What's better, higher FSB or higher multi?

[angry hampster]

I've been playing around with some overclocks on my P35+E6420 setup. Keeping the volts low, I can get 350x8 or 400x7, both good for 2.8GHz. With 400x7, I run the RAM at 800MHz and tighten the timings. With 350x8, I run the RAM at 840MHz and run stock timing (5-5-5-15). The difference between the two is not noticeable, but which one is theoretically better? I'm upgrading to an E8500 in the next few days anyway, but am just curious about my options.

TIA!



Also, if this has been asked before, sorry! I had trouble finding good results with the search function.

 

[DrMrLordX]

Hard to say without benchmarks. Past benchmarks seem to have implied that on Core 2 systems the difference is marginal, though memory straps also came into play on some of those.

I would say you should run a few quick synthetics to see what works best for you. SuperPi is very sensitive to memory bandwidth/latency so that would be a good place to start.

 

[angry hampster]

Sounds good. I'll run 3Dmark's CPU tests, Superpi, and Prime95's benchmarks tonight for kicks to see what I can get. Maybe I'll try tightening down the RAM even more to see what I can do. Should be another fun night of doing nothing productive for school and work. :laugh:

 

[LOUISSSSS]

with the Core 2 systems, that doesn't matter. benchmarks show a small (very small) benefit from higher fst + higher memory speeds, but in real world testing, there is no benefit at all. BUT you WILL wear out your motherboard and ram faster by running them at higher speeds and voltages. You'll likely need to bump up voltages of your FSB + vRam to get those higher speeds.

 

[TurtleCrusher]

I am very interested in this as well. My Q8200 has a modest multiplier (honestly its blows) but i can get a decent FSB for a quad. Certain benchmarks show my oveerclock to be up there with the Q9550 and some even as low as a stock Q6700.

Kind of a bummer when you can't compare my quad to anything..

 

[angry hampster]

Scholzpdx: what is your current FSB on your Q8200?

 

[LOUISSSSS]

393MHz, in his signature. thats not very high for a quad. Most quad can reach 400+, but all FSB readings are related to MB performance. instead of listing his MB in his signature, he's listed his cell phone lol

 

[taltamir]

there is a calculation to determine your ram's bandwidth using the mhz and timings of the ram...
I would hazard a guess that the lower FSB of the ram with the tighter timing will be faster. however it is balanced by the cpu-northbridge link being slower (due to fsb being lower).

overall the difference between the two is insignificant. easiest way to tell is to run some benchmarks.

but in general i prefer higher multi.

 

[angry hampster]

Well I just did some quick benchies, and the results are almost negligible.

Setups:
1. 400x7, DDR2 800 4-4-4-12
2. 350x8, DDR2 840 5-5-5-15
3. 375x8, DDR2 900 5-5-5-14


For SuperPi 1M calculation, I got the following results:
1. 19.313s
2. 19.172s
3. 17.862s



For 3Dmark06 CPU tests 1&2, I got the following CPU scores:
1. 2348
2. 2343
3. 2501




As you can see, the difference between 1 & 2 is less than one percent in both scenarios. As far as I'm concerned, this translates well to real-world usage. I can tell the difference between 3.00GHz and 2.8GHz sometimes, but using lower multis and higher FSB seems utterly pointless unless you're just showing off.


edit: My full system specs are as follows:

CPU: Intel E6420 ES
Mobo: Gigabyte P35-DS3L BIOS rev. f8b (I think)
RAM: 4GB DDR2 PC6400 from G.Skill
Hard drive: WD 320GB 7200RPM
GPU: HIS 4870 512mb

 

[imported_wired247]

Originally posted by: LOUISSSSS
with the Core 2 systems, that doesn't matter. benchmarks show a small (very small) benefit from higher fst + higher memory speeds, but in real world testing, there is no benefit at all. BUT you WILL wear out your motherboard and ram faster by running them at higher speeds and voltages. You'll likely need to bump up voltages of your FSB + vRam to get those higher speeds.

quoted for truth

 

[TurtleCrusher]

Originally posted by: LOUISSSSS
393MHz, in his signature. thats not very high for a quad. Most quad can reach 400+, but all FSB readings are related to MB performance. instead of listing his MB in his signature, he's listed his cell phone lol

Reasoning for the phone is for the phone forums i guess.

Its a Gigabyte EP45-DS3L. LOL... I'm sure i could go higher, but i didn't want to raise volts. I wanted to lower them

 

[TurtleCrusher]

Originally posted by: LOUISSSSS
393MHz, in his signature. thats not very high for a quad. Most quad can reach 400+, but all FSB readings are related to MB performance. instead of listing his MB in his signature, he's listed his cell phone lol

Hey, i got it to 415 and i'm calling it good now.

Is that alright for a quad?

 

[TurtleCrusher]

Originally posted by: Idontcare

Originally posted by: Scholzpdx

Originally posted by: rogue1979
E6400@2816MHz - 8786
E6600@3300MHz - 10,345
E2180@3300MHz - 10,505
Brisbane 4000+ @2.65GHz 6850

I find it odd that the 2180 with 1MB L2 cache can beat the 6600 with 4MB L2 cache at the same speed???????

Like i thought, this benchmarks is FSB hungry.

I ran a test, 200MHz FSB versus 400MHz FSB. This is with QX6700. Same 2.4GHz for both tests (12x200 versus 6x400):

2.4GHz (6x400 Mhz) = 12670

2.4GHz (12x200 MHz) = 11739

So that's an 8% increase in the Nuclearus score for a doubling (2x) of the FSB.

Just pulling up a bench IDontCare did with his Q6700. A higher FSB at the same clock speed should get you higher performance.

These are benchmarks from NuclearMC. I get 16832 with specs in sig.