Trinculo

As to your first point, see the recent threads regarding fab problems both AMD/ATI and Nvidia have been having with their new chips. While historically the price drops have been there.. Until the folks doing the fabrication of the actual chips get their acts together, I don't expect that to be the case, at least for a little while

As to your last point.. very very valid. The overall system performance is going to tend to be constrained by the first bottleneck. Improvements outside of that area will result in very little if any change in performance. Improvements TO the bottleneck will have a very direct and measurable result, right up until something else becomes the bottleneck. This is basic "Theory of Constraints". The tricky part is figuring out where the constraint is, especially when it's not always easy to instrument things like how busy the GPU is.

My general rule of thumbs are:
1) if most of your memory is used when CoX is running, you ought to close down other stuff that's running, or get more ram. (up to the limit of what your OS supports, which is just over 3 GB for most 32 bit OS's)
2) If at least one CPU core is at or very near 100%, when CoX is running, then getting a faster CPU will probably help.
3) If neither CPU or Memory are maxed when CoX is running, the faster video will probably help.

So if TSMC has so badly messed up with both AMD/ATI and Nvidia, it sounds a lot more like a situation where the first of them to come to agreements with a new Fab partner might be the winner in this whole debacle. Or is TSMC the only game in town for cranking out silicon in the 40nm process?

Great info BTW thanks for taking the time to type that up. For the un-initiated, the material for silicon chips is basically a giant crystalline structure that is 'grown' by being pulled out of molten doped silicon material (http://en.wikipedia.org/wiki/Czochralski_process) and there's pretty much a limit to the diameter of the stuff (currently around 30cm). That's then sliced into thin polished 'wafers' that are the raw material used to produce the chips. The smaller the traces on the chip (currently measured in Nano-meters) the smaller each chip will be (and also the less power it consumes, heat it generates, etc). The more transisters on the chip, the larger it will be. And of course the smaller scale you are trying to work at (e.g. 40nm vs 55)nm, the more precise everything has to be and the smaller the margin of error in order to produce working chips.

The 'per wafer' fab expenses and fab time tends to be relatively constant, so the more (working!) chips you can get per wafer (yeild), the cheaper your production costs per chip, and also the higher your potential output . So the drive to get to a viable 40nm process is huge, but doesn't pay off pricewise (and capacity wise) until your process is reliable enough to get more working chips per wafer. Especially when you consider that you have to test each chip, and don't get paid for the ones that don't work. 40nm has thus far 'backfired' because the yeilds have been so low, that the costs have been higher, and output lower, something nobody is happy about.

Both Nvidia and ATI need 40nm process chips in order to keep cards inside reasonable limits for power consumption and heat output, and the new stuff has all been designed around that, so it's likely very impractical for them to 'punt' and just use 55nm process stuff until the bugs are worked out of the other fab lines.

OTOH, the older 55nm stuff is reliable fab wise, and since that process is mature, the yeilds per wafer are pretty good, meaning it's probably a lot cheaper (until those 40nm yeilds get better) Especially if you've managed to pay off your investment in the design of those chips.

The result of that is that I'd personally expect both companies to hold the prices of newer stuff that uses 40nm process a lot higher, especially if there's not currently capacity to generate those parts in quantity anyway (no point to lowering the price to sell more, if you don't have the 'more' to sell) Meanwhile I'd expect the prices of the older stuff to stay low, and represent a majority of their sales.

I think you mean AMPS where you say Ohms.. an Amp is a measure of current flow, or in the case of a power-supply the amount of current it can support and still maintain the proper voltage etc.

An Ohm is a measure of resistance or impedance, You're more likely to run across that when selecting speakers for a home theater system than you would be looking at ratings of power supplies.

Also be aware that many higher end power supplies have a PAIR of 12v rails, to try and distribute the load. That can be really handy for folks with a lot of hard drives, with draw high power when spinnng up. I'm not sure in that case if you can (presuming the card has two 12v connectors) connect both rails to the video card in order to provide it with enough power, of if that would be a 'bad idea'.

Used to be I had to worry about he power drawn by the CPU, and keeping it cool. These days that's small potatoes compared to the power and cooling needed by the video card. Of course most video cards currently represent a massivly parallel computing capacity that would put many older mainframe systems to shame, so i guess that's not entirely unexpected.

(Times like this I shake my head and wonder how the hell I ever got by on an 8086, 640K of memory, and a 20Mb hard disk. Then again my little Sansa media player probably has more computing power than that old Compaq Deskpro did. It's enough to make a geek feel Ancient it is.. _