[Beowulf] Variable precision arithmetic

[John Hearns via Beowulf]

I did not want to hijack the thread on Nvidia cards.

Doug Eadline as usual makes a very relevant point:
"BTW, I find it interesting one of the most popular codes run
on Nvidia GPUs is Amber (MD). It has been optimized to use
SP when it can and many Amber users turn off ECC because it
slows down the GPU which translates to runs really well on
Nvidia video cards."

I would like to start off a discussion on choosing the appropriate
precision for calculations.
I believe that there is a generally agreed limit for an exascale machine of
30MW of power consumption (I could be wrong). So people are looking at the
energy consumption per computational operation.
I believe computational scientists will have to start paying a lot more
attention to the choice of data types, rather than declaring double
precision.

Just my two pennies worth.

[Nathan Moore]

Is GMP vectorized for CUDA?

https://gmplib.org/

Post by John Hearns via Beowulf
I did not want to hijack the thread on Nvidia cards.
"BTW, I find it interesting one of the most popular codes run
on Nvidia GPUs is Amber (MD). It has been optimized to use
SP when it can and many Amber users turn off ECC because it
slows down the GPU which translates to runs really well on
Nvidia video cards."
I would like to start off a discussion on choosing the appropriate
precision for calculations.
I believe that there is a generally agreed limit for an exascale machine
of 30MW of power consumption (I could be wrong). So people are looking at
the energy consumption per computational operation.
I believe computational scientists will have to start paying a lot more
attention to the choice of data types, rather than declaring double
precision.
Just my two pennies worth.
_______________________________________________
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf