The Radeon HD 7950 DirectCU II from Asus, only just available in stores, suffers from a fault with the fixture for its cooler that makes the GPU heat up and can rapidly lead to crashes, especially if you’re overclocking your GPU. We recommend you to avoid this model while waiting for Asus to correct the problem and recall the faulty batches.
When we started work on our Radeon HD 7900 report (currently only available in French: translation to come), Asus was one of the first manufacturers to send us a semi-customised Radeon HD 7950, based on a stock PCB (which was however made in Asus factories) and equipped with an alternative cooler, the famous DirectCU II.
Barely had we set it up in our test casing before we noticed frequent crashes. When we looked at the card a bit closer, we noted that the big cooling system is only fixed to the GPU with four screws and that one of them didn’t have any purchase on anything. We screwed the screws to a maximum to try and compensate and noted that on top of this fault in our sample model, the pressure of the cooler against the GPU was very (too) light for each of the screws. We did however continue to run our tests as the crashes stopped and the GPU temperature reported by the sensor was relatively low.
The infrared photos showed however that the back of the GPU was getting extremely hot, up to 118 °C. There was obviously a problem and we decided not to include the tests for this model in the report before finding out what was going on. We shared our observations with Asus France, reporting that there seemed to be a pressure problem, in addition to a defective screw.
That was when things started to get complicated. On one hand in Europe, Asus told us, without fully spelling it out, that it had found the same issue on other cards and that adding a thin spacer under the screws improved the pressure issue, though it didn't admit clearly that there was an issue that required correction. We went back to them to get a clearer response: either there wasn’t a problem or all the cards on sale needed to be corrected. Our contact at Asus France seemed to be squirming somewhat when we insisted on getting an answer. We didn’t and Asus Taiwan played dumb, claiming our infrared photos were inaccurate. It was of course impossible to get hold of a second sample model to test...
The weeks went by and the cards began to appear in stores, still with no reaction from Asus. We suspected that there was a problem, though we weren’t certain as one of the screws was defective on the initial sample model. We therefore got hold of another card on sale in a store and ran our tests again. The first thing we noticed was that the card was identical in all ways to the first that we had seen (except for the faulty screw of course) and that the pressure of the cooler on the GPU was still as light as ever.
During the first part of our test protocol, on the bench table with the card in vertical position, things went fine. The first test in the casing, with the card in horizontal position, went okay as well... at least at first. After 30 minutes in load, without overclocking, and although the GPU sensor was reporting a low temperature, the system crashed. We started the test again, with the casing open, so we could observe the card temperature with our infrared camera. Surprisingly, it increased quite fast at the back of the GPU and once it had reached +/- 80 °C, which probably made softer the thermal paste used to link the cooler to the GPU, it quickly rose to over 115 °C and brought about another crash. The log of the GPU sensor however only showed 70 °C.
The last infrared photo we had time to take before the crash.
We’re convinced that the problem comes from the fact that the pressure of the cooler on the GPU is too light, as the hot area observed with the infrared thermography isn’t centred on the GPU but rather slightly to the right, or where most of the weight of the cooler is. To demonstrate this, we launched the test again, this time holding the infrared camera in one hand and holding the cooler with the other. To get the GPU to heat up faster we overclocked the GPU a little. As we expected, there was no longer any problem and the temperature at the back of the GPU was normal +/- 75 °C. When we removed our hand, with no support for the cooler however, the temperature increased by 20 °C in a few seconds and the system crashed within a minute. If however we applied a slight pressure (such as even only a slightly rigid cable would exert) on the top of the cooler, it detaches itself even further from the GPU and the system crashes immediately.
The weight of the cooler is sufficient to move it slightly away from the GPU, a problem that we accentuated by exerting slight pressure so that this was clearly visible in this photo.
It looks very much as if the Radeon HD 7950 DirectCU II has a design fault. While it’s strange that the temperature given by the GPU sensor doesn't show this (the sensor is perhaps situated on a side of the GPU die that remains in contact with the cooler, or the value that it reports isn't properly registered), the GPU does heat up as soon as contact with the cooler lessens. Such a fault can happen on any card to any manufacturer and is generally detected on the early samples. All that's then required is to rectify the problem.
What annoys us in the present case is that Asus seemed to be aware of the potential fault (indeed it was confirmed by certain members of the personnel internally) and yet chose to ignore it so it didn't have to delay the release of what was, among Asus fans, a much-awaited product. This behaviour is obviously very disappointing, especially as we’re talking a high-end model that is supposed to have been carefully refined to give superior quality!
We obviously hope that Asus will now take responsibility, recall all the cards, both standard and TOP factory overclocked models, not put them back on sale until it has resolved the problem and not market doubtful prototypes in the future! In the meantime we can only advise you to avoid this model.