Page 1:An Eye For Power
Page 2:Performance Per Watt
Page 3:The Tests
Page 4:Test Setup And A Side Note
Page 5:Test System
Page 6:Benchmark Results: Crysis, The Classic Approach
Page 7:Benchmark Results: Desktop Usage, Less-Than-Ideal Conditions
Page 8:Benchmark Results: Cinebench R11
Page 9:Benchmark Results: Cyberlink PowerDVD 9
Page 10:Benchmark Results: Cyberlink PowerDirector
Page 11:GPU Vs. CPU
Page 12:Measuring Power Consumption: Let's Recap
Page 13:Don't Forget Idle Power Consumption
Benchmark Results: Cyberlink PowerDirector
Next up is Cyberlink's PowerDirector. We actually tested all Stream-supported filters for this article. However, for simplicity's sake, we’re only displaying results from four filters (Bloom, Chinese Painting, Kaleidoscope, and Light Ray), while encoding H.264 video with the default profile (AVCHD 1080p, 15 Mbit). We've selected the aforementioned filters because they seem to respond most readily to GPU acceleration.
With a Radeon HD 4670, GPU utilization stays below 50% most of the time with just GPU filters. The Radeon HD 4670 power consumption levels here are actually quite close to levels when running Crysis (172 W).
We see around 60-70% utilization on both the Radeon HD 5670 and HD 5770 during these tests. Any GPU supporting the use of hardware encoding really stands out, offering very significant improvements beyond CPU-only performance (in our case, with AMD’s Phenom II X4 955 BE).
On that note, even the Radeon HD 3300 offers some savings, but the benefits are limited to filters that it can handle. In this case, only the Kaleidoscope filter qualifies. The three other filters don't show much improvement. One thing to note is that, although there’s virtually no difference in the results, CPU utilization is indeed lower with the Radeon HD 3300. Employing integrated graphics, the CPU doesn’t have to decode the stream and process the filters.
Looking at the total power consumed, both the Radeon HD 5670 and HD 5770 consume roughly the same amount of power throughout the entire process. So, unlike in Crysis and Cinebench, the Radeon HD 5770’s additional performance offsets the slightly higher power consumption. We wouldn't be aware of this fact if we only looked at average power consumption.
We made an interesting observation about the two Radeon HD 5870 cards in this test, too. Instead of operating at full-speed, both boards run at UVD clocks (400/900 MHz). This allows the cards to manage their power consumption very cleverly. The Radeon HD 5870 1 GB is only consuming about 15-21 W more power than the HD 5770. When performance is taken into account, total power consumed between the HD 5670, HD 5770, HD 5870 1GB, and HD 5870 2 GB is similar. The Radeon HD 5870 1 GB edges out the others, but the lead is very small.
On the flip side, there is another lesson to be learned here. Although the Radeon 2900 XT is still faster than a pure software approach, the higher power consumption of the system with the card actually results in higher total power consumed. This is a prime reason why reducing graphics-oriented power consumption is so important in applications that don’t fully utilize the hardware. With its latest-generation boards, AMD has done very well in this regard. Based on the Chinese Painting and Light Ray numbers, a modern GPU architecture like the one used to create AMD’s Radeon HD 5000-series actually cuts down your power usage by more than half compared to the older design.
- An Eye For Power
- Performance Per Watt
- The Tests
- Test Setup And A Side Note
- Test System
- Benchmark Results: Crysis, The Classic Approach
- Benchmark Results: Desktop Usage, Less-Than-Ideal Conditions
- Benchmark Results: Cinebench R11
- Benchmark Results: Cyberlink PowerDVD 9
- Benchmark Results: Cyberlink PowerDirector
- GPU Vs. CPU
- Measuring Power Consumption: Let's Recap
- Don't Forget Idle Power Consumption