Community
cancel
Showing results for 
Search instead for 
Did you mean: 
btrotter
Beginner
102 Views

Hyper-threading to discontinue???

Hi,

Does anyone know if the article on "The Inquirer" at:

http://www.theinquirer.net/?article=30087

is correct?

The article implies that Intel will be discontinuing it's hyper-threading technology.

Can anyone here either confirm or deny this?

The Inquirer isn't known for it's ability to report facts, and I'm hoping the article is a hoax...


Thanks,

Brendan
0 Kudos
3 Replies
TimP
Black Belt
102 Views

It's a partial truth, and partly a matter of opinion. Mobile multiple core models such as Core/duo, Merom/Conroe/Woodcrest do quite well without HyperThreading capability. Itanium/Montecito introduces a replacement for HT, called SOEMT.
The form which "many-core" will take is still a matter of research. There is a hope that work which has already been done to adapt applications to HT will show to greater advantage on multiple cores.
btrotter
Beginner
102 Views

Hi,

Firstly, thank you for the _very_ prompt response!

Mostly I'm wondering about desktop/server IA-32 chips (like the "dual-core with hyper-threading" Pentium Extreme Edition). If hyper-threading is discontinued for all IA-32 CPUs, then it'll become harder to justify supporting hyper-threading for my project due to the smaller range of CPUs that support it. I guess my real question is whether or not supporting hyper-threading is worth the development costs, for now and in the future.

To be honest, I'm hoping for a quad-core 64 bit 80x86 CPU (running at between 1 GHz and 2 GHz with a 1 GHz or faster front side bus) with 4 logical CPUs per core (and a nice 4-socket motherboard to go with it). It'd go well with Intel's virtualization technology...

If it's still a matter of research, then I guess I'll have to wait to see if I'm on Santa's "good" list... :-)


Thanks,

Brendan
SHIH_K_Intel
Employee
102 Views

Here's my perspective

The degree PC users can benefit from multi-threading (and their interest to purchase new hardware and/or software) depends on two factors:

1. The type of PC hardware support for multi-threading. This can be in the form of MP system (expensive and small volume), Hyper-Threading Technology (Mainstream pricing, mass volume with moderate MT performance benefits), dual-core processors, and quad-core processors in the future.

2. The availability of commercial software infrastructure that can effectively use processor hardware with multi-threading capabilities.

Back in 2003, the market place is such that (a) there are few MT applications for mass market; (b) Intel has sufficient manufacturing volume but can only devote limited die space in a processor package on hardware multi-threading. Hyper-Threading Technology is an efficient (die-space-wise) way to enable mainstream processor hardware with multi-threading capability, given a small amount of die-space per processor package. Even though HT only provides moderate performance benefit to multi-threaded applications, it improves UI and I/O responses. As a result, even though there are few MT applications, consumers can benefit with multi-tasking usages in the absence of wide varieties of commercial MT applications.

Moving the clock two or three years forward with silicon process improvement and feature size reductions, Intel can devote ~50% of die space on hardware multi-threading in a processor. However, this pace that Intel can introduce dual-core processors out-paces the speed that commercial software ecosystem can migrate from single-threaded paradigm to multi-threaded design. So the priority of enhancing hardware multi-threading feature at this time is on providing more MT performance benefit per processor package, versus the number of threads (logical processors) a processor package can support.

With respect to the availability of Hyper-Threading Technology in future processors, you should not rely on conjectures or innuendos purported by some web articles.

Although I am not at liberty to discuss future product roadmaps, I'll leave you with these thoughts:

In the IA-32 Intel Architecture Software Developer's Manual, we strive to provide guidelines to help multi-threaded developers to write code that works well with any form (or combination) of HT and multi-core capability. The latest version provides guidelines for MT software to use shared L2 on an Intel Core Duo processor efficiently, and it will continue to support MT software with guidelines to use HT features effectively because Intel values greatly the engineering investments that software developers had made in the past on HT.

As commercial software ecosystem for multi-threading proliferates in the future, I believe the priority in future processor feature enhancement relative to adding number of cores per package vs. supp orting more threads (logical processors) per package through efficient die-space, can grow (or alternate) in either direction. Either way, it will enable growth of multi-threaded software eco-system and end-user interests.

You can find the most recent edition of IA-32 Optimization Reference Manual at

http://www.intel.com/design/mobile/core/duodocumentation.htm

Together with the IA-32 Software Developers Manuals, they provide a unified solution for multi-threaded software to detect and recognize different processor topology and/or cache topology across any combinations of hardware multi-threading capability.

Reply