3 ms·
>The ratio can now be as much as 1:100, so a supercomputer is forced to spend the vast majority of its time moving data to and from memory. This is spurring th
by cmiller1 11y ago
>The ratio can now be as much as 1:100, so a supercomputer is forced to spend the vast majority of its time moving data to and from memory. This is spurring the adoption of data-tiering hardware and software that try to maximize CPU effectiveness by storing frequently accessed data in high-speed tiers such as solid-state drive
Err... this isn't new, wouldn't the high speed tiers be processor registers and cache and we've been doing this for over 30 years?
- kirrent 11y agoAbsolutely it's nothing new, it's just a case of the scale of the problem being larger than ever. IO speed increases haven't kept up with floating operation performance of processors. If you're performing operations on a large, distributed dataset it's easier than ever to have your processor be data starved. Of course, that's why any decent HPC system will include high speed local data stores such as an SSD and plenty of RAM. That's why cache size on these processors is important. That's why systems are more heterogeneous than ever. I feel like some writers in HPC (ironically) miss the free lunch of faster and faster systems with not much extra effort. With more complexity in your supercomputer comes more power and more difficulty in writing software. That's unavoidable.