Power Consumption, Temperature, Noise, Overclocking
The single-processor Radeon HD 5870 did well in our power consumption tests and delivered excellent results considering its high performance. It is going to be interesting to see how much power the Radeon HD 5970 needs. We perform our power consumption measurements on a special testbed configured like follows:
- Intel Pentium 4 560 CPU (3.6GHz, LGA775)
- DFI LANParty UT ICFX3200-T2R/G mainboard (ATI CrossFire Xpress 3200 chipset)
- PC2-5300 SDRAM (2x512MB, 667MHz)
- Western Digital Raptor WD360ADFD HDD (36GB)
- Enermax Liberty ELT620AWT PSU (620W)
- Microsoft Windows Vista Ultimate SP2 32-bit
- Futuremark PCMark05 Build 1.2.0
- Futuremark 3DMark06 Build 1.1.0
Our measurement method is somewhat outdated when it comes to modern graphics cards but we have to use Futuremark 3DMark06 as the 3D load for the sake of comparison. It is for this load that we have accumulated a large amount of data on the power consumption of lots of graphics cards. Comparing new results with older ones would be impossible if the measurement method were changed.
Anyway, we are already working on building a new testbed that would fully automate the measurement procedure. We are going to retest all graphics cards we have at our disposal on the new testbed to get enough data for making comparisons with newly released products.
So, following our old method, we used the first SM3.0/HDR test from 3DMark06 to create 3D load. We ran it in a loop at 1600x1200 with forced 4x FSAA and 16x AF. Additionally we loaded the cards by running the OpenGL-based FurMark. The Peak 2D load was created by means of the 2D Transparent Windows test from PCMark05. We’ve got the following results:
According to 3DMark06, AMD’s new dual-processor flagship is even more economical than the Nvidia GeForce GTX 295. We are not surprised, actually, after the Radeon HD 5870, and the measured power draw of the Radeon HD 5970 is about the same as we anticipated. As usual, the peak power draw in FurMark is higher but only by 12.4W. Perhaps there is some protection mechanism in the ATI Catalyst driver that must safeguard the device under extremely high, unrealistic loads. Although overclockers and benchmarkers may argue, we do think that the OpenGL FurMark test is an unrealistic load for a graphics card.
It is the 8-pin PCIe 2.0 connector that is loaded the most: up to 120-125 watts. The 6-pin connector is loaded less. The load on the power section of the PCI Express x16 slot is low. Summing it up, the Radeon HD 5970 is not an economical device but you can’t expect that from a premium-class solution. As a matter of fact, this card sets a new record of power efficiency in its price category.
The Radeon HD 5970’s new cooler does well, according to Catalyst Control Center:
We tried to use the latest version of GPU-Z available at the time of our tests (0.3.6), which can monitor a graphics card’s temperature too, but it made our testbed hang up. Anyway, the temperature of 55°C in idle mode and 75°C under load during a long session of Crysis Warhead is an excellent result. GPU-Z version 0.3.7 reported that although the second (by GPU-Z’s reckoning) GPU was indeed 75°C, the first GPU was 86°C hot, notwithstanding the common heat chamber. Anyway, even this result is very good for such an advanced solution like the Radeon HD 5970. It is sad that AMD did not use the single-heatsink solution with a common vapor chamber earlier. Perhaps this would have made the Radeon HD 4870 X2 more popular among demanding gamers some of which, being disappointed with the card’s acoustic characteristics, had preferred to build a discrete CrossFireX subsystem with two Radeon HD 4870 and, later, Radeon 4980.
Next we measured the card’s noise with a noise-level meter Velleman DVM1326. The reference point for our noise measurement tests is 43dBA which is the level of ambient noise in our test lab as measured at a distance of 1 meter from the testbed with a passively cooled graphics card inside. When we installed the tested graphics card, we got the following results:
We can’t say that the Radeon HD 5970 is bad in terms of noisiness. It is virtually silent in 2D mode and you can’t hear it against the noise from the other components of a closed system case. But as soon as we switch to a 3D application, the noise grows louder and you can easily hear the card. Well, we could not really hope for a dual-processor premium-class graphics card with over 4 billion transistors on board to be absolutely silent.
The noise spectrum of the Radeon HD 5970’s cooler is quite agreeable to the ear. In fact, it is the hissing of the air passing through the long heatsink. There is no rattling of the fan that used to irritate many owners of older Radeon HD cards.
Then we also investigated the overclocking potential of our sample of the new card. And at first we managed to increase its clock rates to 800MHz for the GPU and 1180 (4720) MHz for the memory. After that AMD kindly offered us a tool for software control over the GPU and memory voltages within a small range.
Using that tool, we achieved much better results:
We benchmarked our card at these overclocked frequencies, too. By the way, many reviewers tested the new card at the maximum speed of its fan, but this approach is not practical. The high level of noise makes it uncomfortable to be near the computer for everyone save for deaf overclockers. Of course, the default cooler can be replaced with a liquid cooling system. It would allow to achieve better overclocking results without hurting your hearing.
Now that we’ve examined our Radeon HD 5970 in all detail, it is time to test it in real games.