3 sticks...dual channel

[fleflikr]

currently have two sticks corsair 32oo (2 x 256) 6 3 3 2 10 x 220
would like to increase ram to 1gig
two sticks are currently in channel 1 and 3
could i put those two sticks (256) in channel 1 and 2 and the third stick of 512m in channel 3
and run dual channel without hindering performance?

also could i mix in a kingston hyper instead of corsair for third stick?
(got a gift certificate for "best buy")

a7n8x deluxe ver2
2500 barton

 

[Duvie]

don't take my word for sure cause I mainly have P4 systems, but I remember hearing 3 sticks wil revert the process back to single channel DDR...They need to be in pairs....

However that being said I though it was widely acepted that dual channel DDR does very little for the athlon xp as the cpu is the bandwidth limiter....

IE barton 2500 has a 166fsb for 333dr or 2.7gb/s of bandiwdth...1 stick of pc2700 will fill that or nearly....2 sticks gives you a theretical cpacity of 5.4gb/s which is way over the cpus ability....

Now if you bump the fsb up to 200fsb but used say an asynch ram ratio like 3:2 (which I don't know if you guys even have!!!) then the cpu would be 3.2gb/s but single channel pc2700 would not fill it all...a secind stick in dual channel could help more in this situation...

I think the above situation may not be able to happen on most boards so I would say weigh the difference of the added ram versus what you really get in performance from the dual channel...

1) most instance dual channel requires looser timings

2) may require more vdimm

3) may hold you back on an OC...


You really should test single channel versus dual channel vs the added ram....The best way to find out is to see it yourself with your unique uses...

 

[Doh!]

For nforce2, having 3 sticks is possible for a DC DDR.

 

[Duvie]

Good to know...Be careful about cross mixing!!! In dual channel mode things need to be in sych amongst the ram...I have heard of issues again on p4 systems as being not to forgiving for this, but I will say get an AMD opinion for sure....general rule try not to do it....

 

[fleflikr]

i thought i read somewhere that if you use 3 slots, slots 1 and 2 combine into a single channel while the 3rd slot becomes the 2nd channel keeping the dual channel enforced .....i guess this is what i need to know for sure.
thanks

 

[pspada]

This ^ is true, but AFAIK you don't have to balance the two channels. Anotherwords, it matters not which slots you have the various chips in. My experience in the specific case of my main machine has been using 3 identical memory modules, all 512Mb. CPUz shows a Size of "1536 Mbytes", and Channels # as "Dual", on the current FIC AU13, and the prior Epox 8RDA+.

 

[myocardia]

Originally posted by: pspada
This ^ is true, but AFAIK you don't have to balance the two channels. Anotherwords, it matters not which slots you have the various chips in. My experience in the specific case of my main machine has been using 3 identical memory modules, all 512Mb. CPUz shows a Size of "1536 Mbytes", and Channels # as "Dual", on the current FIC AU13, and the prior Epox 8RDA+.

Wrong, when you have 2x256 sticks, and one 512mb stick, you'll have to move the stick that's currently in slot #3 to slot #2, so you will have 512x2, as far as the motherboard can tell. I personally wouldn't mix types of ram, but if you do, you may very well have to "backoff" on the ram timings to something like 2.5-3-3-6. BTW, you know, don't you, that nForce2 boards actually run faster when you run the memory at 2-3-3-11. It's been tested by many people, including me, and for whatever reason, it's faster. I run mine at those speeds, even though my memory is good for 2-3-3-5.

 

[Rafael]

Once I ran in my system a pair of 256 Mb GEIL GOLDEN DRAGON PC3200 in slot 1 and 3, and my 512 Mb CORSAIR XMS PC3200 stick in slot 2. It ran DC, not a single problem.

But I dont know if theres a gain or loss of performance running 2 x 512 Mb sticks or 2 x 256 Mb sticks + 1 x 512 Mb sticks. And I didnt try running the 2 x 256 Mb sticks in slots 1 and 2, and the 512 Mb stick in slot 3, like myocardia said.

But it worked here like I said in my system. Abit NF7-S.

Hope it helps!

Raf

 

[fleflikr]

thanks for the help guys and "myocardia" thanks for the heads up on memory settings!

 

[myocardia]

Originally posted by: Rafael
Once I ran in my system a pair of 256 Mb GEIL GOLDEN DRAGON PC3200 in slot 1 and 3, and my 512 Mb CORSAIR XMS PC3200 stick in slot 2. It ran DC, not a single problem.

But I dont know if theres a gain or loss of performance running 2 x 512 Mb sticks or 2 x 256 Mb sticks + 1 x 512 Mb sticks. And I didnt try running the 2 x 256 Mb sticks in slots 1 and 2, and the 512 Mb stick in slot 3, like myocardia said.

