View Full Version : nForce 2 dualchannel bandwidth

12-11-04, 07:30 AM
Is the nForce 2 memory controller really bad or what?

I thought dual-channel would be a huge performance boost for me. With my old VIA KT266A board I could get up too around 2.4 gb/s in singlechannel with the memory running at 333 mhz base speeds. Now with my new board I run dualchannel at 400 FSB with no dividers and the bandwidth is only 3.1 gb/s, less than the theoretical bandwidth of a single module. What's up with that?

12-11-04, 07:39 AM
sure you got em in the right slots for DC?
BIOS reports Dual Channel at boot?

Anyhow, I dunno if the acctual bandwidth went up, If the FSB could take in more the 3.2Gb/Sec..
Theoreticly 2x64bit Memcontrollers @ 400Mhz should yield 6.4Gb Sec.

Singel Channl 333Mhz should be about 2.6Gb/sec.
Singel 400 should be 3.2Gb/sec..

The main benefit of NF2 DC, if i remember right, was reduced latancy as it used the RAM in Interleave, read/Write channel1, channel2, channel1 etc..
sorta like RAID..
Dont remember if it would acctually yeild double bandwidth, depends on if the CPU FSB can take it in i guess, or if its only 64bit wide like a singel memory controller..

12-11-04, 08:10 AM
Hey. Yeah, i'm pretty sure they are running DC. I Don't think it says anything at POST but Sandra reports 2 channels and when I had them in other slots, I only got about 2 gb/s bandwidth.

I've definitely noticed an increase in responsiveness of the system so I guess it's doing something right. I'm just a bit dissapointed because I recently setup a Socket 939 AMD64 system for a friend and Sandra memory benchmark was insane on that system. I was expecting something similar.

12-11-04, 08:36 AM
I recently setup a Socket 939 AMD64 system for a friend and Sandra memory benchmark was insane on that system. I was expecting something similar.

That's because of the Athlon 64's on-die memory controller. Your current Barton score is pretty much normal because the chipset/CPU is not going to be nearly as efficient.

12-11-04, 08:40 AM
socket 939 isnt the same type of DC as Nforce2.
Nforce2 has 2 independant 64bit memory controllers, while A64 DC has one 128bit wide memory controller.
and that is ON the CPU so it can chugg down the 6.4Gb/sec for sure on 2x 400Mhz RAM modules.
Not sure how the AXP FSB works, if it can indeed fit 6.4Gb/sec or if its just 3.2Gb/sec max..
Even if the Chipset Memory Controller uses 2x64bit controllers, the CPU might still be limited to the equivalence of 64bit x 333/400 (FSB effective clock)

So.. Im not sure you should get higher bandwidth.
3.1Gb/sec is a tad lower then the theoretical max for SingelChannel atleast, which is often the case.

This shows 2.5gb/sec on 333Mhz RAM, sais 2 sticks so I assume its Dual Channel..

Pandora's Box
12-11-04, 08:49 AM
yeah i just test my athlon xp 2700+ and pc2700 ram. got around 2.2gb/s. yuck. omfg hurry up fedex! 2 days of this **** left!

12-11-04, 02:14 PM
Before you start complaining about how bad your system performs you should know a couple of thing. just because you switched from ddr333 to ddr400 does not mean that you are going to get a lot higher scores. Many of the *cheaper* ddr400 modules are actually worst than a high quality ddr333 module (say OCZ for example). The other thing is that unless you install the nvidia drivers and then run the test and you get low results you shouldn't even begin to think of complaining. Your chipset could very well be recognized as a generic chipset by windows and as such you will get horrible results. You should also know that windows xp is optimized for a i486 processor and previous versions go even lower than that. ( i486 is not even a pentium) So install your drivers, run the test again, make sure that you have the latest bios, and if you still get low result then you have something to complain about. Oh yeah and you are never going to get the results that you friend did on the athlon 64 because the memoy controler to cpu link on his computer is probably 1600Mhz and your's is 400Mhz. (like saturnotaku said the mem controler on athlon 64 is on-die).

12-11-04, 05:52 PM
I know the 64's intergrated controller is superior but I expected more bandwidth from a nForce 2 with dual-channel at 400 mhz than a VIA KT266A at 333 mhz single-channel. The memory modules are the same and I use 2-2-2-5 timings.
Also running nForce Drivers v5.10, or the newest official ones anyway.

But in the end, I guess it doesn't matter what Sandra thinks as this system runs way faster than my previous one, so whatever the nForce 2 is doing with it's dual-channel thingy, it's good. :)

12-22-04, 12:43 AM
DualChannel on Athlon XP is only meant to icrease efficancy not speed, it does this by accsesing the memory at different times.

So when one is beening acsesed the other is waiting for next clock cycle