Article: Parallelism at HotPar 2010
By: Richard Cownie (tich.delete@this.pobox.com), August 6, 2010 4:05 pm
Room: Moderated Discussions
Mark Roulo (nothanks@xxx.com) on 8/6/10 wrote:
---------------------------
>What is interesting is to watch AMD with the upcoming Bulldozer core(s). This
>*really* looks like a chip with a bias in favor of throughput computing.
It doesn't look that way to me: if you run a single
thread on each module, it seems as though you get a
pretty decent 4-issue OoO integer unit plus a really
good FPU, and a 1MB L2 cache. That's not obviously
slow for running single threads, though it's
going to depend on the details of clock speeds and
latencies and all kinds of other stuff. But if it
clocks high and they haven't screwed up, I don't see
why it couldn't match SandyBridge for single-thread
performance. We'll have to see.
It should be really good at throughput as well, but
I don't see anything that clearly compromises
single-thread performance.
---------------------------
>What is interesting is to watch AMD with the upcoming Bulldozer core(s). This
>*really* looks like a chip with a bias in favor of throughput computing.
It doesn't look that way to me: if you run a single
thread on each module, it seems as though you get a
pretty decent 4-issue OoO integer unit plus a really
good FPU, and a 1MB L2 cache. That's not obviously
slow for running single threads, though it's
going to depend on the details of clock speeds and
latencies and all kinds of other stuff. But if it
clocks high and they haven't screwed up, I don't see
why it couldn't match SandyBridge for single-thread
performance. We'll have to see.
It should be really good at throughput as well, but
I don't see anything that clearly compromises
single-thread performance.
