So, I thought I would report in. I've been doing some messing around with my 7870, with a tinge of overclocking and have found some disappointing results. I don't know if it specifically relates to Gigabytes card or 7870s in general (I have a GV-R787OC-2GD), but it seems to suffer from a GPU throttling cap. Specifically when the GPU hits 70c it throttles it down 450mhz for about 2 seconds, after which it goes back to the normal clocks and the temperature continues to climb back up.
This is at +20% power.
This is only part of it though. I noticed my VRM temperatures get really hot and decided to pop off my heatsink out of curiosity and to replace the thermal paste. After removing it, I can safely say the card has absolutely no VRM cooling. Nothing makes contact with the VRMs, which would give credit to TRs recent 7870 overclocking results. It appears the after market cooler has no VRM cooling and from the design of the card, it would be extremely hard for the heatsink to actually make contact with the VRMs (they're buried lower and next to capacitors which have about double their height).
After replacing the thermal paste, I can keep the GPU temperature relatively stead under 100% fan load, but the VRMs quickly rise to +110c at which point I don't feel comfortable continuing the test to see how hot they can get before the card locks up. It appears as though there is no thermal throttling in place to protect the VRMs from heat damage. Keep in mind the VRMs never reach these temperatures normally, only when running OCCT which seems to do the best at making the card suck more juice.
This sort of puts a damper on the overclockability of the card. I have to keep it under 70c and I can't draw more then 14.5~ amps through it on VDDC in without the VRMs overheating regardless of how cool the GPU. I'm not entirely sure why there is 70c throttle on the card either.
I've done a couple things to get around this as the VRM temps prevent me from completely stress testing the card. It appears as though the GPU is fine with extra voltage and higher clocks, it just makes the VRMs overheat exponentially faster (this also coincides with why the cards crashed so easily when TR overclocked them). Quite literally I bumped the voltage by .04 and it lasted about a minute at 100% fan speed before the VRMs once again became too hot even though it seemed perfectly stable.
Now if I turn down my power limit to like 10% it allows the card to stay cooler, but this also limits my FPS in OCCT to less then what the GPU overclock gives me. So it's really a net loss even though the GPU is running faster. Really it seems like 14.6a is the point at which the VRMs no longer stay cool enough. Getting there through any combination of GPU speed, power limit setting, or core voltage.
Ideally I could run my card at +20% and higher clocks in games as they don't fully utilize the whole card (GW2 for instance only draws +6a on my VDDC in at my normal clocks under 100% load),but I still have no way of stress testing my card without the VRMs overheating to see what is actually stable.
Point of all of this. Third party coolers, at least for this specific Gigabyte model and the ones TR tested blow donkey nuts and don't cool VRMs properly. The 7870 may have a imposed thermal limit of 70c either in drivers or in hardware, I'm not sure which. There are only a handful of posts on the internet that reference this limitation, bit it very much exists. One post I read stated that reverting to 12.6 drivers fixes the 70c thermal cap. I'm not sure if this is or isn't intentional by AMD, there aren't enough people noticing it to generate buzz.
My guess is the 7870 is too good at OCing for its own good and it's nipping at the higher end cards. For instance my card is clocked at 1100Mhz, but easily hit 1150 without any voltage tweaking and hit 1200 with that small bump in voltage, but resulted in VRM overheating (I didn't try going higher then that because it wasn't worth it due to the VRMs). Oh and the memory hit 1450mhz (limitation of CCC) without throwing any sort of fuss. That I could stress test as for some reason it added absolutely no heat to the card... or at least very little compared to the GPU. The memory did have thermal pads on them that connected them to the heatsink.
A little more on the heatsink testing. I originally had a 6970 this card replaced with a traditional blower. From messing around with the fan speed settings on this card, I can safely say subjectively that it isn't all that great at dissipating heat. The heatsink may be good, but it has no where to blow the heat so it quickly becomes saturated (the heat is also blown into the card and doesn't have good direction). The difference in heat dissipation at 20%, 50%, and 100% is marginal even though the CFM drastically increases between those settings. Testing my 6970 on the other hand produced extremely good and also loud results when changing the fan speed settings manually, even keeping the GPU bottomed out at around 40c when testing the overclocking capabilities of that card. They both used the same thermal paste.
I understand that it's two different cards, but there really is something wrong when increasing fan speed doesn't have a noticeably direct impact on heat dissipation. I want to say case blowers like these truly ARE inferior (this has nothing to do with VRMs). Even if you had a box fan hooked up to the side of the case they still aren't good at putting the CFM through the fins efficiently. It may be a giant block of metal, it may have more heatpipes, and it may have more spinners, but it's overall quite a bit less efficient, not to mention the heat it dumps in your case.
Buy a card that blows outside, not in the case. I would've if I had a choice.
Some final notes. This card has a thin plastic shroud that it uses for the fans and they have thin blades. It produces a noticeable whine/rattle and appears as though the fans are of very low quality.http://www.newegg.com/Product/Product.a ... 6814125418