Hello, I have a Sapphire HD4870 (the latest one, with the black fan case and the fan in the middle, a non-reference layout), and I had several problems with it unmodified (750/900 GPU/MEM stock): 1. It worked ok with stock values and stock cooler, even if a VPU recover kicked in sometimes. As HD4870 is supposedly easily over clocked, I started playing with CCC a little bit. I couldn't reach anything amazing, even with fan speed at 100% (see below), 780/960 was almost the best I could get without crashes every 10 or 20 minutes. 2. The fan noise (from 60% to 100% speed) is just unbearable unless you play with loud sound and earplugs, but that would mean to become deaf really fast... 3 intake 120mm fans, 3 exhaust 120mm fans, CPU fan and PS fan, and this single 80mm fan is of course making more noise than all the others together. So, I basically had a not so bad HD4870 working ok at stock values, that I could over clock just a little bit, with the problem of the fan noise. Needless to say I was not really satisfied, as I have drooled on great over clocking values that you can find on different sites, or even stock values from some factory OC'ed cards. Soooooooo, as the temperature values from GPU-Z were a little bit high, I decided to get an after market cooler. Arctic-Cooling Accelero Twin Turbo. Installation : The main GPU heatsink, and the RAM heatsinks are really obvious to install, but it wasn't really clear where to put the VRM ones (remember, non-reference layout). I ended up using my dremmel (cutting tool) on the 4 VRM heatsink included to cover all the remaining places on the card. 1st Test : Awful. All temperatures raised, and got a VPU recover after only 10 seconds of small GPU stress. That was with the pre-applied thermal patch on the main heatsink. 2nd Test : Removed everything again, cleaned correctly both heatsink and GPU (asaka cleaning stuff with 2 different products). Applied Arctic-Cooling MX-2 (ceramic, non conductive) correctly, and I finally managed to let my card run for more than 1 hour with 750/1100 values. The DISPIO T° is not going over 68°C, the SHADERCORE T° is reaching 75°C and the MEMIO is reaching 92°C after one hour. Idle temperature are DISPIO/SHADERCORE/MEMIO = 40/45/55 I'm quite glad right now even though I didn't have time to test GPU OC yet. But, I would still like to get any opinion on the temperatures (especially MEMIO). If I understand it right, the MEMIO sensor is on one corner of the GPU (can't remember which one). So does a high T° mean the RAM is too hot(the RAM heatsinks are not even hot at all) or does it mean that the main Heatsink is not correctly applied on the GPU ? Thank you for any answer/advice.