To what degree does the transistor count in a chip affect a processor's performance and increased power consumption/heat dissi

cveticvetkov

Honorable
Dec 5, 2013
2
0
10,510
Hello, everyone

I am a University student, and I was inspired to do my own research on assessing to what degree does the transistor count in a chip affect a processor's performance and increased power consumption/heat dissipation. Any model would do.

But I am finding that this topic is really challenging and I would like to turn to someone as knowledgeable as you guys, for advice. Would it be possible for anyone to link me to any experimental/evaluational/research data that is relevant to either processor's performance or increased power consumption/heat dissipation, or provide me with any guidelines to follow? Basically any data with experimental results that I could use to support my claims?

Thank you,
Kindest regards,

Cveti
 

swilczak

Distinguished
I don't know how technical you want your answer but it's really a pretty basic concept the way I understand it. The more transistors that are put in a small area reduces the amount of space between each transistor reducing the resistance producing less heat and allowing the processor to calculate more instructions per cycle and it allows engineers to put more cores in a smaller space. So each time a new chip is made with more transistors it can get more work done using less power and it's also more reliable when it produces less heat.
 

cveticvetkov

Honorable
Dec 5, 2013
2
0
10,510
Dear swilczak, thank you for your response. Quite technical, because the biggest challenge for me is finding any piece of data/paper/statistics that includes experimental/comparison results that show the relationship betweeen the different variables (transistor count in the chip, performance increase/no change in performance, clock speed, effect on heat dissipation, basically any variable that might be relevant) and the effects. Anything in that regard that I can use as proof. I have been educating myself on hardware CPU information for the past month, but at this point I am pretty stuck. My professor warned me that its extremely hard, but I am determined to prove him wrong! :)
 
Dont know about performance, but I do know that power consumption has more to do with the CPU architecture, physical die and transistor size and clock speeds than it does with transistor count. The more efficient the architecture, smaller the process node, smaller the die and lower the clock speed results in lower power consumption.

Heat Dissipation definitely has nothing to do with transistor account, that comes entirely down to the thermal transfer between die to heatspreader to cooling device (the power of which plays a big part). Going a million transistors either direction wont change your load temps, but spending $30 on a heatsink better than stock definitely will.

Your probably best off looking at the transistor count of the past 10yrs of equivalent CPU's (Intel Extreme Editions would be a good option) and compare that to their performance. Make a line graph, figure out if it has any correlation at all and determine the strength of it.
 

COLGeek

Cybernaut
Moderator

chrisso

Honorable
Nov 17, 2013
1,333
0
11,660
If you want a good comparison (this is off the top of my head)
look at the performance indexes and comparissons between an intel core 2 duo E6300
(conroe 65nanometer) and an E6600(same) The transistor counts, clock speeds, cache ram, etc etc are all listed on intel arc. I will post in a minute.
 

chrisso

Honorable
Nov 17, 2013
1,333
0
11,660
http://ark.intel.com/products/27248/
http://ark.intel.com/Products/Spec/SL9s8

You will notice the higher numbered chip has nearly twice the transistors. 2 million of these i expect are in the extra cash ram. It also runs at a higher frequency.
It doesnt have twice the performance of the smaller chip though. The cpu passmark for the E6300 is
1111,
and the E6600's cpu passmark is
1500.
so about 40 %. It seems extra transistors is an expensive way of increasing performance. Architecture is everything. Take a look for yourself.
 

swilczak

Distinguished

If transistor count didn't matter we would still be using 90nm chips or something much larger. Transistor count, architecture and how the software is designed to work with that particular CPU are all major factors in performance.
 

chrisso

Honorable
Nov 17, 2013
1,333
0
11,660
You will notice transistor counts go up from a certain point , to a certain maximum point in each generation of cpu's.
Smaller transistors and better architecture design are the main changes, and result in higher performance
and cheaper sku's. The maximum number of transistors is dictated as much by profit margins as by design
limits. The number used will be just enough to better the last generation of cpu's and enough to sell new chips in the market place.