Which GPUs support Double precision floating point natively?

yash_16

Commendable
Jul 15, 2016
3
0
1,510
Which GPUs support native double precision floating point calculations?

This question stems from the "Star citizen's 64 bit conversion of Cry engine and GPU's that can support it" I know it can be that its just CPU conversion and GPU is passed casted floats, but nevertheless.Its not very relevant to the question

The best I can find from google search is all GPUs from nvidia above GTX480 , but that sounds too good to be true.

Also I don't know about AMD and Integrated GPUs , but I expect a GPU integrated with 64 bit CPU to support 64 bit floating point precision
 
Solution
it depends on how the gpu being configured. for gaming gpu DP performance can be limited by design or purposely being limited by gpu maker. since DP are not used in gaming application most often gpu maker purposely limiting DP performance for their gaming card to encourage professional user buying their much more expensive professional line up. take AMD Hawaii/Grenada for example. the fire pro version of the card have it's DP rated at half of SP performance (i think this is the highest configuration that possible on gpu?) but in 290X/390X the DP performance is purposely limited down to 1/8 of SP performance. also yes theorical performance (Tflops) also being cut when using DP. hence that will make it slow for gaming application to use it.
i think you were confuse with certain things like native 64bit application and gpu single precision/double precision operation? AFAIK there is no game using double precision and gaming application itself did not need such precision. also what do you mean with this?:

The best I can find from google search is all GPUs from nvidia above GTX480 , but that sounds too good to be true.
 

yash_16

Commendable
Jul 15, 2016
3
0
1,510


umm yeah sorry about mixing things up in the way how worded the question but I was referring to GPU double precision floating point operations.

following is to give context to my question:
I understand the game "star citizen" uses 64 bit CPU calculations to support a large world that requires more precision for physics after the limit of 32 bit float data type is reached , but is it possible to render the data generated using this 64 bit representation using a single precision GPU without first converting it into a single precision format in camera space or something?

"The best I can find from google search is all GPUs from nvidia above GTX480 , but that sounds too good to be true."

Well if you think about it in the context of star citizen then GPUs with double precision capabilities existing for so long without that mode being exploited is weird. but I recently found out that at double precision almost all GPUs work 24x slower which is the reason why its not used in games to simulate extremely large map size
 
it depends on how the gpu being configured. for gaming gpu DP performance can be limited by design or purposely being limited by gpu maker. since DP are not used in gaming application most often gpu maker purposely limiting DP performance for their gaming card to encourage professional user buying their much more expensive professional line up. take AMD Hawaii/Grenada for example. the fire pro version of the card have it's DP rated at half of SP performance (i think this is the highest configuration that possible on gpu?) but in 290X/390X the DP performance is purposely limited down to 1/8 of SP performance. also yes theorical performance (Tflops) also being cut when using DP. hence that will make it slow for gaming application to use it.
 
Solution