I really don't understand at all what you're getting at with heatsinks. Cooling Ivy is easier than cooling Sandy (if we assume identical setups other than the die itself). It really is. Even on air: #1, The flux of heat (heat per area per time, too much mushing technical terms in iffy usage. Like where you talk about FINFets and say more power but it's not, it's more current but at a lower voltage. I'll admit I should have said "heat is power over time" earlier) may be higher. However, Ivy vs. Sandy, total energy dissipation is lower. #2, The die won't reach Tjmax - the chip itself is fine to very high temps. The packaging is what fails. #3, The thermal mass and thermal conductivity of the HSF are many times higher than the die and its packaging material and are identical between Ivy and Sandy. That basically means that your overall heating of surroundings is lower regardless of the increased die temp. The total wattage is still the most important consideration. Conceivably you can scale down to some area and power dissipation and yeah surface area is then too small and the heat flux is too high. But that's not were we're at. Before that, in real CMOS, leakage fubars everything anyway. (Put me in for 14nm as the last/second-to-last of the "traditional" Si-based microelectronics processes.)
Does 3D finFET require less voltage than traditional 2D planar design? I didnt realize that it did. Regardless: Current produces more heat than voltage, so it still works out to be increasing the temps. One example more people are familiar with is standard electrical wiring in your wall. It will be running 115v (give or take) but to run more current you must have exponentially thicker wires to get the power through otherwise the wires will get too hot and melt the insulation and then start a fire. It is the current going through that causes this increase in heat. Onto the processors now, with stock processors (lets take an i5 3570K for this example) if you have two of them and one has a stock voltage of 1.2v and the other has a stock voltage of 1.3v then the lesser voltage CPU usually has higher stock temps than the other higher voltage model. Also the lesser voltage one will drastically increase temps much faster with its voltage increases than the higher stock voltage one will. However, the lesser voltage one will also tend to overclock better if you can keep its higher temperatures in check. This is why when big time overclocker's get a tray or a bunch of some CPU model they bin the processors based on VID and concentrate on the lower VID chips because they know those have a much higher chance of OCing to a higher level. The lesser stock voltage one compared to the same model that has higher voltage has more current running through it. On the CPU binning that way, it isnt ALWAYS the case. More of a good general starting place. Because even though one can have a lower VID than the other it could just have some bad silicon in it that wont let things run as fast, and another CPU that has higher VID could end up having much better silicon and end up running faster. But in general the best starting place is VID because you cant actually look at the silicon to see which is which, the VID is really the only indicator of different properties within the same model of CPU.
I'll point you to the wiki for multigate transistors like Intel's Tri-gate http://en.wikipedia.org/wiki/Multigate_device#Tri-gate_transistors Basically there's two sort of issues. Thinner gate oxide widths mean less threshold voltage. That's how many volts you need to switch the thing. But thinness also means more leakage when it's off. Going to 3D finFETs gives you more surface area for the gate electrodes so you can flow more electrons when you need to but fewer when you don't want to. A CPU with a lower threshold voltage typically means things have gone right for it in manufacturing. You'll have (probably) less leakage and faster switching and less lower consumption.
3D gate may lower the threshold voltage required to switch it and allow more current through, but I am pretty sure you are thinking of leakage to VID ratio backwards. Lower VID of a CPU means it has higher leakage, not lower. As well as the lower VID chips are using more amperage which is why they tend to run hotter at the same speed as higher VID chips. Anyway, the 3D FinFET design is very good don't get me wrong here. It just seems to contribute to higher temperatures than a 2D Planar design.
It's a confluence of factors. A lower threshold with the same transistor geometry would produce more leakage. But the finFET is a thing specifically because it limits leakage. It's a PITA to pattern that and wouldn't have been done if leakage didn't get much lower.
We are going far away from that AMD / NVIDIA discussion We just need graphene chips and we should have much better heat management. It'd be dirt cheap in terms of material as well, just need to be able to harvest it. Now i think for Mantle and all, we'll have to wait for benchmarks early next year with games developed for it (like BF4) to see if AMD cards really destroy Nvidia. Not to mention that AMD needs to convince PC devs to actually use mantle, considering that they would have to develop in OpenGL or DirectX anyway (unless Mantle can be decyphered by Nvidia too, in which case it'd still make ur code much more complex) ? It's an interesting read but i'm not sure that it'll actually be super popular.
Well, the speculation is that Mantle is the low-level API for the XBone (and highly similar to PS4's as well). It's conceivable everything that's cross-platform could put it to use to some extent.
Lol Plant vs Zombies using frostbite 3 ? WTF ?! We'll have to see benchmarks if mantle makes any kind of difference.
It may have some pretty good power for something like that. Ubi's 2D-ish engine is getting a number of titles for it...perhaps EA decided to put in that functionality.
just noticed this thread ... I actually had the honor of being at an AMD training event the day before at Yankee stadium to go over all the new tech. Shit looks pretty legit from first hand experience. Might be droppin this 670 soon !!!
So...fun fact that sort of mutes excitement: Only the R8 290 and 290X are not rebrands. Only those two and the R7 260X have the True Audio. The R9 270 is a 7870ghz. The 280 is a 7890.