Quote Originally Posted by matose View Post
Ok, so I tested some more GTX 580 with and without IHS. If you make good contact (Freeze or GC-Extreme thin layer, backplate and high pressure) and have a good card without IHS you can do ~1525MHz 3DMark03, ~1450MHz 3DMark11 and <1400MHz Heaven.
In the same condition but with IHS you gain ~125MHz in low intensity benchmarks (3DMark03, 3DMark01, 3DMark05, AQ3,etc) and more in high intensity ones (Vantage, 3DMark11, Unigine). The only card that have IHS is a Matrix that's a poor clocker on air and on LN2, but I managed to do 1520MHz 2k3 and 1375MHz Unigine easy...

The way to go is definately with IHS but I still don't understand why? The question is how do I mount now the IHS back on my very good Lightning?!
I have yet to test a 580 Lightning, but on the 480 Lightning, no IHS appeared to result in better core frequencies while the memory became pretty much un-overclockable. Actually did a test one step further with an IHS and a shim on top of the IHS, and I ended up getting stunningly high memory clocks, higher than 1400MHz, whilst normally the card would struggle to break past ~1220 on LN2. So, it appeared that the memory clocked worse the closer the pot was to the PCB. With the IHS off, not even 1200 was doable, with the IHS, a bit past 1200, and with a shim on top of the IHS, over 1400. It might not necessarily be the memory causing problems in the case of the 580 Lightning, but my guess would be that it's something on the PCB that's getting just a little too cold because the pot is closer to it when IHS-less. (yes, that 2mm is significant)

It might be worth a shot to pull out the hair dryers when going without the IHS. It's no secret that it helped the 480L.