I'm interested in learning more about GPUs besides performance. Thanks!
What is die-size?
What is a wafer?
How are the above two correlated?
What does electricity have to do with die-size (PSu requirements)?
What does 'yield' mean?
What does it mean to bin chips?
Why do some chips OC better than others?
How does memory run at 3+ghz effective? What does effective mean?
What kind of stepping is an A-1 to an A-2 stepping?
What kind of stepping is an A-1 to a B-1 stepping?
In general the smaller the manufacturing size (90nm, 65nm, 40nm, 32nm NanoMeter) the less power that that chip needs.
Yield is the amount of chips that are successfully made and work compared to the total chips made. (50% yield means that 25 out of 50 chips work)
To bin chips... Ive heard of the term but I just cant remember what it meant right now. Ill be back.
Typicaly later versions of chips come out that have been redesigned to allow for better overclocking. (Q6600 B3 was one of the first while the later G0 came out and had much better overclockability.)
I think "effective" means that you added together the speed of each RAM stick together. (I could be wrong) so DDR2 running at 800mhz will have a effective rate of 1600 mhz, DDR3 @ 1000mhz will be effective 3000 mhz.
And the difference between each of the steppings, I have no Idea.
Binning is the practice of putting chips that don't make the grade to one side to be released as a different, usually lower end model of a GPU.
Take the 5870 and 5850. The 5850 is the same chip but it didn't pass the tests to make 5870. This could be anything from a stability issue to a fault in the silicon, but its fine when run at lower clocks or with the faulty silicon disabled. That's why there are less SP's in a 5850.
Some times chips can be binned when they are of exceptional quality and so are put aside for overclocked versions or for special editions but i believe this practice is quite rare.
Also some times if a process is very good then you get chips that have no defects being used as lower end chips with bits disabled. This is where the products that have unlock able cores (CPU's) come from.
DDR stands for Double data rate. This means that if you have a displayed speed of 800 in CCC in ATI's case your effective speed is 800 doubled because its DDR which gives you 1600 effective.
GDDR5 is basically Quad pumped so you just X the base clock that would be displayed in your CCC or Nvidia equivalent to get the effective rate. Its a little more complicated than just being quad pumped as there are latencies involved but its more or less right.
You tend to get the effective, bigger number on the packaging which is why we get questions on the forum saying "my memory is running way slower than the advertised speed"
I'm not sure about steppings as far as do the numbers mean any thing, all i know is each stepping is basically an improvement on the original design, usually means better efficiency which then gives higher OC potential or just using less power, some times as with the D0 stepping of the i7 920 you get a decent increase in performance as well.
A1- A2-B1 etc would be considered as Silicon revesion numbers as far as i am aware. I dont know exactly what they mean but i think each time you have to go back and make a batch of chips due to defects etc that make the chips non viable you end up with a higher number which is bad as it means your chips/process didnt work as you thought and you had to try again. I thing Fermi is on B1 but thats just what i read on the net. If anyone could clarify this point i would appreciate it.