Memory and FSB in Sync?

Ok my since my pc2100 memory runs at 133mhz is it better to have my FSB at the same speed? I currently have it set to 154 but if the memory is being a bottleneck is it really worth the high temps?

Are you using an FSB memory ratio other than 1:1? If not your memory is running at 154 just like your fsb

Always run in Sync... no exceptions.

What about with an NFORCE with dual channel DDR? Since the memory bandwidth far outstrips the processor's, it shouldn't be an issue to run your memory slower than FSB speed should it?

I was just thinking that the other day, and if I could get my damn corsair XMS 3200C2 to work with this epox 8rda, I would test it.

In theory, yes. Running two sticks in dual mode at 50% should have the same bandwidth as one stick in single mode at 100%.

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?

Only if you had the PC3200 or PC2700 memory running synchronously to the FSB as in a 1:1 FSB:RAM ratio.

Eg.
PC3200=DDR400=200MHz FSB if the processor can handle it
PC2700=DDR333=166MHz FSB if the processor can handle it

Using the same multiplier for both memory types would result in the PC3200 showing better results because of higher overall system clock speeds.

Am I making any sense?

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?

Click to expand...

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.

Good points, and in order to maximaize throughput of the CPU memory combo ideally one could then run 250x9.5 and 200+. Unfortunately though at least on the EPOX mobo anything higher than 75% does not work, and any higher fsb than 230 does not really seem feasible either without major mods...

On AMD systems the bottleneck is the CPU architecture. Theres little gain when the ram runs at higher fsb than the cpu. Dual channel also very little gain cause AMD cpu cant handle the extra bandwith effectively. Intel architecture is the opposite their cpus can handle more bandwith and dual channel.

I have a question in the same vein... altho I think my understanding is a little behind dreamtfk's. I am running an AMD xp2000 on a Soltek SL-75DRV5 333 mobo. I have micron PC2700 ddr333 ram (but suspect I actually have PC2100 due to SiSandra report). But the question is: My bios has different settings for FSB and dram clock. When I set FSB to 133 (stock) and ram to 166, I can't boot. Do I need to kick my vendors arse or is it not possible to run the xp2000 with dram clock of 166 w/o unlocking etc? If it IS my vendor shafting me... and I do replace with actual pc2700... will there be much performance increase since FSB will be 133 and dram @ 166? Any hard fact info about this would be great. Thx.

Dr.Octopus said:

I have a question in the same vein... altho I think my understanding is a little behind dreamtfk's. I am running an AMD xp2000 on a Soltek SL-75DRV5 333 mobo. I have micron PC2700 ddr333 ram (but suspect I actually have PC2100 due to SiSandra report). But the question is: My bios has different settings for FSB and dram clock. When I set FSB to 133 (stock) and ram to 166, I can't boot. Do I need to kick my vendors arse or is it not possible to run the xp2000 with dram clock of 166 w/o unlocking etc? If it IS my vendor shafting me... and I do replace with actual pc2700... will there be much performance increase since FSB will be 133 and dram @ 166? Any hard fact info about this would be great. Thx.

Click to expand...

Basically since your cpu is only running @ 133 fsb your only get PC2100 speeds from that PC2700 ram.
Your PC2700 can run @ 166 fsb but the 2000+ doesnt run 166@ stock. Option #1 unlock the 2000+ then lower its default multipler (x12.5) then it can run @ 166 fsb cpu /166 fsb memory. Option #2 buy a cheap 1700+ Tbred cpu that already comes unlocked and run it @ 166 fsb.

thx

thx for the succinct and helpful info