best chipset

[jogutz]

I know nforce2 is the best, how would you rate other chipsets?:)

[BipolarBill]

What's left to rate? The KT400a is good too.

What about the Intel chipsets? the 845PE, 865PE and 875P chipsets are plenty good for Intel CPUs.

[Izdaari]

Yep, nForce2 and KT400A are the only real contenders for the top AMD chipset. On performance they're very closely matched. For me the choice between them is obvious because I trust Nvidia to make a good chipset and I don't trust VIA as much, but that's just me and your mileage may vary.

[NDD]

nForce2 is the only "worth" chipset today which supports 200MHz FSB CPUs. KT600 will do it too, but nForce2 still got dual-channel DDR :)

[deadkenny]

SiS 748 chipset might also be worth checking out. Of course it could also end up being another 'stealth' release by SiS, kinda like the 746FX.

[Someone Stupid]

Dual Memory controllers really doesn't benefit the processor much at all. If it was a SMP setup then having dual controllers would be something that would truly boost performance. Dual Controllers is like IDE raid speed wise, not a big boost at all. Though I have to agree, excellent chipset.

[stix_kua]

I've had good times with VIAs KT266/A but I've heard good and bad about nForce2....

Sis, and VIA have always been nice to me...:)

[mjrtoo]

Well, I disagree that dual channel DRAM is not that big of a deal. from 3.4 Gig per second to 6.8 is pretty big to me.

[Someone Stupid]

But that is not how it happens. This is how the 2 controllers work (actually they work the best when you are as asynchronous as you can get, and even then, it's only around a 30% overall increase). The first controller gets the first fetch command, based on that command, the second controller GUESSES at what is needed. When running in synch, on average on 4 to 5% of the time what is pulled is needed, as that is the true gain. You have to run so far off Async to get the "maximum" benefits of the dual controllers that on an Althon you wind up running slower even though you are supposedly now going 30% faster.

Remember, the chip itself is still designed for one controller - if they can improve their algorimths to where it gets it right 30% of the time when running in synch then it is really something to make something of. As of now, 5% on a 2.2ghz chip to me is only 2.3 ghz. Not a huge change there considering that's 3200+ speeds.

If the Althon had two memory pathways, one for each controller and used them that way it would work great, or for an SMP platform with each processor using a controller, but as of now, with the current algorimth it isn't a huge deal. Mostly included in there probably for the graphics chipset then they decided to intergrate it into all chips anyhow as it is good PR hype.

Don't get me wrong, the THEORETICAL bandwidth you mentioned is there, just the REAL bandwidth that is being used is far below that.

[mjrtoo]

well, it's still twice the bandwidth no matter how you slice it....

[L8NightHype]

Currently, nforce2 and via are it for amd. The argument concerning dual channel is irrelevent. It is proven in benchmarks and general use...period...both sets are perfectly good options either way you slice it.

I personally have Intel's 875 (dual) and she's a screamer...

L8

[Izdaari]

On nForce2 SPP chipsets (i.e., not using integrated video), benchmarks show dual channel makes about 5% difference in memory performance. That's not insignificant, but it isn't earthshaking either. It makes more difference, something like 20-30% if using the onboard nForce video.

And of course, with P4's, which are chronically starved from memory bandwidth, it makes more difference than it does for Athlons which are less dependent on memory bandwidth.

Sorry, I didn't bookmark any of those benchmarks. Just memories from a while back, so I could be a little bit off, but I'm pretty sure that's at least generally correct.

[NDD]

You can never have too much bandwidth :t :cool:

[causticVapor]

IF the athlon had an 800MHz, 64-bit FSB, then it would be able to splice 64-bit data alternately into two 400MHz, 64-bit channels, and thus theoretically maintain the same bandwidth. Another way would be to have a 128-bit, 400MHz FSB going to two 64-bit, 400MHz channels. Unfortunately, the Athlon (3200+) has a 400MHz, 64-bit FSB, so the northbridge's branch predictor algorithms have to work overtime, and they do actually give a boost.

There's really no way, though, to fill a cup with water if you only have half a cupful of water to fill it with. This is (one reason) why the P4 is so far ahead of the Athlon now.:(

Think of a dual memory as giving an athlon as big of a bandwidth boost as going from CL2 to CL1 ;)