a7n8x - Dual Chan. DDR 2700 -

[ciscobee]

I recently purchased the deluxe mobo and 2 512's from Crucial - DDR2700. I booted, it detected ... etc.
The ram is there, and the settings are "optimal" ... but now ... I'm sick of the ordinary, and ready to push the envelope.
I 've read through the official thread ... which is over-bearing ... and spend a couple of days surfing through searches, and I still cannot locate a decent thread on the dual channel ddr. I'd love to take advantage of this mobo's features, but honestly, when I look at the BIOS, I do not see a glaring red beacon that says "Dual Channel RAM" - "ENABLE". My sticks are in slots 1 & 3 per the recommendations of the collective, and the CPU is a 1700+ which I purchased at newegg , per the recommendations of the collective.
Here is some additional info. I've got two sticks of Crucial 2100DDR 256 that are collecting dust. The settings for the mobo are factory spec. Basically, it's assembled and running. Nothing more. The freq is 133 and the mult is 11. Saw another post with someone running like 1950 out of the same proc, and I'm still a little fuzzy on the 400 Mhz FSB - considering the manual says 333/266 or else 200.

Anyway - looking for some quality info on the dually with a side of processor specs.
PEACE

 

[mechBgon]

Switch all the "Optimal" settings to manual, set the AGP frequency to 66MHz, set the memory to run synchronously with the CPU's bus speed, and you're on the starting line. Your CPU bus speed and memory speed will be 133MHz at this point. Raising the CPU's bus speed, or the CPU multiplier, will raise your total MHz. If it'll run 200MHz, you've got that DDR400 setup.

If your memory is in 1 & 3 then you are running dual-DDR. It's not going to make a large difference in system performance versus single-channel DDR because the AthlonXP is usually getting all the bandwidth it wants even with single-channel. The primary benefit of dual-channel is when you've got a motherboard that has an onboard video processor, and the 3D gaming performance of the onboard video will jump very greatly when running dual-channel, because the onboard video processor needs that bandwidth badly.

 

[squidman]

No. Dual DDR only works (according to nVidia's engineers) with 64 bit DDR 2100 memory. It DOES NOT work AT ALL with nowadays 72 (thats why they say 128 bit - 64+64) bit memory chips. I was shocked too - according to CPU-Z i had mine disabled. No matter what i did (also, it doesnt matter, whether the sticks are in 1 or 2, as long as the other one is in 3) it stayed disabled. Then i consulted nVidia, and calmed down.

 

[LuDaCriS66]

Originally posted by: squidman
No. Dual DDR only works (according to nVidia's engineers) with 64 bit DDR 2100 memory. It DOES NOT work AT ALL with nowadays 72 (thats why they say 128 bit - 64+64) bit memory chips. I was shocked too - according to CPU-Z i had mine disabled. No matter what i did (also, it doesnt matter, whether the sticks are in 1 or 2, as long as the other one is in 3) it stayed disabled. Then i consulted nVidia, and calmed down.

Are you sure? THow come companies like Corsair and Mushkin have 2700 and 3200 modules designed for dual channel?

 

[ahfung]

I'm not sure if squidman's right but I got very similar experience and came to the same conclusion.

This is my thread posted a few minutes ago

Sandra reported around 3.2GB/s only for my dual DDR400 setup. Do the maths, this exactly PC3200. Took a DIMM away then re-run, the number didn't decrease.
I could get around 3.0GB/s with my previous P4 845 setup with memory running at over 400MHz single channel.

Something fishy

 

[mechBgon]

The nForce2's dual-channel memory controller can indeed run in dual mode with PC2100, PC2700, PC3200, PC3500, or anything inbetween if you pick an inbetween speed. You don't need to look any further than an IGP-equipped nForce2 board's gaming benchmarks to confirm this... the onboard video will give about 80% higher framerates when the memory's in dual-channel mode, due to the extra bandwidth, and it will do it at memory speeds other than just PC2100.

Sandra makes me look like this -->

but if you want to see dual-channel in action, toggle the buffering to the alternate setting (on --> off if I recall correctly) and try your tests again. I think the reason is that nForce2 has two independent memory controllers and load-balances between them as needed, so Sandra is really just testing one at a time in buffered mode. Don't take that as "official" but I suspect it's correct.

To reiterate, the benefit of DCDDR is mostly when using an nForce2 IGP board (onboard video). The video core needs that extra bandwidth, the AthlonXP really doesn't. Here are some benchmark results comparing single-channel DDR to dual-channel DDR for nForce2 non-IGP: Anand's benchie results. If you're disappointed, remember that even in single-channel mode, nForce2 was still significantly faster than KT333 and KT400 in almost everything, particularly when running a 333MHz-fsb processor where it has a major latency advantage on its side.

Hope that helps

 

[ciscobee]

-?-
Seems like I am treading water on the RAM issue. My existing, 2x crucial 333 512's are 64x64 .... not sure if I included that before .... but where will I see the performance gain? Looks like the bios is 7,3,3,2.5 and it seems like the only real question is whether or not I can use the existing RAM or a mash of the 512 and 256 to get the latency down. The one post out here has a 5,2,2,2 which is insanely fast ... and I'm confident that windows would catch on fire if I ran that ... so ...

With my present setup, and a bios rev of ...2001g (sure about the g, but not the 2001) - what should be my goal? ... I want to have a realistic target in sight, - basically, if trying to achieve a 5,2,2,2 with my current setup and 1950 out of the processor is a lofty goal at best ... I'd like to know. This board has caught on like wildfire at my job and so far myself and 5 co-workers have jumped on the deluxe bandwagon. No one else is pushing the envelope yet ...

Thanks for the heads up on the Nforce2 - and the graphics info ... I had no idea that all of the throughput was basically for onboard gfx etc...

