Honestly, you probably won't see too much improvement over stock. MSI has been known to occasionally improperly apply TIM at the factory, but that's rare and only for MSI.
To start off, AS5 is VERY BAD EVEN FOR CPUs. It's pretty old, and though it has good performance, it has a LOOONG cure time. Modern pastes like MX-4 have the same performance, but it's safe to use on chips without an IHS (like GPUs) and there's very little cure time. I'd recommend MX-4 or PK-3. I also hear that diamond particle stuff is pretty good on CPUs as well, but I don't know if it's capacitive or conductive. They should warn you on the product page about things like that.
Same as the best method for CPUs, a small dot in the middle. Use a much smaller dot, you'll cover the whole chip with a very small dab. Perhaps a blob 1/3 the size of a grain of rice for a GPU.
AS5 is unsafe for GPUs and chipsets. While it's not conductive (that would be very bad), it is slightly slightly slightly capacitive. This could short the GPU infrequently, and eventually end up killing the card. Basically, whatever's recent and cheapest. Cooler Master has some pretty bad ones, but the cure time is near instant. I think MX-4 is only like an hour compared to the strict two weeks of exacting power cycles for AS5 curing.
here is a roundup of popular compounds until Jan 2012. They're used on CPUs with an IHS using the grain of rice method. It's been proven that the grain of rice method works best by that same site here:
Less is more in this case. The tiniest dot method didn't even cover the whole CPU IHS surface, but still tied for first place. Slathering it on and using one or two lines were of varying degrees worse than the tiny dot and small dot method. I personally would use enough to cover the whole die, but just barely, so it doesn't clog up the heat transfer.
The site I mentioned didn't seem to test MX-4, but the PK-3 is listed as "prolimatech thermal compound" which is at the very top of the delta graph. I've read reviews that bash on AS5 and Ceramique with results that don't wait for the Arctic cure time. MX-4 and PK-3 perform well in these tests, showing the Arctic Silver weakness. Even when cured (in Hardware Secrets testing) Ceramique was worse than PK-3. Worse performance, AND worse cure time? Sounds like a real winner. Still, Ceramique and AS5 is cheap, and I justify the $4 saved with the 2 week cure time, especially since I build PCs frequently for others. They won't notice the cure time, but I still use an old tube of MX-3 on my personal rig (similar performance to MX-4 except it's thicker and has a longer cure, though not as long as AS5).
Just my two cents. More like $200 actually. Wow, that was long.
EDIT: actually, MX-4 was tested, but it tested relatively bad, though still in the top few C. This could be put down to a margin of error. All the top 2C pastes are pretty good, just look for cheapest and least cure time.Edited by 267267123 - 12/11/12 at 9:08pm