Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Single Precision vs Double Precision GPUs? #96

Open
spimager opened this issue May 14, 2016 · 3 comments
Open

Single Precision vs Double Precision GPUs? #96

spimager opened this issue May 14, 2016 · 3 comments

Comments

@spimager
Copy link

We're looking to buy several GPUs for our cluster and were wondering if double precision is necessary for large datasets or if we can get away with single precision?

@psteinb
Copy link

psteinb commented May 14, 2016

Hi spimager,
We are facing similar questions on a daily basis. And the answer is typically: it depends. If you are somewhat certain that your payload is only SP, gaming cards offer the best price per performance within certain limits. The latter being the amount of hours used and the memory accuracy. For example, if you have users that run SPIM_Registration once or twice per day or less, then gaming cards maybe good. After all, if one card fails, replacing them is financially feasible. If you wanna be sure to cover other apps that require DP, there is no way around HPC cards. Rest assured that many other institutions wonder about the same question. If you wanna, discuss more details in terms of what servers to choose and what cards offer a good price per value etc, feel free to ring me up starting Tuesday next week.
Best,
P

@StephanPreibisch
Copy link
Member

Hi, specifically the SPIM Reconstruction uses only single-precision for memory efficiency (CPU & GPU versions). The precision is not that crucial in this case.

@spimager
Copy link
Author

Great! Thanks, guys!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants