Page 1 of 1
Asymmetric multi-threading
Posted: Mon Apr 19, 2004 11:29 pm
by quantus
http://www.eet.com/semi/news/showArticl ... 97&kc=2515
Intel recently demonstrated a technique it calls asymmetric multithreading on a prototype Itanium processor that uses as many as 24 short "helper threads" generated by a compiler to pre-fetch and speculatively execute data, improving a CPU's single-thread performance. The technique will appear in Xeon and Itanium CPUs "very soon," said Justin Rattner, director of Intel's microprocessor research lab.
Jonathan, would you happen to have pointers to papers describing the performance benefits from this, or should I just try looking on my own?
Posted: Tue Apr 20, 2004 12:45 am
by Jonathan
I have no links. But I am intrigued.
http://www.intel.com/research/mrl/news/ ... ecture.pdf
shitty pay link deleted
This is what Google has to say.
Shen!!!
Posted: Tue Apr 20, 2004 2:52 am
by quantus
damn these journals which want you to pay
Posted: Tue Apr 20, 2004 3:16 am
by quantus
What's with the random slide at the end of the presentation showing Intel processors with Netburst architecture? Those are the ones that implement dynamic threading I suppose?
Posted: Tue Apr 20, 2004 3:19 pm
by Jonathan
Read the link you posted. It's really a compiler feature which can be exploited by SMT or CMP systems (anything with shared resources). They happen to have implemented it in a research Itanium compiler they've been developing. The way I read it, there's no hardware support at all.
Posted: Tue Apr 20, 2004 9:07 pm
by quantus
I don't think it's quite as simple as just writing a thread that runs in parallel. For example, one of the requirements is "low-cost thread spawning". Wouldn't there be certain things that aren't necessary for these special threads?
Posted: Tue Apr 20, 2004 9:13 pm
by Jonathan
Then that sounds like an OS issue to me.
The Sun architect is quoted in your article as saying Intel's approach is software only.
Posted: Tue Apr 20, 2004 9:18 pm
by Jonathan
http://www.intel.com/research/mrl/resea ... pilers.htm
Here's the web page of the compiler team responsible for this. They don't seem to have any of their publications online.
Posted: Tue Apr 20, 2004 10:45 pm
by quantus
Ok, I'm back to my original question of "what's with the random slide at the end of the presentation?"