But it worked here like I said in my system. Abit NF7-S.

Hope it helps!

Raf

Rafael, your system may have ran fine the way you set it up, but it wasn't running as fast as it could have been, since you were only running 2x256 in dual-channel, even though you had 1GB installed. You had 768 in one channel and only 256 in the other channel. The channel that has the lesser amount of ram forces the other channel to use less than it has, because, as the name implies, "dual" has to have the same amount in each channel. Try it that way again, and do a benchmark with Photoshop, or Photodeluxe, and you'll see that your times won't drop any with it setup wrong, and they will drop dramatically with it setup right.

 

[pspada]

Originally posted by: myocardia

Originally posted by: Rafael
Once I ran in my system a pair of 256 Mb GEIL GOLDEN DRAGON PC3200 in slot 1 and 3, and my 512 Mb CORSAIR XMS PC3200 stick in slot 2. It ran DC, not a single problem.

But I dont know if theres a gain or loss of performance running 2 x 512 Mb sticks or 2 x 256 Mb sticks + 1 x 512 Mb sticks. And I didnt try running the 2 x 256 Mb sticks in slots 1 and 2, and the 512 Mb stick in slot 3, like myocardia said.

But it worked here like I said in my system. Abit NF7-S.

Hope it helps!

Raf

Rafael, your system may have ran fine the way you set it up, but it wasn't running as fast as it could have been, since you were only running 2x256 in dual-channel, even though you had 1GB installed. You had 768 in one channel and only 256 in the other channel. The channel that has the lesser amount of ram forces the other channel to use less than it has, because, as the name implies, "dual" has to have the same amount in each channel. Try it that way again, and do a benchmark with Photoshop, or Photodeluxe, and you'll see that your times won't drop any with it setup wrong, and they will drop dramatically with it setup right.

So, you are saying that my setup, nForce 2 in dual channel mode with 3 512Mb sticks will only use 1gig of the memory, since it needs to be balanced? Then why does CPUz (and any other utility you can name) show dual channel mode with 1.5gigs of memory?

 

[myocardia]

No, CPU-Z says you have 1.5GB of ram, because you have 1.5GB of ram installed.

And it says that you're running in dual-channel, because you have a stick in slot#1, and a stick in slot#3. That's what's called dual-channel.

 

[Rafael]

Myocardia, it makes lot of sense what you said.

I didnt know about the timings settings 2-3-3-11 were faster than 2-3-3-6. Thanks for the hint!

Pspada what Myocardia is saying is that in channel 1 you have 1 Gig of ram, and in the channel 2 you have 512 Mb of ram. Its unbalanced, so you wont get the same performance as if you had like 768 Mb in channel 1 and 768 Mb in channel 2.

Raf

 

[mechBgon]

My understanding is that due to the crossbar nature of the nForce2 SPP's memory controllers, it really doesn't matter which modules are where. Controller 0 can access modules 1, 2, and 3. Controller 1 can access modules 1, 2, and 3 also. This ain't Intel land.

I looked for a while to try to find where I dredged that up from, but with a 49k dial-up connection, it's slow going to slog through every blasted LostCircuits mobo review If I happen to be incorrect, then we are talking about a ~5% drop in performance on the last module... or to put it another way, we have two modules running at dual-channel plus a bonus module that's running nearly as fast The glass is at least 97.5% full, the way I'm looking at it, and I do in fact run 1.5GB on my nForce2 and benefit significantly from that last 512MB, whatever channelization it's running at.

 

[joe2004]

The answer to the original question is decidedly NO. Three sticks will cancel dual channel.
Some time ago I actually tried that and the performance was immediately down.

 

[pspada]

Originally posted by: mechBgon
My understanding is that due to the crossbar nature of the nForce2 SPP's memory controllers, it really doesn't matter which modules are where. Controller 0 can access modules 1, 2, and 3. Controller 1 can access modules 1, 2, and 3 also. This ain't Intel land.

I looked for a while to try to find where I dredged that up from, but with a 49k dial-up connection, it's slow going to slog through every blasted LostCircuits mobo review If I happen to be incorrect, then we are talking about a ~5% drop in performance on the last module... or to put it another way, we have two modules running at dual-channel plus a bonus module that's running nearly as fast The glass is at least 97.5% full, the way I'm looking at it, and I do in fact run 1.5GB on my nForce2 and benefit significantly from that last 512MB, whatever channelization it's running at.

Indeed, it does not matter what modules are where, which is what I've been saying all along. Simple enough to prove, just put your 256Mb sticks into the "first" channel, and the 512Mb stick into the "other" channel, and run some benches. then swap the 512Mb stick with one of the 256Mb sticks and run some more benches.

