NVIDIA and ATI HDCP Compatible Graphics Cards Roundup

CPU Utilization

Of course gaming performance is only part of the equation when it comes to looking at these HDCP compliant cards, the other major aspect is CPU utilization during high definition movie playback. Today we're only able to provide a small subset of HD movie playback performance as we're only testing with a MPEG-2 encoded Blu-ray title. We're still waiting for a PC HD-DVD player which will let us test VC1 and H.264 decode performance as well, but for now we're only able to look at high bitrate MPEG-2 content. VC1 and H.264 encoded content will put more stress on the CPU and GPU as a whole, but we'll unfortunately have to wait a little longer before testing it.

Just like when graphics cards started becoming important for offloading graphics processing with games like GLQuake, we are in a kind of transition period where it is becoming necessary to also have cards that can process our video playback for us. For the past couple of years ATI and NVIDIA products have been handling video decode acceleration, but it hasn't started to be really necessary until HD-DVD and Blu-Ray came around. The complex video formats they provide require more processing power to decode, meaning that slower processors won't be able to play them back without help from a graphics card.

Right now, since Blu-Ray titles are predominantly MPEG-2, having lots of extra power in a graphics card to accelerate the decode process isn't extremely important, but we still want to take a look at how much load the cards can take away from the CPU. With this in mind we put together a benchmark, recording the average CPU utilization of a period of about one minute of Blu-Ray movie playback. The movie we used was Click, and we tested each of the cards with the exact same one-minute segment of the movie. Audio was also enabled for this test.

Here are the CPU utilization results from each of our cards.

CPU Utilization	Avg	Min	Max
NVIDIA Gigabyte GeForce 7600 GS	51.5	41.4	58.2
NVIDIA ASUS GeForce EN7600 GT	45.5	38.8	50.8
NVIDIA MSI GeForce NX7600 GT Diamond Plus	46.9	38.3	52.9
NVIDIA MSI GeForce NX7600 GT	45.8	39.1	51.6
NVIDIA Albatron GeForce 7900 GS	45.8	36.7	54.7
NVIDIA EVGA e-GeForce 7900 GS KO	44.5	37.5	52.3
NVIDIA Leadtek WinFast PX7900GS TDH Extreme	44.8	36.7	51.6
NVIDIA MSI GeForce 7900 GS	45.9	38.3	52.3
NVIDIA MSI GeForce NX7900 GT	44.9	38.3	51.6
NVIDIA EVGA e-GeForce 7950 GT KO	43.9	35.9	50.0
NVIDIA Gigabyte GeForce NX7950 GT	44.4	36.7	51.6
NVIDIA PNY GeForce 7950 GT	44.3	36.7	52.3
NVIDIA XFX GeForce 7950 GT HDCP	44.1	35.2	53.1
NVIDIA Sparkle Calibre 7950 GT	44.1	35.9	64.1
NVIDIA BFG GeForce 7950 GX2	46.3	36.7	53.1
NVIDIA EVGA e-GeForce 7950 GX2	46.2	39.8	53.1
NVIDIA GeForce 8800 GTX	38.7	29.7	46.9
NVIDIA GeForce 8800 GTS	39.8	31.2	48.8
ATI Powercolor Radeon X1600 PRO HDMI	40.6	28.1	50.0
ATI Sapphire Radeon X1950 XTX	36.3	28.9	44.5
ATI Radeon X1900XT 256 (reference)	34.2	28.1	39.8
ATI Radeon X1650XT (reference)	38.3	28.1	46.1

Video decode acceleration on NVIDIA GPUs is handled by the PureVideo processor, which is tied directly to the core clock speed, so the CPU utilization of each card will reflect this. The end result is that an NVIDIA card with more pipelines that is better at 3D performance will not necessarily be better at video decoding. With ATI, its AVIVO decoding is also tied to the processing power of the card, but is not quite as related to the clock speed as it is with NVIDIA. We also found that there was a bit of variance between multiple runs of the same tests, but these tests give us a general view of the CPU utilization of each of these cards.

We can see that the X1900 XT 256 gets a very low average CPU utilization compared to the other cards. Also, the 8800 GTX and 8800 GTS offloaded more processing from the CPU than the other NVIDIA cards, which isn't very surprising given that NVIDIA mentioned that the PureVideo core is a bit faster in G80. For reference, we measured the CPU utilization of the Blu-Ray playback benchmark with hardware acceleration disabled, and we got an average of 51.0%, giving us an idea of how much work these graphics cards take off the CPU. The Gigabyte 7600 GS doesn't seem to help in this area at all, and it makes sense when we consider that it's the slowest clocked NVIDIA card of the group. It would appear that a 400MHz clock speed doesn't provide enough power with PureVideo to make a difference in CPU utilization.

Even taking into account these results, CPU utilization isn't going to make a big difference between which of these cards would be better choices than others. Until we can look at H.264 and VC1 decode performance we will have to focus on other important factors to consider such as power, heat and noise.