 

[mechBgon]

The memory should always run synchronously (100%) with the FSB. Period. If you raise the FSB, then you will raise the memory's speed. Going from 133MHz (DDR266 or PC2100) to 166MHz (DDR333 or PC2700) nets you a 5-7% system-level performance boost just due to the faster memory speed alone. The fact that this also raises the CPU's speed proportionately will bring your CPU up from 1.47GHz to 1.83GHz is another separate boost.

However, if you really want a snappy system, get yourself a Seagate Cheetah 15k.3 and an Ultra160 SCSI controller. If I had a choice between another 800Mhz on my CPU speed or going to the 15k.3 from an IDE drive, I'd take the 15k.3 for sure. Warranted for five years of punishing 24/7 server usage, too. At Hypermicro, about $300 gets you an 18Gb 15k.3 and an LSI Logic U160 controller kit.

 

[squidman]

Go here:

http://www.nvidia.com/view.asp?PAGE=nf2_tech

and download the file thats called "NVIDIA nForce2 Memory Architecture Tech Brief
Read in detail about the NVIDIA nForce2 DualDDR memory architecture in the attached tech brief (2.3MB PDF)"
So dont worry. Dual channel doesnt exist on 72bit banks.
Also, i tested (out of interest) the "duality", by sticking memory in 1 and 2. Haha, PCMark gave 4450 as opposed to 4590 when in 2 and 3. Also, SiSoft sandra does utilize both channels. It benchmarks them with a 128mb file, but windows also loads about 128 MB of crap in memory. The other hing, is that architecture-wise, banks dont get loaded one at a time. THey both get into work, and the load is divided between them. This "dual channel" crap is just a mareting campaign (*cough* hyperthreading from intel, *cough*) that cost extra $, but doesnt do much. Im satisfied with my systems performance. Yes, i get 4600 mem marks, so? I got a 7-3-3-2.5 mem, not 5-2-2-2, which costs 2 times more. Rather spend the extra 80 bux (mem price difference) and invest into a better video.

 

[ciscobee]

Alright,
Looks like some education is on order on my part. Tonight I will install Sandra and see how she rates the system in it's current config, and then take out a stick of 512 and put in two 256's ... and then run the test again. Then I will run the a 1-512 2-256 and 3-512 setup and see the bench. I'll read the doco from nvidia also.

- TANKSALL -

 

[mechBgon]

Originally posted by: squidman
Go here:

http://www.nvidia.com/view.asp?PAGE=nf2_tech

and download the file thats called "NVIDIA nForce2 Memory Architecture Tech Brief
Read in detail about the NVIDIA nForce2 DualDDR memory architecture in the attached tech brief (2.3MB PDF)"
So dont worry. Dual channel doesnt exist on 72bit banks.
Also, i tested (out of interest) the "duality", by sticking memory in 1 and 2. Haha, PCMark gave 4450 as opposed to 4590 when in 2 and 3. Also, SiSoft sandra does utilize both channels. It benchmarks them with a 128mb file, but windows also loads about 128 MB of crap in memory. The other hing, is that architecture-wise, banks dont get loaded one at a time. THey both get into work, and the load is divided between them. This "dual channel" crap is just a mareting campaign (*cough* hyperthreading from intel, *cough*) that cost extra $, but doesnt do much. Im satisfied with my systems performance. Yes, i get 4600 mem marks, so? I got a 7-3-3-2.5 mem, not 5-2-2-2, which costs 2 times more. Rather spend the extra 80 bux (mem price difference) and invest into a better video.

Err, I don't quite see where you keep getting "72-bit" from. The speed of the modules (PC2100, PC2700, etc) has nothing at all to do with the width of the data path (64 or 72 bits). There are PC1600, PC2100 and PC2700 modules available in both 64-bit (non-parity) and 72-bit (error-checking-&-correcting or ECC) modules. nForce2, like almost all desktop-oriented chipsets, uses 64-bit data paths. Put 72-bit modules in, and it will simply ignore the parity info and run them in 64-bit width anyway (this is confirmed, a person used ECC modules successfully on his nF2 board).

I don't see how the dual-channel feature can be dismissed as marketing baloney when it provides real and very large performance improvements to people who are using the onboard video, such as the customers of Compaq, HP and Shuttle. If you're offended at having two memory channels just because someone else benefits and you don't, well then buy an nForce2-ST 400 non-Ultra board when they come out (single-channel DDR400, lower price).

I might add that yes, there is such a thing as a dual-channel ECC (72-bit) memory implementation I can name several chipsets that use it right off the top of my head: Granite Bay, nForce3 Professional 150, nForce3 Professional 250, i865, i875, and i7505.

 

[ciscobee]

Status Update:

10:00pm ... and I just tried every combination possible with my 256 and 512's. Sandra and I need to become better friends before I can understand this data? ... WOW ...
but, of interest... the largest performance gain where my software was concerned seemed to be in Pshop. The 2 x 512's in slots 1&3 took about 10 seconds to run a series of KPT Filters (if your unfamiliar, these are huge graphics filters). On the other hand, the 512, 256, 512 combo took about 15 seconds, and the 256,256 combo was garbage at around 20 seconds. Something changed, and it wasn't Pshop...had to be the RAM. But I really want the "raw" data. Just because of the whole quest factor. It seems like it's impossible to clock that speed and then measure it's performance, but yet nvidia and the rest of the big pimpin companies must have a piece of software that tells them, "use this ram, with this proc, at these speeds and viola' - it's ultra fast".

Next up to bat ... Belarc ... we'll see what they say.

 

[squidman]

I dont care about having 2 channels, or one. As i said, "im satidfied with my mobo performance". It might have 3 channels, doesnt matter, as long as the performance satisfies me. Or otherwise i'd habe returned it back to the store.