Avatar28 said:
I realize that it won't be 100% since the board is probably not 100% efficient at it, though I hear the NForce 2 isn't as good with 1 stick either, so maybe it would.
My real question is more if I had two systems with the same Athlon cpu at the same clock speed using the same mb, etc. The only difference is the RAM. One is using two sticks of PC3200, the other two sticks of PC2700. Would there be any meaningful difference between the performance of the two?
Quick answer:
Assuming both systems have same FSB and CPU, RAM size, ..
The dual channel 2700C2 running 75% ASYNC will lose out by about 10% in 3D Mark compared to the 3200C2 running 100% SYNC. But the 2700C2 will save you 50% on memory cost, or delaying buying new memory.
I have done some detailed analysis and benchmarks, details can be found in
Some analysis and benchmarking for nforce2 dual channel async mode
...
Case 1: Running SYNC, single channel (best score)
FSB = 211 MHz
CPU = 211 x 9 MHz @1.75 V
RAM = 211 MHz, SYNC, 6-3-3-2 (PC3500 C2)
SiSoft 2003 BW = 3071 / 2764 MB/s
3D Mark 2001 SE = 10652
UT 2003 = 59 / 40
Case 2: Up FSB as high as possible, ASYNC with slower/cheaper memory
Result: 3D Mark -9% from best score, almost no hit for UT 2003
Advantages: Use old memory, cheaper memory, BIG saving than using top speed memory, < 10% hit in 3D performance, 50% saving on memory (check the price PC2700 vs PC3200, you'll see)
ASYNC 75%
FSB = 230 MHz
CPU = 230 x 8 MHz @1.75 V
RAM = 172 MHz, 5-2-2-2 (PC2700 tight timing)
SiSoft 2003 BW = MB/s
3D Mark 2001 SE = 9321 (9619 @ 1899 MHz CPU) (-9% from SYNC score)
UT 2003 = 56 / 37 (58 / 38 @ 1899 MHz CPU)
ASYNC 66%
FSB = 222 MHz
CPU = 222 x 8.5 MHz @1.75 V
RAM = 148 MHz, 5-2-2-2 (like using PC2400)
SiSoft 2003 BW = 2641 / 2424 MB/s
3D Mark 2001 SE = 8644 (8698 @ 1899 MHz CPU) (-18% from SYNC score)
UT 2003 = 56 / 37
...
For AMD MB, the SYNC memory bandwidth is the upper bound for that MB, since there are overhead and efficiency involved in running dual channel. I estimate that the efficiency for Dual Channel is around 50% of the difference between FSB and memory speed.
RAM_dual_channel_effective_speed = RAM + (FSB - RAM) x 0.5
Conclusion:
Contrary to many belief that nforce2 MB should be run in SYNC memory mode, the above analysis and data show that it is very cost-effective by runing in ASYNC Dual Channel mode by setting the FSB as high as possible with slower/cheaper memory. Max FSB speed for nforce2 seems to be a matter of "luck", a wide range between 180-230+ MHz. But the result should hold regardless of how high the FSB can be overclocked.
If you want to chase the last drop of game and 3D performance, then you will get the best RAM to match the max FSB and run them in SYNC. One can save 40-60% in memory cost (e.g. PC2700 instead of PC3200, save $100 per 512MB) by trading off 10% of the 3D performance, worst case 20% off. One may not have to buy new memory for a new MB. The data suggests that the 75-80% ASYNC dual channel seems to be the best cost-performance trade-off. You can pick you own tradeoff, 66% may be acceptable too.
For certain games and non-3D applications, such performance hit is almost unnoticeable. E.g. UT 2003 is less sensitive to slower memory in the test. Even with the 10-15% hit in 3D benchmark and game applications, I think it is acceptable, and one can readily get it back by overclocking the video card and CPU.