Soltek NV400-64: Single Channel nForce2 400 Review

[KF]

> we used the same mainboard in single-channel mode to test the single-channel NVIDIA chipset.

Single channel vs dual channel tests have been done many, many times over, with this same result. That's why the results with the new chipset seemed incredible. Even the Soltek reviewer coudn't believe it, and contacted Nvidia to attempt to get some answers.

 

[Hauk]

Regardless, problem solved for me! :beer:

I dumped $$ into a single 512MB stick of Corsair XMS PC3500, which kept me from going Nvidia. Was a cat hair away from buying a KT600. Now I can dump my KT400, and join the crowd!

Any word if this or future boards with this chipset will allow manipulation of the multiplier (XP 2500 or XP 2600)??

 

[KF]

Originally posted by: SteelSix
Regardless, problem solved for me! :beer:

I dumped $$ into a single 512MB stick of Corsair XMS PC3500, which kept me from going Nvidia. Was a cat hair away from buying a KT600. Now I can dump my KT400, and join the crowd!

Any word if this or future boards with this chipset will allow manipulation of the multiplier (XP 2500 or XP 2600)??

to BIOS pics

"The multiplier goes up to 22x and the voltages have highs at around 0.2v higher than the defaults, so AGP: 1.8v, DIMM: 2.8v, CPU: 1.85v, VDD: 1.8v. "

 

[DAPUNISHER]

Originally posted by: KF
> we used the same mainboard in single-channel mode to test the single-channel NVIDIA chipset.

Single channel vs dual channel tests have been done many, many times over, with this same result. That's why the results with the new chipset seemed incredible. Even the Soltek reviewer coudn't believe it, and contacted Nvidia to attempt to get some answers.

Precisely, Hence why I'm very interested in seeing the Soltek reviewed elsewhere, so we can see if like Kyle's review of the KT600 it's an aberration or if indeed there is something (the implementation of DASP perhaps???) that is boosting the performance on the 400 chipset.

 

[NesuD]

Originally posted by: DAPUNISHER

Originally posted by: NesuD
check this socket A chipset roundup. They compare an abit board with the nforce2 u400 with a the soltek single channel board as well as the new via kt600 chipset and the newest SIS socket A chipset. Their results clearly put the U400 on top of the single channel 400 although it is by a pretty slim margin in most tests. Nforce2 anything basically manhandles the rest hands down. Socket A chipset roundup

I thought that at first glance too but if you look closer

For NVIDIA nForce2 Ultra 400 we selected ABIT NF7 2.0 mainboard. This solution demonstrates almost the fastest performance today among all mainboards based on the same NVIDIA core logic. For this reason and also because the mainboards based on the single-channel nForce2 400 are not that widely spread yet, we used the same mainboard in single-channel mode to test the single-channel NVIDIA chipset.

Until we see more reviews using the NF2 400 SC and not a ultra400 in SC mode it's still not possible to say conclusively that the ultra400 is faster even though all the known spec indicate that should be the case.

This is what the have to say about the differences between Nforce2 ultra 400 and Nforce2 400

this way the company could also find an application for the defective NVIDIA nForce2 Ultra 400 chips, which had only one memory channel working properly: they simply disabled one memory channel in such chips and then they were sold under NVIDIA nForce2 400 name.

According to them nforce2 400 is simple a ultra 400 with one of the memory controllers disabled. They don't give a source for this information but they aren't phrasing it as speculation either.

 

[DAPUNISHER]

I'm with you NesuD, and that's why Evan jumped right in to say it can't be faster than the ultra400, but if the numbers from the AMDMB review can be reproduced by others then some benefit is derived by the elimination of the other controller or there is an X factor to account for. Also, The fact that Nvidia didn't provide a clear answer to Ryan's inquiries also lends to the mystery and peaks my interest even more :evil:

 

[DAPUNISHER]

Well the Inquirer weighed in on the topic LinkInteresting stuff despite the source. There's a crappy review linked but it tells me nothing. The guy, knowing the overclock limitations of his CPU beforehand, doesn't even drop the multiplier enough to see what the board has in the way of fsb potential!?!? :disgust:

 

[Tab]

It looks alright, but I HATE purple!

 

[First]

Soltek's nF2 400 board will be here by Monday (maybe Saturday). We'll see if this turns out to be true. This'll certainly be a great buy for enthusiasts, though I'm sure NVIDIA will be tearing their hair out.

 

[KF]

>this way the company could also find an application for the defective NVIDIA nForce2 Ultra 400 chips, which
>had only one memory channel working properly: they simply disabled one memory channel in such chips
>and then they were sold under NVIDIA nForce2 400 name.

I'm going to argue against this, although I think it is factual that in some cases, for certain chips, manufacturers do disable parts of the chip to get larger yields (or for marketing reasons.) But generally the disabled sections attach to a common bus, making disabling simple. You turn off the set of buffers that gate the signals onto the bus, a function which has to present anyway. Dual memory controllers have a separate set of wires for each (I believe), not a common bus. (It wouldn't make much sense to multiplex the two memory channels into the chip; you would need an additional external chip to it.) So instead, you would have to hook up an alternative set of a large number (>128?) of data, address and control wires to different pins on the chip, not easy to do.

The premise of the whole thing is to REDUCE manufacturing cost. If the additional memory channel doesn't use a lot of costly chip space, then removing it is not going to save much, and not going to improve yields much. So lets say it takes up a lot of chip space. Your best solution is to reduce the die size by redesigning the chip. You save on the chip package too by not having to hook up an additional channels worth of pins, possibly using a package with less pins.

