multi semi-high, but also a little low, so my ram can fly as well (after all, my D9's need a workout!). I can run a very nice 2.65ghz with a 10x multi and get much better results than a 2.69ghz with multi of 11x.

[EastCoasthandle]

Using a 9x multi for a 400 FSB clocked at 3.60GHz, ram at 1066MHz


Using a 8x multi for a 450 FSB clocked at 3.60GHz, ram at 1080MHz

Photos were consolidated

[Mussels]

thanks for the benches east coast, it backs up what i was saying that overall clocks matter more than FSB clocks.

[EastCoasthandle]

no problem

[Silverel]

I've got Bioshock, CoD4 and Oblivion I'll be testing out.

[CrackerJack]

Originally Posted by Mussels

thanks for the benches east coast, it backs up what i was saying that overall clocks matter more than FSB clocks.

i'll back that up too, i see little to no performance increase either way. I mostly see a slight decrease.

[Wile E]

I've never noticed a difference while gaming, but it seems to help with encoding.

[EastCoasthandle]

I have to wonder the point in it all? Dropping the multi does put you at a disadvantage. In the example I provided it takes a FSB Frequency of 50MHz to regain the 1x multi. That's pretty significant even if your MB can handle it.

What I like to know is it only really worth dropping the multi and doing 450FSB instead of 400FSB if you have DDR2 1200 ram (if your MB can handle it). Then and only then I wonder if there is any substantial gains found?

[Mussels]

the reason its worth it sometimes comes down to the mobo in question.

For example, at 400x9 i get the choices of 800 and 960MHz for ram - i have 1110MHz ram. i'm 'wasting' some speed there.

450x9 gets me 900 and 1200. if i had 1.2GHz ram, that would be the best option - i chose 960MHz with tighter timings since it performed best.

On some systems, you are FSB limited (older chips with really high multipliers, old chipsets) because you need more FSB to use all the bandwidth. Basically, there ARE situations when it helps, but its a minority - you're better off tweaking a lower FSB than aiming for a higher one.

(For the record, many mobos have an FSB strap option, hidden or visible - as you raise the FSB this option changes, and often increases latencies to help you OC, at the cost of memory bandwidth. you have to try and test for yourself what settings are fastest)

[hat]

I noticed a difference in Oblivion when I cranked up my DDR2 533... but then I had an 8500gt.

[Mussels]

oh as a useful bit of advice i forgot: i use superpi to test my changes in this regard. Yes its purely CPU power, but if you notice say, a 5% performance gain, its worth it - if you're OCing a lot higher, needing more volts (and heat) to various components, you can judge if its even worth it or not.

[FR@NK]

Originally Posted by EastCoasthandle

Using a 9x multi for a 400 FSB clocked at 3.60GHz ram at 1066MHz



Using a 8x multi for a 450 FSB clocked at 3.60GHz, ram at 1080MHz





Using a 9x multi for a 400 FSB clocked at 3.60GHz ram at 1066MHz



Using a 8x multi for a 450 FSB clocked at 3.60GHz, ram at 1080MHz

First of all your tRD is set slower at 450FSB. As you should know this will directly effect your ram efficiency. tRD = 6 @ 400FSB would be 15ns and tRD = 7 @ 450FSB is 15.5ns delay. So yea your FSB is higher and your RAM is faster but your MCH is running slower. I think thats why you arnt seeing much of a difference.

Relax your performance level to 7 @ 400FSB and you might see more of a spread on your scores.

Also in CoH it looks like you have vert sync on since your min and max are standard values with a LCD of 60Hz. Although this might just be a coincidence :/

[philbrown23]

I've noticed that with my ati cards the higher the fsb the higher the fps but with nvidia I do not notice that.

[EastCoasthandle]

Originally Posted by FR@NK

First of all your tRD is set slower at 450FSB. As you should know this will directly effect your ram efficiency. tRD = 6 @ 400FSB would be 15ns and tRD = 7 @ 450FSB is 15.5ns delay. So yea your FSB is higher and your RAM is faster but your MCH is running slower. I think thats why you arnt seeing much of a difference.

Relax your performance level to 7 @ 400FSB and you might see more of a spread on your scores.

Also in CoH it looks like you have vert sync on since your min and max are standard values with a LCD of 60Hz. Although this might just be a coincidence :/

The performance is actually better at PL6 not worst. Using a PL6 at 400 does perform slightly better then PL7 based on my setup. There is no reason for me to lower PL from 6 to 7 just to "show" a difference. It's better to use the best setup at 400FSB then the best at 450FSB and review the results. By the same token if I want to decrease delay in the MCH I would need to use PL8 at 450 and so on and so on. It's just not feasible. The MCH is not running slower to a point were performance is hindered. And, I am not using vert sync.

Originally Posted by FR@NK

The MCH is running slower to a point where performance increase is hindered. Based on your bench scores anyways.

Not necessarily true from a stand point of what your ram and northbridge chip is capable of. If you are limited by either then performance increase hindrance becomes forfeit do to the design of the ram or chipset not by the overclocker's decision.

[yogurt_21]

2900 is the bottleneck at the settings I run at, not the q6700. so no matter if I run at 4GHZ or 2.66GHZ, my frames don't change. now load times, benchmarks, and overall desktop speed, thats a different story.

[FR@NK]

Originally Posted by EastCoasthandle

Using a PL6 at 400 does perform slightly better then PL7 based on my setup. There is no reason for me to lower PL from 6 to 7 just to "show" a difference.

Well you could use tRD of 6 for both 400 and 450FSB but you will need ram that can do tCL = 4 @ DDR2-1080. Yea you're right, theres no point in using 450FSB if you have to relax the MCH timings to do so.

Originally Posted by EastCoasthandle

The MCH is not running slower to a point were performance is hindered.

The MCH is running slower to a point where performance increase is hindered. Based on your bench scores anyways.

Originally Posted by EastCoasthandle

Not necessarily true from a stand point of what your ram and northbridge chip is capable of. If you are limited by either then performance increase hindrance becomes forfeit do to the design of the ram or chipset not by the overclocker's decision.

The x38 chipset can run at tRD 6 at either 400 or 450MHz, Its the "overclocker's decision" to relax the timing to 7 when running at 450MHz. If you think about it, why are you raising the FSB? If you are keeping the ram speed nearly the same and the CPU speed the same then whats the point in raising the FSB? The overclocker needs to keep all settings and timings the same to see a linear performance increase in the memory subsystem when comparing 9x400 and 8x450. Otherwise yeah you are right, 400MHz FSB with tighter MCH timings and a lower multi is faster then 450MHz with relaxed MCH timings.

[amd64skater]

Originally Posted by EastCoasthandle

Using a 8x multi for a 450 FSB clocked at 3.60GHz, ram at 1080MHz



Using a 8x multi for a 450 FSB clocked at 3.60GHz, ram at 1128MHz

i know its a bit off topic but what program is that you are using on right with the blue background that tells you all ur info

[OzzmanFloyd120]

Originally Posted by amd64skater

i know its a bit off topic but what program is that you are using on right with the blue background that tells you all ur info

Everest, you can get it at TPU's download section.

[marsey99]

it could be down to the chipsets as i found it to be the other way on my nf650 board, 8x400 was slightly faster in all things than 9x355, but this was with my 4300 when i was tryinng to find the best setup for my 24/7clocks.

not played with this setup that much tbh.

[DanTheBanjoman]

If the final clock is the same the higher FSB will be faster, simply because all other subsystems are faster. If you actually notice this is different per case.