nforce2 is NOT a true dual channel chipset

After 14 hours of straight testing, I have concluded that the nforce2 does not ever really run its memory at 128 bits.

With the memory at 160mhz (320mhz DDR) and the front side bus anywhere from 200-214, the memory scores exactly as it does at 160mhz sync mode. There is no second channel, no matter what I change in the bios or on the board.

I tried the ram in slots 2 and 3 (as stated in the bios) and 1 and 3 (where it is now), and i cannot score anything above what 160fsb scores in sync mode. Used newest version of sandra and 3dmark 2001.

Board is an NF7 v2.0, 1.1 bios

Using 2x 256mb crucial pc2100 which does 160fsb@2-2-2 (which are
the memory settings im using now)

The system does not utilize the extra bandwidth that 128bit mode *should* be making available, the only conclusion left to get is that it is not really running 128bit mode.

i have done similar tests with a Soltek 75FRN2-L and reached the same conclusion

my benches are no different in single channel to double channel

dual channel also crippled my chips overclocking ability. in dual channel i was only able to reach 170fsb and in single i hit 233fsb with a JIUHB 1800XP.

i am using a pair of 256MB GEIL PC3500 sticks

Really? I haven´t noticed any difference either.. Ill change my memory back to single mode and check..

ya the "dual channel" just give it a little extra boost in 3dmark compared to running sync.... like 400 extra points for me

Ok, i didnt notice any difference at all using single and dual.. Accept for that i need much more voltage to run the memory stable in Dual mode..

All i ever took DDDR was as a 50% increase (not doubled, only +1/2, doubled would be a 100% increase) in bandwidth...hehe thats advertising for ya =\

Of course i imagine that some manufacturers will give a different increase ind bandwidth for different types of memory, and some may not....

Another thing may be that with different motherboards comes different things you have to remember to read off of the spec sheet, they may say DDDR (Dual Channel DDR) but it may not in truth be a duobled bandwidth...i imagine that somewhere there is that information on which you would be able to see which offer the biggest increase in mem. bandwidth..I could be some rambling fool that makes some sense or i could be totally wrong who knows..8)


Well..just my thoughts upon the subject

The problem is not with the dual memory controllers, its with the Athlon chips. If you run memory benchmarks you will see that indeed dual channel works and gets close to its theoretical 2x advantage. However the Athlon chips have no use for this kind of added memory bandwith, especially when theyre running at higher FSBs.

In other words, it works, but for overall performance its worthless.

Rocko[DPC] said:

The problem is not with the dual memory controllers, its with the Athlon chips. If you run memory benchmarks you will see that indeed dual channel works and gets close to its theoretical 2x advantage. However the Athlon chips have no use for this kind of added memory bandwith, especially when theyre running at higher FSBs.

In other words, it works, but for overall performance its worthless.

Click to expand...

Thats why I ran the fsb 40% higher than the memory for all the tests. It SAYS my ram is running at 320fsb, but I get no increase in performance whether the FSB is at 160 or 200. If it really was in dual channel mode, then I would get a lot of extra performance when running the FSB at 200. I get like 100mb/sec extra going from 160-200 on the fsb, I should get at least 300 more, at least.

In theory you should be able to have to fsb at 200 with two sticks of pc1600 and get the same performance as single channel ddr400. If this doesn't hold true then it isn't dual channel.

You can't use sandra to test your mem benchmarks, sandra only uses one stick of mem during testing. I heard rumors a few weeks ago that there would be an update for it, but haven't seen it yet.

Rocko[DPC] said:

The problem is not with the dual memory controllers, its with the Athlon chips. If you run memory benchmarks you will see that indeed dual channel works and gets close to its theoretical 2x advantage. However the Athlon chips have no use for this kind of added memory bandwith, especially when theyre running at higher FSBs.

In other words, it works, but for overall performance its worthless.

Click to expand...

That's more or less how I thought it worked. The Athlon has a DDR memory bus, so single channel DDR is all it can handle. A Pentium 4, OTOH, can take advantage of DDDR nicely since it has a QDR bus. This is what I've always thought. Am I right?

http://forum.oc-forums.com/vb/showthread.php?s=&threadid=185118

http://home.earthlink.net/~deeppow/OC_Guide/Mem_Mobo_CPU_Overclocking_Guide.htm#ocnforce

Demont said:

http://forum.oc-forums.com/vb/showthread.php?s=&threadid=185118

Click to expand...

I posted my detailed testing results there FYI. My results are verified, dual channel does not work as it should, at all.

Malakai said:

I posted my detailed testing results there FYI. My results are verified, dual channel does not work as it should, at all.

Click to expand...

I replied there, too. Did you check both links?

Dual channel does not work as it should in your board. One board.... dual channel works fine for me as well as a host of other's.

Malakai said:

I posted my detailed testing results there FYI. My results are verified, dual channel does not work as it should, at all.

Click to expand...

Ed did an article on this a while back. DDDR only adds a few % points to performance on Athlon boards. It's entirely because the Athlon's DDR bus doesn't allow more bandwidth than a single channel of DDR does. The few % points only came from the rare clock cycles where the primary DDR channel can't offer itself up to the processor.

That was the gist of Ed's article.

Demont said:

I replied there, too. Did you check both links?

Dual channel does not work as it should in your board. One board.... dual channel works fine for me as well as a host of other's.

Click to expand...

Yes I did, and my results are almost identical to his. At 200fsb/100memory, we should both get over 2700mb/sec easily. Also, his 3dmarks would not run, where I had no problem 3dmarking at all in sandra.

Sandra did show small improvements running async, during the first 3dmark score (the lower one) that I posted I scored 2500/2400, and I only score 2400/2400 with the fsb at 160. None the less, 3dmark scored much higher while running sync mode. Since he was unable to show a sync 3dmark score, I cannot compare them. I am fairly sure that his scores would have agreed with mine, I have been discussing this around many forums. Everyone so far has had bad results using async, and most people are saying its pretty much marketing hype.
Either way, the NF7 2.0 is a damn nice, fast, and stable board. It just will not run async dual channel at anywhere near its max theoretical bandwidth, hence I say it is not really dual channel

I'm confused now... earlier you were saying no benefit from mem @ 160 and FSB at 200-214 right?

I think you are considering this in an all or nothing type of way... The fact of the matter is that dual channel is neat and works to some extent but its still no substitute for having memory and fsb sync'ed @ 200mhz+ or whatever. but there is some middle ground between dual channel not existing (as you claim) and 100% efficiency!!! Of course its not perfect.... but it IS there.

based on data deeppow is showing on his guide and testing I've done there is something to be gained! If you look at his chart he is running @ 200 FSB..... testing with 2 sticks of ram in dual channel mode shows that running @ 133 makes them even a little fasterthan one stick @ 166! thats 2373/2218.... easily a couple hundred mb/s than you would get with fsb & ram @ 133 synced. doesn't sound useless to me. in fact, would probably be quite useful for folks such as yourself that dont have the latest and greatest ram. again, perhaps it is just your board that has problems.