28 Dec
2017
28 Dec
'17
8:12 p.m.
Will Godfrey wrote:
If I've understood that correctly you can also
ensure that they are
also on the same socket, which apparently improves memory access.
I think this is what is meant by NUMA (Non-Uniform Memory access).
These days, the time it takes for a core to access memory, even on the
same chip, can depend on where the core and memory are on the chip.
So if you want threads to share memory as fast as possible you need to
be able to pin them down.
The downside is that running many CPU-hungry threads close together on
the chip can cause overheating, causing the CPU to throttle everything
back until things cool off. The CPU architecture giveth, and the CPU
architecture taketh away.
Regards,
Jeremy Henty