Power Consumption: TDP vs Reality

Now we have done a lot of measurements and often we pointed to the TDP, the Thermal Design Power. At this point we have to say, that the Thermal Design Power is an Information that doesn´t matter for the end customer. CPU manufacturers have already stated many times, that this information only matters for manufacturers of cooling solutions and for nobody else. With the aid of the TDP those manufacturers should be able to create cooling solutions that can handle the power consumption of the according processors.

But in case of graphics cards it seems to be a little bit different, even if it should not. Standard cooling solutions of graphics cards are manufactured by contract manufacturers. The products will be delivered with these cooling solutions that should sustain the board power. But also third party cooling solution manufacturers need this TDP-value. Currently many of those third party manufacturers fail to develop working products for the new GeForce-GTX and the Radeon HD 4870 series. But why? One reason could be that the specified TDP values simply don´t seem to be correct.

Modell	TDP	Real
Radeon HD 4870 X2	286 W	373.1 W
Radeon HD 4870	160 W	187.2 W
Radeon HD 4850	110 W	148.2 W
Radeon HD 4830	110 W	92.3 W *
Radeon HD 4670	59 W	64.2 W
Radeon HD 3870	105 W	123.9 W
Radeon HD 2400 Pro	25 W	15.5 W *

Modell	TDP	Real
GeForce GTX 295	289 W	316.5 W
GeForce GTX 285	183 W	214.1 W
GeForce GTX 280	236 W	226.0 W
GeForce GTX 260	182 W	166.2 W
GeForce 9800 GX2	197 W	268.1
GeForce 9800 GTX	156 W	186.4 W
GeForce 8800 GT	110 W	111.7 W
GeForce 9600 GT	95 W	68.5 W *
GeForce 8500 GT	40 W	26.5 W *

At this point we can say, that a lot of products in this comparison don´t have any problem with the specified TDP even when using the bad boy FurMark. But the situation seems to change in higher performance-classes where a lot of cards exceed their TDP. But not in every Paper AMD and NVIDIA are speaking of a TDP. Sometimes you can also find the term “Maximum Board Power”, but how should this value be treated, if not as maximum?

But are these to the press communicated TDP-values really those values that are used by cooler manufacturers to design their cooling solution? We don´t know. But we can assume from our experience that third party cooler manufacturers use these values for designing their cooling solutions. The one or other could be caught on the wrong foot this way.

Our measurements show clearly that the TDP-values shared with the press are exceeded noticeably in many cases while using FurMark as load. Manufacturers like AMD and NVIDIA couldn´t know that a tool like FurMark appears? Thats true. They couldn´t know that such a tool appears in the public. But one ought not have lost sight of the fact that the manufacturers – or better their engineers – know better than anybody else how to put a maximum load to their own hardware. The programmer of FurMark might have had an good idea how to stress the hardware, but it's safe to assume that the manufacturers know how to stress their hardware at least the same way (and probably even more).