|
OF COURSE, and as we reported long back, Intel could have implemented hyperthreading in the Pentium 4 a whole year ago. But it didn't. Instead it fused the transistors, except on engineering samples, to test the performance before the launch of its next 3.06GHz Pentium 4 chip on the 25th October. Probably.
It's interesting to have the view of a former Pentium 4 designer on the architecture and on hyperthreading.
Brinkley Sprunt, who is an assistant professor of electrical engineering at Bucknell University is one such. When he worked at La Intella, he was a member of the architecture teams on the 80960, on the Pentium Pro (what a great keyring that was), and the Pentium 4.
So what he has to say about the Pentium 4 in a lengthy learned article on the web is most interesting. The article covers performance monitoring but also provides some invaluable information on hyperthreading and on the architecture itself.
On page 76 of the document (page five of the PDF), he makes the point that SMT (simultaneous multithreading), is a sword with two edges. In fact, he claims that the best case argument for hyperthreading is that performance will double, while the worst case is that performance will half. No-one wants half the performance, we're sure you agree.
It all depends, says Brinkley, basing his arguments on the Xeon, which has had hyperthreading for a while, and which -- let's face it, although it has a different name, is much the same beast as the Pentium 4. Although obviously more robust.
If Brinkley is correct, then the claims made by Intel last week about 30 per cent better performance, or in some cases 75 per cent better performance, or in some cases no better performance, will need some examination when the 3.06GHz is released.
He also frequently refers to bugs in the Pentium 4 which go undocumented and states its current implementation has "problems and limitations". He says "the most significant of these is the lack of documentation for performance monitoring capabilities, implementation bugs, and undisclosed features".
Here's the PDF from Brinkley (_http://computer.org/micro/mi2002/pdf/m4072.pdf). And we notice from Intel's update page, which you can find here(_http://developer.intel.com/design/pcn/Processors/D0102779.pdf), that customers should move to the B1 stepping and to the C0 stepping, as we revealed in August. But if you want a C0 stepping, you're going to have to wait until early next year.
Intel says in reply to a letter we posted last week that there is absolutely no shortage of Pentium 4s at 2.8GHz. But if you want a C0 stepping, you're just going to have to wait. µ
|
Linear Mode
