Historical GPU FlOPs performance

[Nick]

I would much rather expect Extended Precision (32-bit mantissa, I-don't-remember-how-many-exponent-bits) to become standard rather than FP64, if we are to ever move away from FP32 (which I'm slightly skeptical about, frankly, unless we're thinking more than 5 years down the road?) - this is because it allows significant synergies with 32-bit INT for the execution units, although data paths still need to be wider obviously.

Isn't this how they implement DirectX 10 float and int support already? I can't imagine having separate units for them. Or are integers really stored in floats and thus limited to a 24-bit integer range?

 

[3dilettante]

I would much rather expect Extended Precision (32-bit mantissa, I-don't-remember-how-many-exponent-bits) to become standard rather than FP64, if we are to ever move away from FP32 (which I'm slightly skeptical about, frankly, unless we're thinking more than 5 years down the road?) - this is because it allows significant synergies with 32-bit INT for the execution units, although data paths still need to be wider obviously.

Extended precision would mean non power of two alignment in memory.
Loading a value from memory wouldn't be quite as simple, or data structures would be padded to maintain alignment at the cost of space.

 

[Davros]

Would it be good having FP precised monitors also? (I can't believe we can see just 16M colors at max)

well there's allways analog crt
also the matrox parhelia supported 10bit per colour output gigacolour they called it
and my 9800pro (maybe others) apparently had 10bit dacs but I never found out how to enable 30bit colour or 40bit if you include the alpha channel

 

[OpenGL guy]

well there's allways analog crt
also the matrox parhelia supported 10bit per colour output gigacolour they called it
and my 9800pro (maybe others) apparently had 10bit dacs but I never found out how to enable 30bit colour or 40bit if you include the alpha channel

The mode was A2R10G10B10 and was available to full-screen D3D applications, not sure about OpenGL. Destination alpha is rarely used, so only having 2 bits is not a big deal. Lost Planet is one game supports 2101010, I'm sure there are others (TRAOD did ages ago).