Dilemna One: Hippedy-Hoppedy Pricing
Memory prices have been declining even on the high-end lately.
For instance, the price of a stick of Corsair 256Mb PC3500 RAM has dipped about $20 in the last month or so.
More competitors have shown up in at least the PC3200 division lately: Samsung, Kingston and Kingmax have DDR400 sticks out and even Micron is now at least making 5ns memory modules.
The increase in company is no doubt due to Intel wanting PC3200 chips to fuel their 200MHz FSB mobos.
It also makes price predictions on these chips more than a bit iffy. Memory makers pounce on any excuse to raise prices, and the leading song in their hit parade is “Due to increased demand because of Intel.” It sounds even more plausible than usual because dual DDR needs two sticks rather than one.
So the price will likely head downward for at least a little while, then jump up sharply in anticipation of the Intel introduction, then probably slump even more over the summer as everybody piles on to DDR400.
When will the jump occur? Who knows?
Dilemna Two: No Room To Breathe
For AMD overclockers, they’re pretty well set memory-wise. Even if things get no better performance-wise than they are now, they have a good deal of overclocking room, and can always play with the multiplier.
Intel overclockers have a much grimmer choice. Those who buy 133MHz FSB chips will have plenty of FSB overclocking room, but not as much memory bandwidth as those with 200MHz FSB chips and no hyperthreading.
Those who buy the new 200MHz FSB chips will have hyperthreading and tons of memory bandwidth, but little overclocking room.
For example, the most reasonably-priced HT chip, which will go for about $175, will run at 12X200. Presuming you can get 240MHz out of two memory sticks (which is probably asking a lot), that only gives you 12X240 = 2.88GHz.
With that kind of math (and presuming the PIV get a bit better), one starts to look longingly at the 2.8GHz, until one looks at the price tag.
On The One Hand, On The Other
Current high-end memory modules today are essentially overclocked 5ns memory modules (yes, I know a few are claiming “4.5ns,” but I can’t find any real memory manufacturer (no, neither Geil nor OCZ are real memory manufacturers) claiming anything more than 5ns in their datasheets).
A real 4.5ns (or even better 4ns) memory module would be able to handle the overclocking speeds needed to push the 200MHz chips to their max (though how far one can push dual DDR chips is a big question).
The other, more likely possibility, is that the next generation of Intel motherboards will be able to run memory slower than FSB. I’ve gotten some early reports which indicate that this could well be the case.
What I’ve heard suggested in a 6:5 ratio, which means FSB at 200MHz/memory at 166MHz. This would make some OEMs very happy because while they’d have the additional cost of two sticks of RAM rather than one, they could at least buy two sticks of cheaper PC2700 RAM.
This would be even better for an overclocker. A 6:5 ratio could mean an FSB of 266MHz and a memory speed of 220MHz. Given that the Intel mobos are supposed to make the two sticks play well together at 200MHz, 220MHz looks like a doable stretch.
The Obscurity To Look For
What’s important to the rest of the world isn’t always important to overclockers, and vice versa. A 40% price drop in a pricey Intel processor can be a big deal to nonoverclockers, but is laughable to an overclocker who got himself a 70% price drop for that speed a couple months before that.
Here, what’s unimportant to Joe Sixpack is vital to us. The obscure item to look for in the Springfield/Canterwood reviews are the FSB/memory ratios.
If they don’t allow memory to be run slower than the FSB, then the average overclocker should avoid 200MHz FSB PIVs until truly faster memory becomes available.
If they do, then 200MHz FSB chips become a viable overclocker option immediately.