3DMark2001/03/05 .: vs :. 512MB/1024MB/2048MB - Comparison

[bachus_anonym]

Hey,

I have decided to spent few hours tonight benching Futuremark's 3D benchmark against different RAM configurations. I hope that there will be quite a few of you guys that will find this thread interesting. For some it might not be anything new but maybe for many others it will So, here we go...

Starring:

2 x 256MB Mushkin PC3200 LvL2 (Winbond BH-5)
2 x 256MB G.Skill PC4800LA (Samsung TCCD)
2 x 512MB G.Skill PC4400LE (Samsung TCCD)
2 x 1GB Corsair TWINX2048-4400PRO (Infineon CE-5)
2 x 1GB G.Skill PC4000HZ (Samsung UCCC)

7800GTX 512MB @ 550/1700 (stock), Forceware 81.98, High Quality
(Property of a buddy of mine, waiting to be vmodded...)

AMD Opteron 150
(One of njkid32's toys...)

I chose CPU speed of 3190Mhz (290x11) as this is the MAX 3D stable frequency for UCCC and TCCD. All benchmarks were run in EXACTLY same way: 3DMark2001 first, 3DM05 second, then reboot and 3DM03 last. In comparo between 512MB TCCD, 1GB TCCD and 2GB UCCC @ 8-4-4-3.0, I think I managed to use absolutely same latencies and other timmings for both. That example is really useful to see how amount of memory affects (if any) score is 3D benchmarks.

2x256MB BH-5

3DMark2001 and 05 @ 266x12=3192MHz DDR532 5-2-2-2.0-1T --- 36748 / 9795

3DMark03 @ 266x12=3192MHz DDR532 5-2-2-2.0-1T --- 20894


2x256MB TCCD

3DMark2001 and 05 @ 290x12=3190MHz DDR580 8-4-4-3.0-1T --- 35868 / 9789

3DMark03 @ 290x12=3190MHz DDR580 8-4-4-3.0-1T --- 20748

Unfortunately, I could not get this pair to run @ DDR580 6-3-3-2.5-1T stable enough to pass all benches...


2x512MB TCCD

3DMark2001 and 05 @ 290x12=3190MHz DDR580 8-4-4-3.0-1T --- 36289 / 9785

3DMark03 @ 290x12=3190MHz DDR580 8-4-4-3.0-1T --- 20846

3DMark2001 and 05 @ 290x12=3190MHz DDR580 6-3-3-2.5-1T --- 36900 / 9812

3DMark03 @ 290x12=3190MHz DDR580 6-3-3-2.5-1T --- 20969


2x1024MB UCCC

3DMark2001 and 05 @ 290x12=3190MHz DDR580 8-4-4-3.0-1T --- 36261 / 9814

3DMark03 @ 290x12=3190MHz DDR580 8-4-4-3.0-1T --- 20895


2x1024MB Infineon CE-5

3DMark2001 and 05 @ 266x12=3192MHz DDR532 5-2-3-3.0-1T --- 36396 / 9804

3DMark03 @ 266x12=3192MHz DDR532 5-2-3-3.0-1T --- 20901


So there you go... The only RAM that is missing from this comparison would be 2x512MB BH-5 and 2x1024MB Micron -5B D. Unfortunately, I don't have those sets But I think we could expect 2x512MB BH-5 to score better compared to 2x256MB BH-5 due to Bank Interleave kicking in. A config of 2x512MB BH-5 could very possibly outperform TCCD as well, but not by very much. But I would love to see, how 290MHz 8-3-3-3.0 (Crucial Ballistix) compares to Infineon CE-5 @ 266Mhz CE-5 and TCCD @ 290MHz 6-3-3-2.5-1T. If I manage to get one of those I will update this post

Main goal of this comparison is see what differenences are between 1GB and 2GB configs... All of the scores can be bettered by further "tweaking" of some latencies but I chose not to so this time. I just wanted to show the trend

I will leave conclusions to you guys

Thanks for stopping by

EDIT: I'm not sure which section is better for this thread: X3DTeam or XB...
So, at some point I might move it or not Thanks for understanding...