Someone make this into a sticky or FAQ, please.
A simple look at the background of product binning: http://en.wikipedia.org/wiki/Product_binning
Contrary to popular belief, there's no "cherry picking" for GPUs. Not every single piece of silicone will be identical to the next, no matter how controlled or precise the manufacturing process is.
So, how do top tier chips come about? Product binning.
Let's say 1000 GPU chips come in. Reference clock speeds are 500MHz, and the approx. overclocking headroom is around 30% (or as advertised by the GPU maker).
After some internal calculation, it's determined that the GPUs can be overclocked to around 650MHz and meet the 3 year failure rate within acceptable limits to not exceed RMA limits.
Second tier is set to 600MHz, third tier is 550MHz, fourth is reference speeds.
So the batch of GPUs is tested.
Tier 1: Meets 650MHz testing, reserved for top level boards
Tier 2: Meets 600MHz testing, reserved for 2nd level boards
Tier 3: Meets 550MHz testing, reserved for 3rd level boards
Tier 4: Meets 500MHz testing, reserved for reference clock speed boards
Anything that fails the lowest testing is usually returned to manufacturer.
Now, I'm overly simplifying the process of binning, but this is how it usually works. The more important question is, what about overclocking???
In a nutshell, chips are binned BEFORE they are installed on a board. So, the process will not guarantee your GPU to overclock to a certain speed. NOTHING GUARANTEES OVERCLOCKING OF ANY SORT. The only thing guaranteed, is what default speeds the card is clocked to.
Beyond that however, the chances of you getting a higher overclock with a tier 1 video card is better than the chances of you getting a higher overclock with a lower tier video card. Thus, this leads to the misconception that flagship products are "cherry picked".
Remember: when it comes to overclocking, you're playing the silicone lottery. Your chances of winning are higher with a top tier product, as opposed to a lower chance of winning with a lower tier product.
Every manufacturer's selection process varies slightly, but it's mostly the same concept. There is simply not enough profit margin in this industry to go and cherry pick GPUs for enthusiasts.
Edit: How are GPUs tested before installation? Simple: reference PCB test bench.
GPUs, much like CPUs, are delivered to the AIB partner in tray(s) of usually 1K. The GPUs are then put on a test PCB that is a copy of the reference PCB design, except with test points.
Then voltages and clock speeds can be increased to see how viable the chip is for certain specs, and "binned" accordingly.
Q.E.D.
A simple look at the background of product binning: http://en.wikipedia.org/wiki/Product_binning
Contrary to popular belief, there's no "cherry picking" for GPUs. Not every single piece of silicone will be identical to the next, no matter how controlled or precise the manufacturing process is.
So, how do top tier chips come about? Product binning.
Let's say 1000 GPU chips come in. Reference clock speeds are 500MHz, and the approx. overclocking headroom is around 30% (or as advertised by the GPU maker).
After some internal calculation, it's determined that the GPUs can be overclocked to around 650MHz and meet the 3 year failure rate within acceptable limits to not exceed RMA limits.
Second tier is set to 600MHz, third tier is 550MHz, fourth is reference speeds.
So the batch of GPUs is tested.
Tier 1: Meets 650MHz testing, reserved for top level boards
Tier 2: Meets 600MHz testing, reserved for 2nd level boards
Tier 3: Meets 550MHz testing, reserved for 3rd level boards
Tier 4: Meets 500MHz testing, reserved for reference clock speed boards
Anything that fails the lowest testing is usually returned to manufacturer.
Now, I'm overly simplifying the process of binning, but this is how it usually works. The more important question is, what about overclocking???
In a nutshell, chips are binned BEFORE they are installed on a board. So, the process will not guarantee your GPU to overclock to a certain speed. NOTHING GUARANTEES OVERCLOCKING OF ANY SORT. The only thing guaranteed, is what default speeds the card is clocked to.
Beyond that however, the chances of you getting a higher overclock with a tier 1 video card is better than the chances of you getting a higher overclock with a lower tier video card. Thus, this leads to the misconception that flagship products are "cherry picked".
Remember: when it comes to overclocking, you're playing the silicone lottery. Your chances of winning are higher with a top tier product, as opposed to a lower chance of winning with a lower tier product.
Every manufacturer's selection process varies slightly, but it's mostly the same concept. There is simply not enough profit margin in this industry to go and cherry pick GPUs for enthusiasts.
Edit: How are GPUs tested before installation? Simple: reference PCB test bench.
GPUs, much like CPUs, are delivered to the AIB partner in tray(s) of usually 1K. The GPUs are then put on a test PCB that is a copy of the reference PCB design, except with test points.
Then voltages and clock speeds can be increased to see how viable the chip is for certain specs, and "binned" accordingly.
Q.E.D.
