[3dem] [ccpem] advice hardware for cryo-EM data processing

[Marin van Heel]

Dear Dave

We have already converted all our cryo-EM software to 64 bits FP more than
a decade ago.
The reason is simple: if you have a dataset consisting of say 1.000 images
of 512x512 pixels and you would like to calculate the average density of
the pixels in that data set that requires 9 significant digits. In 32 bits
floating point only 6-7 significant digits are available...
Another example: The 3D FFT of an 800x800x800(=512.000.000) voxel 3D
reconstruction will again require more than 9 significant digits. The
tricky thing is that the floating-point underflows are rounded off to "0"
and the users don't necessarily notice that the results are flawed.
More recently we have also implemented the double-precision FP also as an
official IMAGIC disk format, although all FP calculations were already in
64bit FP for more than a decade.
Cheers
Marin


Van Heel, M., Portugal, R., Rohou, A., Linnemayr, C., Bebeacua, C.,
Schmidt, R., Grant, T. & Schatz, M. (2012). Four-dimensional cryo-electron
microscopy at quasi-atomic resolution: *IMAGIC 4D*. *International Tables
for Crystallography*. *Vol. F, Chapter** 19.9,* 624–628.




On Tue, Dec 5, 2023 at 12:04 PM David Lawson (JIC) <david.lawson at jic.ac.uk>
wrote:

> Hi All,
>
>
>
> This a request from a colleague who has been tasked with sourcing hardware
> for cryo-EM data processing. He is trying to work out a compromise between
> NVlink for memory sharing and 32/64bit Floating Point operations. Is 64bit
> FP needed or likely to be important in the next 5 years or so?
>
>
>
> Any advice would be much appreciated.
>
>
>
> Many thanks
>
>
>
> Dave
>
>
>
> -------------------------------
>
>
>
> Prof. David M. Lawson
>
> Department of Biochemistry and Metabolism,
>
> John Innes Centre,
>
> Norwich,
>
> NR4 7UH, UK.
>
> Tel: +44-(0)1603-450725
>
> Web: https://urldefense.com/v3/__https://www.jic.ac.uk/people/david-lawson__;!!Mih3wA!HIEoGvRHTL6Rcxv13u3aXIldgdmnsEQE3XgVp-YZoFIoTtsnHQu9kjsYnNL2uBRmwCck8TMTPknkTM0LO2pgJb5cGF2SJw$ 
>
> Email: david.lawson at jic.ac.uk
>
>
>
>
>
> ------------------------------
>
> To unsubscribe from the CCPEM list, click the following link:
> https://urldefense.com/v3/__https://www.jiscmail.ac.uk/cgi-bin/WA-JISC.exe?SUBED1=CCPEM&A=1__;!!Mih3wA!HIEoGvRHTL6Rcxv13u3aXIldgdmnsEQE3XgVp-YZoFIoTtsnHQu9kjsYnNL2uBRmwCck8TMTPknkTM0LO2pgJb7L2YiQww$ 
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.ncmir.ucsd.edu/pipermail/3dem/attachments/20231205/9018c8b5/attachment.html>