On Fri, Oct 25, 2019 at 02:11:58AM +0200, deloptes wrote:
Boyan Penkov wrote:
Ulrich Drepper's piece on on-chip memory architectures is a fantastic
read, and I recently had the chance to revisit it --
https://people.freebsd.org/~lstewart/articles/cpumemory.pdf
I am writing to ask more knowledgeable folks if the last 13 years have
seen sufficient changes that render parts of this out of date or
misleading on 2019 hardware.
114 pages! Really :) I am not an expert but let me ask you a question - did
something change in the past 13y regarding memory in context of
programming? I think no. Only "developers" became dumber.
Processors have (again) changed a bit: the gap between processor
speed and memory has widened a tad, there are more cores on a
package (putting even more pressure on the memory bottleneck).
Compilers have become smarter (and more insidious, depending on
your problem at hand) [1] to try to keep that illusion of Moore's
"law" upright.
So yes, an update on Ulrich Drepper's paper would be welcome.
A good read, btw.
Cheers
[1]
https://lwn.net/Articles/799218/-- tomás