Sign in with
Sign up | Sign in
Your question
Solved

The smaller the die size, the less heat production?

Last response: in Graphics & Displays
Share
July 21, 2010 11:38:56 PM

This is a noob question about GPU's. I heard that the smaller the die size, the less heat production and energy consumption. Performance and efficiency is also slightly improved. The 40nm ATI Radeon HD 5xxx series will be refreshed to a 32nm "Southern Island" process. Will this lessen the heat output and energy consumption slightly? Allowing for higher clock speeds and improved performance?

Does this same logic apply to CPU's and other micro-chips?
a b U Graphics card
July 22, 2010 4:16:05 AM

Yes - that's why the hex core i7 980X has the same TDP as a Core i7 930, which is a quad core, as it is based on the 32nm process instead of 40nm. Higher clock speeds? Probably.
m
0
l
a c 194 U Graphics card
July 22, 2010 4:22:26 AM

Shrinking the process like from 40nm to 32nm does reduce power as it allows the voltage across the transistors to be reduced slightly, it also increases speed as the channel is now shorter so you can have faster switching transistors or more complex architectures running at the same speed. Die size usually refers to the size of the GPU or CPU itself rather than the size of the transistor so die size does not affect the power usage directly, its just a result of having more or less transistors.
m
0
l
Related resources
Can't find your answer ? Ask !
July 22, 2010 4:56:22 AM

hunter315 said:
Shrinking the process like from 40nm to 32nm does reduce power as it allows the voltage across the transistors to be reduced slightly, it also increases speed as the channel is now shorter so you can have faster switching transistors or more complex architectures running at the same speed. Die size usually refers to the size of the GPU or CPU itself rather than the size of the transistor so die size does not affect the power usage directly, its just a result of having more or less transistors.


Does it reduce heat output slightly?

I would imagine less TDP would result in lower temperatures.
m
0
l

Best solution

a c 171 U Graphics card
July 22, 2010 5:19:03 AM

Quote:
Shrinking the process like from 40nm to 32nm does reduce power as it allows the voltage across the transistors to be reduced slightly,


As hunter said, yes. The more voltage you have to apply to the transistor, the more heat it will emit. 1Billion transistor with a voltage of 1.5v will emit more heat then 1B transistors with a voltage of 1.25. (assuming the same type of transistor and manufacturing process as well of course. You can't compare Intels HKMG tech straight across with AMD/IBMs SOI) As the voltage goes down, so does the heat output of the chip. As the gates get smaller, the faster they can switch on and off as well.
Share
a c 194 U Graphics card
July 22, 2010 5:26:07 AM

Power usage is directly proportional to heat output which is proportional to the temperature it runs at with a certain cooler. Since your CPU has no way to do mechanical work all the power it draws from the power supply gets turned into heat which is why you sometimes see coolers rated in watts like the Thermaltake Frio says it supports 220W, but since they often ship lower power parts with weaker coolers the temperatures end up being very similar between chips with different TDPs.
m
0
l
July 22, 2010 5:40:07 AM

Best answer selected by ambam.
m
0
l
July 22, 2010 5:41:28 AM

The GTX 485 was a 55nm GTX 480 core, which was originally a 65nm. It used almost 40W less power, and ran slightly cooler and more efficiently.

When is the 32nm refresh of the "South Island" HD 5xxx series going to be done? They should put out even less heat than they already do. The HD 5970's can exceed 90 *C under heavy loads in cases with poor airflow, mainly because of the high energy demand.
m
0
l
a c 171 U Graphics card
July 22, 2010 6:39:03 AM

GTX485? I'm assuming you mean GTX285...

SI isn't supposed to be done till the end of this year. Hopefully in time for Christmas, but anything can happen. Its not supposed to be a big leap from the 5xxx series, nor is it supposed to be on 32nm. It should be a 40nm refresh of the 5xxx series. Rumor says you won't see the next set of chips from AMD until GF can get their sub 40nm process up. No telling when that will be. I just read something the other day that GF wants IBM to change the transistors in their HKMG w/ SOI so that they can get it working.
m
0
l
!