R480/430 roadmappery

[joe emo]

"by the way, the resistance of a transitor is decreased with increased temperature. "


that's sortof true, but the effect is much greater for intrinsic semi-conductors vs. extrinsic. Conduction in intrinsic semi-conductors is highly dependant on T [ varying as exp(-Eg/2kT) ], increasing as T increases, however extrinsic semi-conductors have an extrinsic zone where T dependance is quite mild, until excessive T is reached and intrinsic properties take over.

 

[Entropy]

By the way, guys, there is the additional issue that when you increase the temperature of a resistor, its resistance typically increases. This is where you get the power goes as frequency squared, I believe (well, approximately).

There's also the additional issue that transistors aren't actually linear in their response (i.e. V=IR does not apply exactly, only approximately), so that may also be a contributing factor. I'm not actually sure which factor dominates the faster-than-linear increase of power consumption with frequency.

Summing up, there's a lot of issues raising their ugly heads, interacting even, which is ultimately why you measure the damn things instead of just using ever more sophisticated calculations.

Thing is, manufacturers look at frequency vs. power vs voltage curves like the one above, try to take yields into account, market priorities, competition + god knows what else, and decide how the device should be presented to the market in order to maximize profit.

My point is that in the gfx-ASIC business, as well as CPUs, the market is pretty immature, and has been trained to respond to a single stimuli - performance. This was great for as long lithography advances allowed Intel and others to regularly increase that parameter (and applying a single, easily recognizeable figure of merit - Hz). Lithographic technique hasn't allowed as rapid increases in performance lately, and in order to keep the purchase numbers up, manufacturers have felt compelled to creep closer to the right hand side of the curve. (As well as the whole thing shifting towards higher power draw as well.)

Now, these decisions are marketing more than technology driven. A given gfx-ASIC will be clocked differently in the mobile space, where battery life and cooling are more critical parameters for the consumers.
So conceivably, if there's a desire for lower power draw in the market, eventually clever manufacturers might pick up on that and try to offer devices that target that segment even in the desktop space.

I don't really believe the 430 is such a device, I believe it is simply a much cheaper part to produce on TSMCs bulk 0.11um process, than the 480 will be on 0.13um low-K and probably intended to be the new R360. ATI will be able to drop prices on it as the market dictates, just as they have done with the 9800 line of products.

 

[KimB]

No.
P = U*U/R -> linear voltage increase leads to quadratic power increase.

by the way, the resistance of a transitor is decreased with increased temperature.

Yeah, I guess you're right. My bad. But if voltage is kept the same, that means that power increases when resistance decreases (assuming the majority of the resistance is made up by the transistors). Makes even more sense now. I believe this can be often understood as an increase in temperature means an increase in leakage current, though that's only one portion of the whole picture.

Edit:
Ack, forgot about this:

that's sortof true, but the effect is much greater for intrinsic semi-conductors vs. extrinsic. Conduction in intrinsic semi-conductors is highly dependant on T [ varying as exp(-Eg/2kT) ], increasing as T increases, however extrinsic semi-conductors have an extrinsic zone where T dependance is quite mild, until excessive T is reached and intrinsic properties take over.

Yes, that's right, the number of mobile electrons/holes in a doped semiconductor can be often approximated as just the number that have been added by doping. Forgot about that. Wow it's been too long since I took that electronics class.

Anyway, I think that in the end, there are a large number of factors that lead up to the total power consumption (capacitance, inductance, temperature variance, etc.), which, in turn, results in a greater than linear dependence of power on frequency. This is obvious, for example, if you just look at some power consumption graphs.

Yes and no.

Yes, transistors are not linear but no, it doesn't matter here.
We're talking about digital devices where transistors only have two different states: ON and OFF.

Those are only approximate states. I added this as a qualifier for the simple reason that it is known that current microprocessors are really starting to push the limits of the physics, so I thought it was possible that the inherent nonlinearity, which has been ignored for a very long time, may start to become an issue.