DDR400

[time]

LOSTCIRCUITS has a comparison of eight different PC3200 modules. It's particularly interesting for some of the technical explanations that Michael Schuette includes.

In summary, CAS latency has become rather unimportant for both data transfer and operating frequency. It is only the legacy of single data rate SDRAM and the fact that CAS is easier to pronounce than, for example, tRCD (Tee Ar See Dee) that is the main incentive to hi-light the products as CAS-2.

That still does not mean that all SPDs are read correctly now. Many boards read the frequency specs of the memory modules and some of the latencies but apply latency settings for e.g. PC2100 operation simultaneous with an operating frequency of e.g. DDR333 or DDR400. Needless to say that this will often enough cause POST failure, particularly with more expensive, faster modules.

That is, we are sticking with CAS-2.5 since the majority of modules would not support the CAS-3 setting to begin with for the simple lack of necessary pipeline stages in the DRAM output path

I've run some tests that confirm his assertions, at least with my setup. CAS 2.5 and 3 perform identically on my sample, suggesting it only really supports 2.5 anyway. And going from CAS 2.5/3 to 2 has a barely measurable effect - tRCD is way more significant (although still so small I could not care less).

I just picked up some Transcend DDR400 modules. They use Samsung chips, and by running an Epox 8RDA+ asynchronously, I was able to post
at 247MHz. I ran 3DMark 2001 at 243MHz (in these conditions the FSB is only stable to 182MHz). Not too shabby, I thought.

It's all academic for me though. As Jake The Dog showed, synchronous is the way to go, and the Epox is only stable up to 192MHz in this mode - unlike Jake's Soltek at 205MHz.

 

[blakerwry]

on my kt266a, it only supports up to 266mHz FSB/RAM operation... by going from cas3 to cas2 I notice a measurable difference in benchmarks that are dependant on memory (Encoding, and 3D benchmarks).

The difference isn't much, a few percent at best, but why not eek out the most you can?

 

[time]

Yeah, but at higher speeds it seems to fade into the background.

Corsair makes a living selling so-called "High Performance" DRAM for 2 to 3 times the normal price. I think the article shows that their marketing is smoke.

The fact is, vastly cheaper modules equalled or bettered their performance, even without heat spreaders(!) :roll: I think this is an important lesson.

 

[blakerwry]

Yeah, but at higher speeds it seems to fade into the background.

Corsair makes a living selling so-called "High Performance" DRAM for 2 to 3 times the normal price. I think the article shows that their marketing is smoke.

The fact is, vastly cheaper modules equalled or bettered their performance, even without heat spreaders(!) :roll: I think this is an important lesson.

I tend to agree. People who buy Hyper-X or Corsair XMS modules are generally of the "beach bum" overclocker type. If those are the only sticks that meet my timing requirements I would buy them, otherwise there's no reason to spend extra on heat spreaders and brand name when that extra could be buying me something more beneficial.

 

[blakerwry]

btw, corsair uses alot of micron modules... Kingston uses Samsung... not sure what each use in their "high end" stuff.