Dual channel.. where's my performance?

D

dobbz

Limp Gawd
Joined
Sep 4, 2002
Messages
339
this is my first dual channel bench using 2x256 BH-5. the reference bench's are single channel using 1x512 CH-5. at the same settings (230MHz, 2-2-2-11, CPC on) i notice i get get a crappy 130mb/s gain. this is wrong isnt it?

dcmyass.jpg
 
Are you using the system in your sig? If so, Athlon XP systems offer almost no real performance gains when using dual channel. The gains are about 3% and that's about it.
 
oh. i never knew that. :(

is it a limitation or did AMD just kinda say screw dual channel?
 
Well, nVidia's nForce2 chipset allows for dual channel, but the way the Athlon bus is designed, I don't think it needs to be constantly saturated by bandwidth, and it only needs data when it's needed, hence the low performance gain. I have no idea what I just said, it's late and I'm not very sure why AthlonXP chipsets are kind of weak in the dual channel area.

However, it is a very real fact that the dual channel architecture vs. single channel architecture only does offer a slight gain. Compared to the single channel Via KT400A chipsets, the dual channel nForce2 chipset only edged out the KT400A very very slightly in memory benchmarks.
 
The FSB is the bottleneck. If you want more performance you are going to have to OC your FSB.
 
The bottleneck is the bus WIDTH, not the frequency. The Athlon has a 64-bit bus. The Memory controller in dual channel mode uses 128 bits.

CPU -> NB is saturated by one channel.

Dual channel is only useful for the IGP motherboards.
 
Originally posted by mwarps
Dual channel is only useful for the IGP motherboards.
Click to expand...

ugh. that makes it sound like a 'fix' for crappy mobo's and video.
 
Originally posted by mwarps
The bottleneck is the bus WIDTH, not the frequency. The Athlon has a 64-bit bus. The Memory controller in dual channel mode uses 128 bits.

CPU -> NB is saturated by one channel.

Dual channel is only useful for the IGP motherboards.
Click to expand...

The bottleneck is with the amount of bandwidth the FSB can provide. Both the width in bits and frequency along with other factors like how many times the bus can transmit data per clock cycle all contribute to bandwidth. That is why a 200mhz 64-bit QDR Pentium 4 FSB (800mhz effective) can transmit more data than a 200mhz DDR Athlon FSB (400mhz effective). Although increasing the number of bits (width) the bus can transmit per clock cycle would help, he does not have this option. He does have the ability to increase the frequency of the FSB, which will increase the bandwidth and help alleviate his problem. However, because of the greater amount of bandwidth dual channel memory can provide and the limitations of the Athlon's 64bit FSB, the FSB will always be the bottleneck for his PC when looking at memory performance.

Looking at your SiSoft Sandra graphic, you can see that your theoretical maximum bus (FSB) bandwidth is about 3.7MB/s. Looking at the theoretical maximum memory bus bandwidth it is 7.4MB/s. That 7.4MB/s of memory bandwidth is bottlenecked by your slower FSB which can only provide 3.7MB/s. Looking at your system’s RAM bandwidth Int Buffered score, it’s about 3.6MB/s which is about the theoretical limit of your FSB.

If your memory and CPU are already running at their limit, you may be able to decrease your memory multiplier, decrease your CPU multiplier, and then overclock your FSB some more to increase it's bandwidth. You would then be pushing the limits of your north bridge to get the best memory performance. Even if your RAM ends up running a little slower after making these changes, it will still provide more bandwidth than your FSB can handle and your overall system memory bandwidth will increase. This is all in theory of course. There are other factors like running your FSB and memory bus out of sync can cause other inefficiencies and you may not get the gains you had hoped for.
 
You must log in or register to reply here.
Back
Top