Part of the saving with the single channel is the cost of the mobo. A less complex mobo, with fewer wires, is liable to be somewhat cheaper. If you leave out a DIMM socket, you save some more. (Signal quality is easier to maintain with fewer wires and fewer DIMM sockets too.)

The new Solek is cheaper, but what they leave out compared to thier other nForce mobos accounts for a chunk of that. The Nvidia chip with sound and firewire seem to be expensive, and lots of mobo makers cut that.

>It looks alright, but I HATE purple!

I normally avoid purple. It's OK for kids and women I suppose. (Although I once got some purple underware as a gift during the hippy era.) But I rather like the yellow and purple combo on this mobo. Blue LEDs flashing all over it would make quite a picture.

 

[NesuD]

Hmm i don't know simply not having traces for the second memory controller would effectively do it wouldn't it. i don't see where it would be any different than simply leaving dimm slots for one of the channels unused. That in effect disables one of the controllers. If the nforce2 400 really is an ultra 400 with a disabled memory controller than it would seem to me the best way to do it would be to simply leave the connections to one of the controllers unused. It wouldn't surprise me at all that both memory controllers by themselves are fine for 400mhz operation but won't pass muster in dual channel so it would make perfect sense to do this with such chips. I think cost reductions would be far better served by utilizing low yield chips(as in not dual channel capable at 400mhz) that would otherwise be trashed. that seems more economical than taping out another chip sans i memory controller and fabbing it separately. Higher yields of saleable chips does REDUCE manufacturing costs and this is what you would have if you were able to use some of those non qualifying chips by using only a single memory channel.

 

[KF]

>Hmm i don't know simply not having traces for the second memory controller would effectively do it
>wouldn't it. i don't see where it would be any different than simply leaving dimm slots for one of the
>channels unused. That in effect disables one of the controllers.

It doesn't work if the channel wired to the mobo happens to be the defective one. In addition, if a channel is defective, maybe the defect will prevent it from automatically disabling itself.


>seems more economical than taping out another chip sans i memory controller and fabbing it separately.

I don't know the costs involved. Maybe it is different with chips - I doubt it- but in mass producition, generally developement costs are dwarfed by production costs. They will spend whatever it takes to reduce production costs, because it always pays. They have guys looking how to save 1/4 cent by using a cheaper screw in an automobile that costs $20,000. If he finds one, his salary is in effect paid for.

Chips are always undergoing revisions. People often use the jargon for one kind: stepping. Therefore I don't think small revisions, updates, and fixes to chips are so cost prohibitive as people make it out to be. They just don't do it if it doesn't pay.

With foresight they could put one memory controller in a rectangular section that is omitted in the second chip design, making a complete revision unnecessary. If you look at the layout of chips, they generally do confine certain functions to certain areas. That way only one section needs to be revised, substituted or removed.

Balanced against revision costs and separate production:

Yes, saving defective chips from the junk pile is cost effective. But for objects that represent MORE sales than complete version, such as a cheaper version, it seems unlikely to me that you will be saving anything. Probably 90% or more of the chips will not be defective, so almost all of the chips that have a disabled section will in reality not be defective. You will be creating more problems, because defects go up in direct proportion to chips size. And you will be spending more on each chip because the number of chips per wafer will be reduced.

Defects go up in direct proportion to chips size, so anything you can do to reduce chip siize reduces cost per chip. You don't just save by salvaging the inevitable, but small number of, defective versions, you save on every single chip.

Things other than CPUs and GPUs do not use bleeding edge and scarce processes. CHipsets for instance. They use routine processes that are readily available in large quantity. Separate production therefore is not necessarily cost prohibitive. The foundaries of chip companies are always making a variety of chips and looking for more contracts, because making chips, not designing, is their business.

 

[NesuD]

You make some pretty good points. It would be intersting to know if they are actually manufacturing a different Northbridge that has only one physical memory controller in it or if they are really using the standard SPP northbridge with one disabled as was described in the Xbit article.

Evan maybe you could compare chipset revisions when you look at the board or even get a statement from nvidia directly to see what they have to say about it.

 

[Hauk]

Evan got the motherboard in yesterday. Said he'd have a review posted by next week. Wonder who else will be putting out a board with this chipset. I've seen Asus thus far...

 

[DAPUNISHER]

HarryC the soltek mod at AMDMB showed some benchies of his SL-75FRN2 and they were a little higher than the NV400's. I think MARMADUKE had it right in that comparing it to the Asus was not an indicator of how it'll do against other NF2 ultra400 boards since the Asus isn't the fastest among them. I suspect Evan will find that XBit's results of 2-3% slower will be about right but that's still fast enough to keep users from choosing VIA over Nvidia for performance on a tight budget.

 

[mdcrab]

The Soltek NV400-64: Single Channel nForce2 400 is back in stock at Newegg at $76 shipped.

mdcrab

Edited 7/31: Corrected Link.

 

[colonel]

got mine last night after 15 days of waiting, I m a Soltek freak and I will upgrade from my Soltek Via kt266a.

 

[BDSM]

Personally I am not surprised at all that a single channel system could be faster than a dual one because the Athlons fsb is very much saturated by one 64 bit ddr channel running in sync with the fsb. So dual channel won't do much for memory throughput but it could very well affect latencies negatively which is what I think happens with the nf2 ultras.

Running the ultra single channel is most likely nothing like runnign the non ultra sc because the ultra is most likely optimized for dual channel..

Anyway.. As someone already brought up.. It could be that the Asus is just a slow nf2 ultra board.

 

[KF]

Originally posted by: mdcrab
The Soltek NV400-64: Single Channel nForce2 400 is back in stock at Newegg at $76 shipped.

mdcrab

That link took me nowhere.

relink