DRAM:FSB Ratio Test

antas

Distinguished
Feb 22, 2008
274
0
18,780
Hi all,

I was doing a little experiment to compare a different DRAM:FSB ratio to see the performance difference just for the sake of my curiosity. Actually I would like to post this @ Kraynor threads:

Which ram:fsb ratio will see best performance

But somehow I couldn't make a new post in this thread. So I decided to post a new one. I'm really curious about this issue, but never had time to do real test of my own, until last night :) OK, here we go:

specs
Processor Intel E6750 @2.66 GHz
Memory Corsair XMS2 PC2-6400C4 2x1GB
Mobo ABIT IX38
VGA ATi HD3870 GDDR4 512MB (Gigabyte)

test configurations
This is the test configurations (reading from CPU-Z v1.43). I was only playing with the DRAM frequency & ratio, as it is the purpose of this test, to see the effect of DRAM freq & ratio in reality. The processor is set to stock speed @2.66 GHz


Test no DRAM(MHz) FSB : DRAM CL tRCD tRP tRAS CR DRAM Voltage
1 400 5 : 6 4 4 4 18 2T 2.1
2 400 5 : 6 5 5 5 18 2T 1.8
3 333 1 : 1 5 5 5 18 2T 1.8
4 408 5 : 6 4 4 4 18 2T 1.8

*note:
All using default multiplier (x8).
number 1: I set the tRAS at 12 at BIOS (according to Corsair XMS2 spec), but according to CPUZ it's 18.
number 1-3: The FSB is set to 333 MHz, achieving proc speed at stock 2.66GHz.
number 4: This is the default (auto) setting of the mobo, the FSB rate is set to 340MHz instead 333. The proc speed is 2.72GHz, a little higher than stock.

Benchies
I used Sisoft Sandra XII for benchies, specifically at Memory bandwith and Processor to Memory latency. Also I tested the config using EZ video converter (MPEG to AVI) to convert 108MB files (10mins length).

Results

Test no Mem Bandwith Mem Latency Video conversion (m:s)
1 6137 86ns 1:26
2 6012 95ns 1:26
3 5738 96ns 1:27
4 6251 87ns 1:23

note:
- mem bandwith: higher better
- mem latency: lower better
- vid conversion: quicker better

Temporary Conclusions
From the result, out of no 1-3, as I expected the best result is when I set the RAM at 400MHz (effective DDR2 800MHz) with 4-4-4-18(12) timing. The difference not too much at the same DRAM speed with 4-4-4-18 and 5-5-5-18. However this is confirmed my belief that the best configurations is the faster DRAM speed with the tighter timing. It's really not necessary to achieve 1:1 ratio for best performance. In this case we have the 5:6 ratio perform better than 1:1 ratio. Even, it's proven that higher mem speed do help raising the performances. Although I also believe that there is an 'optimal' settings, not necesarilly the highest speed. It must be remembered that all of this at stock speed. If we're talking a 1:1 ratio at 400MHz, that's a different story. Since it means OC FSB, which of course will increase the performance even further. As it's shown in figure no. 4.

The no.4 is showing the highest result out of 4, but this is achieved because it's a little OCed from stock speed. So I didn't really count on no. 4 result as the fair comparison to others. On number 4, the FSB is pushed to 340MHz instead of default 333MHz. For the fair comparison I used only the 333MHz FSB (no 1-3). But I would like also to show the no 4 since it's a MOBO default settings.

I have not experimented yet with OCing memory, pushing out of it's stock speed (400MHz). In my personal opinion, there is an optimal settings that we can achieve. But at the moment I'm content with default speed, as proven at point no 4 (default mobo settings) was the best result. This ABiT mobo for some reason OCing a little bit out of it's stock (from 400MHz to 408MHz DRAM, and from 333MHz to 340MHz FSB). Which is look good to me. I don't bother to OC as I don't feel I need it. And the other reason is my casing circulation is not really good at the moment.

Additional note
For you who might confused with the FSB & DRAM speed terms:
I wrote here all the actual speeds. My FSB is 1333MHz, which is actual FSB speed is 1333/4 = 333MHz.
My RAM is PC2-6400 or DDR2-800, with actual RAM speed is 800/2 = 400MHz.

ps: I love this ABiT & CORSAIR combo, it's really easy to set and be experimented with.

Please share your thoughts, as I understand that this is a really basic benchmarks. I didn't count on an overall performance using other applications or in actual gaming. Since it would consume a lot of time, which is not my luxury at the moment :) Having to do around 30 mins for each configurations and testing is sucky :) Meh ..
 

kad

Distinguished
Feb 29, 2008
524
0
18,980
There is an excellent test of this issue in anandtech in the review of Asus Formula Rampage
And it is true 1:1 is not the best always

Good work and thanks for sharing info
 

antas

Distinguished
Feb 22, 2008
274
0
18,780
Wow, should've read that review before, so I don't have to go all through the work :(

Could you kindly share the link for the article you mentioned? I couldn't find it, but I really want to read that one.

Thanks :)
 

TRENDING THREADS