Does AMD/Nvidia bin their GPUs?

[karma831]

Just wondering if AMD/Nvidia or their partners bin their GPUs. And how exactly does the process work? I know that CPU manufacturers bin their processors with tester machines (ex. http://www.teradyne.com/std/ ). Do they do the same with GPUs?

Thanks

 

[eyefinity]

It's a good question.

The obvious answer would be yes, they do bin gpu's. That's why the same chip can end up with 3 different SKU's (stock keeping units).

A 5870 and 5850 are both of the same chip (Cypress), and The gtx 480 and 470 are also the same (Fermi).

Sometimes chips are bad but can be salvaged through using higher voltages and cutting off bad silicon. Higher volts = more stable performance when the silicon isn't good. These are chips like the 5830 (Cypress) and gtx 465 (Fermi) - physically identical to their more illustrious parents except cut down into slower working parts.

 

[eyefinity]

As a physical test, I believe they use certain kinds of dye which shows how good or bad chips are. That is probably what the "binning" name came from. A chip gets "binned" into a certain bucket depending on how it performs on the dye (voltage) test.

 

[karma831]

How about GPUs of the same SKU? Companies often have many different models of the same GPU. Do they use the higher binned for the OC editions and lower binned for the standard?

 

[eyefinity]

Chip binning will be done at AMD/nVidia I think (splitting a cypress into a 5850 or 5870), although further binning can then be done at their partners, eg Sapphire or EVGA, and that will theoretically allow them to keep better parts for more expensive cards.

 

[mactronix]

How about GPUs of the same SKU? Companies often have many different models of the same GPU. Do they use the higher binned for the OC editions and lower binned for the standard?

EVGA do exactly this for their FTW 460 cards.

Quote.

"Some go for advanced cooling, some change some of the extraneous components, while some even go for completely redesigning the PCB the chips sit on.

Here though EVGA has done none of that. Instead it has tested and retested it's stock to find the fastest, most stable versions of the chip and overclocked them to within an inch of their lives."


Source.
http://www.techradar.com/reviews/pc-mac/pc-components/graphics-cards/evga-geforce-gtx-460-1gb-ftw-edition-900025/review

Some companies will do this and some wont. Some companies use better than stock components besides the GPU and some dont, things like memory and power regulation circuits.

Mactronix

 

[dertechie]

The 460 is particularly well suited to that. NVidia clocked it absurdly low to not eat into their 465/470 sales, so some of those GPUs can clock very high without effort. It's equivalent to what would happen if AMD decided to make all Barts GPUs into 6850s and didn't make a 6870. You would see OC 6850s clocked to the sky since all the chips that would be 6870 would be in the pile. You're already going to see people OC 6850s close to 6870 speeds because the 'almost 6870' chips are all 6850s.

 

[mactronix]

Yes you are correct in what you are saying, its not a new thing though, i have seen this kind of thing happening since the 8800 series, the first ones the 320 and 640 cards.

Mactronix