I tried this when I first moved to a nForce2 board, as I had some concerns. The benchmarks were virtually identical. Realize that the RAM should have the same latency ratings regardless of size, or it may invalidate your testing methodolgy.

 

[myocardia]

Originally posted by: mechBgon
My understanding is that due to the crossbar nature of the nForce2 SPP's memory controllers, it really doesn't matter which modules are where. Controller 0 can access modules 1, 2, and 3. Controller 1 can access modules 1, 2, and 3 also. This ain't Intel land.

I looked for a while to try to find where I dredged that up from, but with a 49k dial-up connection, it's slow going to slog through every blasted LostCircuits mobo review If I happen to be incorrect, then we are talking about a ~5% drop in performance on the last module... or to put it another way, we have two modules running at dual-channel plus a bonus module that's running nearly as fast The glass is at least 97.5% full, the way I'm looking at it, and I do in fact run 1.5GB on my nForce2 and benefit significantly from that last 512MB, whatever channelization it's running at.

Well, it does matter which modules are where, BUT, you're right about the glass being 97% or so full, also. I was merely answering his original question, which was "Can I run dual-channel, if I add another stick of ram...". The answer is yes, but only if you put the two 256 sticks on the same channel, and the 512 stick on the other channel. I'm planning on adding a 512mb stick pretty soon, since Thugsrook showed me how to run 1 gig with Window 98se.

 

[pspada]

Ok, I'll test it with 3 identical modules of Kingston PC2700, 2x256, and 1x512. Any particular benches you'd care to see?

 

[mechBgon]

UT2003 Demo botmatch would probably be a good one that doesn't take too long to run, if you think it's repeatable enough.

 

[myocardia]

Originally posted by: mechBgon
UT2003 Demo botmatch would probably be a good one that doesn't take too long to run, if you think it's repeatable enough.

Agreed. I'd like to know what the difference will be myself. I'm sure it's not gonna be huge, since dual-channel on an Athlon only gives you a max of roughly 5% better performance.

 

[pspada]

Ok, with the 512 stick in the offset slot (Bank 0), and the 2 256 sticks in the 2 close together slots (Bank 1), so the memory is "balanced", UT2003 Citadel botmatch is:

9.461353 / 31.059847 / 84.692574 fps -- Score = 31.087503 rand[23108]

I then swapped the 512 stick into the middle slot, putting that 256 stick into the offset slot, and the UT2003 Citadel botmatch is:

9.418547 / 31.061878 / 88.517540 fps -- Score = 31.089642 rand[23108]

so, I got a better score with the memory "unbalanced".

 

[myocardia]

Originally posted by: pspada
Ok, with the 512 stick in the offset slot (Bank 0), and the 2 256 sticks in the 2 close together slots (Bank 1), so the memory is "balanced", UT2003 Citadel botmatch is:

9.461353 / 31.059847 / 84.692574 fps -- Score = 31.087503 rand[23108]

I then swapped the 512 stick into the middle slot, putting that 256 stick into the offset slot, and the UT2003 Citadel botmatch is:

9.418547 / 31.061878 / 88.517540 fps -- Score = 31.089642 rand[23108]

so, I got a better score with the memory "unbalanced".

Yes, but that's only because the 2 are less than 1% difference. Run it three more times, and it's likely to change every time. Remember when I said that there's not much advantage at all to running an Athlon in dual-channel? Thanks for proving it for us, though. I had always wondered just how much of a difference it makes.

 

[mechBgon]

I'm not convinced that any of those results were single-channel results. How about a follow-up using one 512MB (definitely single-channel) versus two 256's in DC mode, to see what that does?

 

[pspada]

You are correct, none of these results was single channel. But I'm not convinced that the botmatch is a good benchmark, since it does not run identically each time. Any suggestions for a better memory benchmark?

Meanwhile, here's what I got (I ran each twice)

Single Channel (one 512mb stick):

9.456475 / 31.056349 / 87.504974 fps -- Score = 31.083906
9.486269 / 31.058781 / 84.684944 fps -- Score = 31.086283

and Dual Channel (two 256 sticks, one in back0, one in back1):

9.445383 / 31.066645 / 91.849991 fps -- Score = 31.094511
9.431480 / 31.058996 / 88.747932 fps -- Score = 31.086664

 

[mechBgon]

If you have the time for it, there's SETIBench, which I made and put up here. SETI@Home does heavy mathematical number crunching on radio-telescope data, as most of us know, and uses the memory a fair amount. It's somewhat sensitive to latency and bandwidth. The instructions for the SETIBench kit are in the Zip file.