Intel's MPF Presentations

[Conrad Song]

http://www.intel.com/pressroom/kits/...2001/index.htm

[bobgod007]

Thanks Conrad. I personally can't wait till tommorow. Intel and Mircosoft's Q3 conference calls and more of MPF. It should be very interesting. Also when is AMD anouncing the K8 details? I thought it was today or tommorow.

------------------
Count Baron of the OC Crusaders

[Conrad Song]

Originally posted by bobgod007:
Thanks Conrad. I personally can't wait till tommorow. Intel and Mircosoft's Q3 conference calls and more of MPF. It should be very interesting. Also when is AMD anouncing the K8 details? I thought it was today or tommorow.

http://www.amd.com/us-en/Corporate/V...5_3923,00.html

[Moridin]

I notice a lot of stuff on spawning speculative threads. I had always felt that this would not be included in the SMT implementation in the P4. Does this mean I was incorrect or does the information presented here not necessarily apply to the P4?

I was excited about this technology to begin with and that was before I underestimated it once. Now it seems I may have underestimated it a second time.


------------------
Life is complex. It has real and imaginary parts.

[Conrad Song]

Originally posted by Moridin:
I notice a lot of stuff on spawning speculative threads. I had always felt that this would not be included in the SMT implementation in the P4. Does this mean I was incorrect or does the information presented here not necessarily apply to the P4?

I was excited about this technology to begin with and that was before I underestimated it once. Now it seems I may have underestimated it a second time.

From what I understand. The SMT implementation on the Pentium 4 is "true" SMT. Apparantly, there are some restrictions, but the execution pipe is multithreaded.

In this sense, there is no implementation difference between supporting a truly coded multithreaded application, or a coded single threaded application that is recompiled for dynamic threading under special conditions.

Oh, I see your point...

Obviously there is overhead in generating even a lightweight thread, but it is also possible to generate an execution context without support by the operating system. Now I'm speculating...

Actually, the presentation talks about using spawned threads to spawn more threads... hiding the overhead!

[This message has been edited by Conrad Song (edited October 15, 2001).]

[This message has been edited by Conrad Song (edited October 15, 2001).]

[Moridin]

I guess my questions what is it doing with these threads it is spawning. I am wondering if the P4 does indeed have some form of vertical multithreading. Basically this would give it the ability to eliminate branch prediction penalty by spawning threads that executed down both paths of a branch.

This isn’t clear for the power point slides though. It also isn’t clear if this was P4 specific information. It may only be theory, and not something that exists in the P4 as it is now.


------------------
Life is complex. It has real and imaginary parts.