MSI R7970 graphics card

This test is related with the review of HD 7950, as we changed test platform when doing power consumption test, so we had to retest it. At that time, though frequency and performance of HD 7950 is close to that of HD 7970, but the former featured much lower power consumption. In spite of four CUs disabled, it’s impossible to get such a big change, what happened with it?

Both based on GCN while with big difference in power consumption

The obvious difference between HD 7970 and HD 7950 is voltage, idle voltage is both 0.85V, but working voltage of HD 7970 is 1.174V, only 0.993V with HD 7950(displayed with 1.09V in Furmark, however, the result is incorrect, which is shown below.). That’s to say, 0.1-0.2V lower voltage  would lead to dozens or even a hundred watt decreasement in power consumption, while there comes with not too much performance loss.

The most commonly-used graphics adjustment software is Afterburner, which could get more voltage regulating settings when matched with MSI graphics card. Our test was based on MSI R7970-2PMD3GD5, we set various voltages and frequencies to test power consumption of the whole system, test items included graphics recycling loading test 4 in 3DMark 11 X mode and Furmark 1920×1080 loading. As two items featured small power consumption fluctuation which was quite stable in certain range.

Although Afterburner supports GPU core and memory voltage adjustment, here we only adjusted core voltage and frequency, while memory voltage kept at default value of 1.6V.

The test platform was basically consistent with that in HD 7950 review, it included Intel Core i7-2600K, which was overclocked to 4.5GHz at voltage of 1.408V, 8GB dual-channel DDR3-1600MHz memory, MSI R7970 graphics card. However, different from the last test, case and power supply were changed to Super Flower SF-1000 and Golden King 1000W power supply, respectively.

The voltage range was at 0.8V-1.3V with MSI Afterburner software, starting voltage was 0.85V set by AMD. Here we firstly solved voltage control issue. We used GPU-shark (just like Furmark), AIDA64, GPU-Z and Afterburner to monitor voltage change, Afterburner was mainly used for regulation, take 0.85V as example, the result was shown as follows:

Getting various GPU voltage results with different software

In Afterburner, the voltage displayed as 0.85V surely, 0.828V/ 0.829V shown at AIDA and GPU-Z, respectively. It displayed 1.17V with GPU-Shark, so we excluded the result in Furmark and GPU-Shark, that’s why we said it was inaccurate as we previously mentioned. In addition, the voltage shown in GPU-Z was 0.01V-0.05V lower then that in Afterburner.

Apart from whole system power consumption, it attached with GPU temperature and GPU Power draw. Temperature was got in Furmark loading test, indoor temperature was set at 20-20.5 ℃, while Power draw was calculated by Furmark through VDDC voltage and current detected by GPU-Z in real-time, which could be considered pure GPU power consumption. In digital circuit, P≈CV²F (V and F stand for voltage and working frequency, separately). That’s to say, compared with frequency, voltage improvement surely would bring more power consumption promotion. The following test could account for it, did voltage have how much influence on power consumption?

We simply classified the test result into three classes: low noise and power, better performance and enthusiast. Within 0.9V, HD 7970 could be overclocked to up to 800MHz, loading temperature was below 69℃, and fan rotation speed was about 2000RPM, we hardly hear any noise. As for system power, it was no more than 200W in 3DMark 11 X mode, while in Furmark loading condition, it was only 266W at most.

In the range of 0.95V-1.05V, graphics frequency could reach 1.05GHz, surely power consumption would rise to about 360W gradually, and GPU temperature improved to around 77℃. However, fan rotation speed also lifted to 2400-2600RPM, there featured obvious fan operation sound, but frequency was about 1G, so it was a balanced solution with better performance.

If we continued improve voltage, it won’t be easy to get frequency promotion upon voltage of above 1.10V, under 1.10V it could only secure stable operation at 1.05GHz. While it required to promote voltage to 1.112V in the event of 1.1GHz normal running, additionally it couldn’t pass Furmark loading at 1.15GH in spite of voltage at 1.3V.

In enthusiast level, it featured lower increasing rate of frequency, but power consumption soared by a large margin. Within 1.2V there didn’t feature obvious 3DMark 11 power advancement, while in Furmark loading condition, power of 379W at 1.10V soared to 475W at 1.2V. Once voltage was above 1.2V, 3DMark 11 power consumption also increased rapidly, however, in such state, Furmark loading couldn’t work, system would halt upon operation.

All in all, voltage increasement had much influence on power consumption of HD 7970, the gap was not so obvious when voltage was below 1.1V. After that, we went on improving voltage, there featured more narrow room for frequency improvement, while power consumption and temperature increased sharply, at which time the loss outweighed the gain.

Through the test, we found that core frequency of 925MHz at 1.17V set by AMD was quite conservative (maybe for securing absolute stability of graphics cards).

The best matched solution between voltage and frequency

In view of above test data, we recommended two matching solutions. Firstly it was 800/1375MHz@0.9V (at low power and low noise class). Under the frequency, its performance won’t be lower than that of GTX 580, total system power was 266W at most, 200W lower than GTX 580 in the same platform.

The following was the screenshot for Furmark condition after half an hour:

At 800/1375MHz@0.9V setting, it passed half an hour of Furmark loading test, which was in compliance with silent need.

For another grade, we chose 1000/1375MHz@1.05V, as it could be overclocked to 1GHz, loading power consumption was only 352W, and temperature was 76℃, in such condition, it only featured some fan sound, the noise was acceptable.

At 1000/1375MHz@1.05V setting, it passed half an hour of Furmark loading test with better performance.

When voltage was higher than 1.10V upon overclocking, it would be difficult to pass Furmark loading test. So it’s mainly suitable for enthusiastic players. The more appropriate setting was 1150/1375MHz@1.174V, actually its overclocking capability was more than that, in order to keep parameter consistently, we didn’t adjust fan rotation speed and memory voltage. For enthusiasts, it could be basically overclocked to 1.25-1.3GHz at 1.2V or even 1.3V, provided that we don’t care about the state of power consumption, heat and noise.