And the IDF winner is…

Atomic’s IDF competition has finally come to conclusion. Many sleepless nights were had, both here at Atomic HQ as well as at Intel headquarters in the US, pouring over entries from finalists, choosing our favourites and arguing cases. After much

Atomic's IDF competition has finally come to conclusion. Many sleepless nights were had, both here at Atomic HQ as well as at Intel headquarters in the US, pouring over entries from finalists, choosing our favourites and arguing cases. After much debate, argument, a few hundred meg in emails and about two thousand dollars in international phone bills, Atomic HQ and Intel HQ have settled on one lucky person to go to the next Intel Developer Forum in San Jose, California.

And the winner of a trip to the Intel Developer Forum in San Jose, California, courtesy of Atomic Maximum Power Computing and Intel Corporation, is…

…Peter Sbarski!

Congratulations Peter on doing a fine job explaining what Hyper-threading is and why it's important. Explaining such a complex topic in an easy to understand manner is hard by itself. Making it an enjoyable and entertaining read at the same time is almost impossible. Almost.

We've already notified Peter, and we're happy to say that he was literally speechless after being informed we'd chosen him as winner. A quick reminder that he had the chance to write an article for print in a future issue of Atomic broke the stunned silence and brought forth a barrage of w00ting. He's obviously looking forward to the experience, and after you read Peter's Hyper-threading explanation we're sure you'll be looking forward to his IDF coverage too.

Every entry received was of top notch quality, making the choosing of an eventual winner an absolute killer of a task. Thankfully, it's almost the end of the week so we should soon have a chance to catch up on almost a weeks worth of lost sleep. In the meantime, checkout Peter's winning entry below.



Hyper-Threading: Maximum Power Computing

Hyper-Threading technology (codename: Jackson Technology) has been in development by Intel for more than 4 years. Back then we were excited over SSE instructions, Windows 98 and the upcoming millennium and when speaking of concurrent multi-threading we generally thought of multi CPU systems. However, back then in 1998, Intel set out to change that, so please stand up, take out your wallet and welcome Hyper-Threading.

Oh, what a complex thread we weave…
Multi-threading is a method used by the Operating System to split tasks over multiple threads of execution. On multi processor machines these threads can be executed independently at the same time by different CPUs thus providing greater performance and higher processing efficiency. With single CPU machines, however, the scenario is quite different. The Operating System can still create multiple threads but only one thread can be executed by the CPU at any time (single threading). We've been used to this paradigm for quite a while but Intel is trying to shatter our beliefs.

Hyper-Threading (HT) works by creating a virtual duplicate of your processor which in turn makes your Operating System think as if you have two totally separate CPUs. This forces the OS to work as if it was in a dual processor mode and send threads to both virtual CPUs. Now your processor instead of waiting patiently for each thread can work on two at the same time. This provides greater CPU utilization, speeds up all processing tasks and uses resources which would have otherwise been idle.
However a single processor with Hyper-Threading will not achieve the same results as a true dual CPU computer. Both virtual processors must share the same cache and there is still only one bus. This also may lead to nasty effects such as collision which happens when multiple CPUs attempt to write to the same memory address or two instructions being given to the same execution unit at the same time. However processors equipped with HT will have a couple of safeguards to deal with such problems.
Note: only dual CPU compatible systems such as Windows XP, 2000 and Linux will take advantage of Hyper-Threading. Windows 9x systems will not support HT.

Hyper-Threading technology is a variation of simultaneous multi-threading technology (SMT) which works by creating a virtual duplicate of the “architectural state” of the CPU. Hyper-Threading provides thread level parallelism (TLP) which is different to instruction level parallelism (ILP) used in all current desktop processors. Hyper-Threading is transparent to the Operating System; no updates should be needed to use it. All current software which supports multi-threading (e.g. Quake 3) will be able to take advantage of Hyper-Threading. However, further optimization is possible and highly recommended by Intel. If you would like to read more on the specifics of Hyper-Threading (including a few interesting whitepapers) have a look at: http://developer.intel.com/technology/hyperthread/

Real World and Performance
According to the latest rumours from sites such as www.theinquirer.net and www.xbitlabs.com desktop Pentium 4s with Hyper-Threading should be coming out in the very near future. As a matter of fact, all current Pentium 4 CPUs and Xeons have the necessary HT circuitry included; alas it is disabled at the factory and there is no way to turn it on once it has been turned off. The first processor with HT enabled will be a 3.06GHz Pentium 4 (Northwood Core, 0.13 micron). It should come out in the 4th quarter of 2002 (possibly in November) or the start of 2003 at the latest. Every CPU after that (Northwood, Prescott, etc...) will feature HT. All motherboards supporting 533MHz FSB will be completely compatible with Hyper-Threading. Motherboards with 845GV, 845GE, 845PE, 845G-B, 850E and 845E chipsets will support it.

Intel claims an “average improvement of 40% in CPU resource utilization”, ability to handle larger workloads and an increase in the number of transactions that can be processed. Intel's whitepapers show performance increases of up to 20% in certain tasks. Sources on C|NET claim a 30% improvement in graphic application tests (e.g. Maya) on Xeon chips.
A few months ago Anandtech (www.anandtech.com) ran a small number of server related benchmarks comparing Athlon MPs and very early Xeons with and without HT enabled. The benefits of having HT enabled ranged from 3 to 17 percent although one test showed a drop of 5% in performance. Have a look at the website for some interesting details.

Will it make a difference to you?
No matter what the actual performance benefit is, and there will be one especially if developers optimize their code, Hyper-Threading technology is a winner. Even a modest increase in performance of 3% can't have too many critics especially since it will be free and HT can always be turned off if need be. Kudos to Intel for not, blindly, concentrating on the megahertz race but for trying to innovate.
The first P4 with HT, at launch, will cost about $635 US which is about $1200 AU at the current rate and hopefully after the IDF (9-12 September) we will have a lot of interesting benchmarks to analyze.

By Peter Sbarski

• All Judges decisions are final. No correspondence will be entered into.