Originally Posted by Jpmboy
yeah - I just put the second ("better" ) card in the loop - running two uniblocks, it's "loopy". What I've seen so far is that ya can;t predict the Freq_max from what the card maintains while being downclocked from the TL (well, and PL etc). They all work together and the weighting of each is hard to figure out when I can only remove the thermal effect. One thing for sure, the clock holds much more constant. I had the weak card at 9C (max t of 18C under load) and it will hold 2101 in timespy, heaven and valley .without dropping a clock bin. Just on the rads, the cards never > 35C.
I posted the IR gun scan earlier... the hottest things on the board are the 2 R22 mofsets thingys up by the PCIE power connectors.
The two R22 "thingys" are chokes (inductors). That is the memory VRM up there. Not surprising that they run so hot being that we are overclocking into the 11GHz range on all those dense memory chips.
You can easily solve your power limit problem, especially since your cards are already naked like that. Just put some CoolLabratory Liquid Ultra over the 5MO resistors near the power connectors on the cards. Cover up the whole resistor to connect the two sides of the resistor together. That should about double your power limit on the GPUs.
Originally Posted by Jpmboy
the IR gun temps were measured as shown. The 2 R22s I pointed out will hit very high temps (70C+) but that is still below the spec T_max for that part. Regardless, the performance/stability declines with higher temps.
The mosfets (typically the limiting factor of the GPU VRM)
in the VRMs of this card have basically no de-rating with higher temps. Same rating at 25c as 100c. Which is odd for a mosfet, but the reason it is like that is because the package is rated extremely low and is the limiting factor. They do become a little bit less power efficient at higher temps, but they dont lose performance or anything like that on this model. The good news is that the memory VRM current capability is way overbuilt for the memory, can easily put 150 watts through the mosfets (though, 150w would probably over saturate the memory chokes)
. Just buckets more power than you would ever use. it is surprising Nvidia would build up the memory VRM so much considering how badly they cut costs and make a crap core VRM on all their cards, and because GDDR5X is supposed to use even less power than normal GDDR5. IDK. Maybe the new X RAM requires a more intricate power delivery to achieve its lower draw and higher bandwidth.
I am curious though, how can you claim the parts are not near their T_Max when you dont even know what the parts are called that you are referring to? Seems like you would not be able to look up the spec sheet on something you dont know what it is.
Originally Posted by Dr Mad
I wish Watercool make the new Titan waterblock available soon as it has the better looking in my opinion but it's not announced yet.
Watercool told me yesterday that the Titan Heatkiller block is being designed. My card performs well enough as is for the time being so I am just going to leave it and wait for the Heatkiller block