Power Consumption
With 956 million transistors compared to 666 million for the RV670 (Radeon HD 3850/70) — both engraved at 55 nm — let’s see the variations in power consumption at the power supply (which includes, just as a reminder, the power consumption of the entire computer and the 20% lost by the power supply).
Consumption at idle has increased significantly, which is not a pleasant surprise – a total increase of 22 W (approximately 18 W attributable to the graphics card) for the same usage level (more or less nil), which is pretty disappointing, especially since it puts the 4850 back behind the GeForce GTX 200 for consumption at idle, which is a highly appreciated strong point of existing AMD cards. At idle, the frequencies of the 4850 drop from 625 MHz (GPU) and 1000 MHz (memory) to 500 MHz and 750 MHz, which is a much less aggressive throttle-back than on the GT200.
On the other hand, under load, running Test Drive Unlimited, the increase compared to the 3870 was no more than 12 W for a performance that was 26% better in this specific case, which is very good (and that’s despite the slight overclocking of the Asus test board, remember). Note also that the consumption peak we measured for the card (still running Fillrate Tester) was only 270 W.
But admittedly, again the 9800 GTX + did as well, and that cancels out AMD’s advantage. Note that the G92’s change to a 55 µm process does result in a slight reduction in power consumption, both at idle and under load (up to 8 W), which is predictable given the very slight increase in frequency. We should also underline the fact that in this specific test case (Test Drive Unlimited at 2560x1600), the 9800 GTX +’s performance/watts ratio was 11% better than the HD 4850’s.