When Intel bought the technology behind Compaq's Alpha processor last week, it may have done more than simply eliminate a potential competitor.
The deal includes the transfer to Intel of Compaq engineers whose expertise working with a non-proprietary chip technology could speed processor performance -- a constant Intel goal. But what remains to be seen is how long it will take Intel to integrate the technology, Simultaneous Multithreading (SMT), into its chip lines.
SMT technology makes a single processor act as multiple processors, spreading and handling the workload more efficiently. For example, when a standard processor multitasks, it has to drop everything from one task before it can pick up the next task, but a multithreading processor can take multiple streams, or "threads," of information simultaneously. Using these threads of data, an SMT processor can process from four to 10 instructions per clock cycle, while a standard processor would only be completing between one and four instructions per cycle. There is also no need to port applications to take advantage of SMT technology.
Compaq is largely viewed as much further along in its work on SMT than are other vendors. The company planned to add SMT technology to the EV8 version of its Alpha processor, but with Intel's purchase of the Alpha technology, what would have been the EV8 will now be built into Intel's Itanium architecture. Intel is so far along in developing future versions of its Itanium 64-bit processors that contributions from Alpha engineers may not show up in its chip families anytime soon -- soon, that is, in chip time. It could be a couple of years at least before Intel begins using SMT in Itanium, analysts said.
"That's something we'll start talking about somewhere further along this year," Otto Pijpker, a spokesman for Intel said. "We're not disclosing where and when you'll first see it, but we'll talk about it."
Ongoing industry talk has it that Intel wants to add SMT technology to its 32-bit processor series. Many in the processor community expect Intel will roll out SMT with its next Pentium 4 release. Intel has remained mum on the topic, but following last week's purchase of Compaq's Alpha technology, Intel confirmed that it has been working on SMT.
One of the theories behind Intel's code of silence on SMT is that the company was worried that SMT on its 32-bit architecture would take a bite out of its long-delayed 64-bit Itanium rollout. That theory doesn't hold water because one of the key advantages of 64-bit over 32-bit is that the former can access more than 4G bytes of RAM, said Nathan Brookwood, principal analyst with Insight 64.
"The difference between what Intel's doing with Itanium and what it's doing with the 32-bit line is not one where absolute levels of performance are going to make any difference in what people are going to buy long term," Brookwood said. "I don't think they'd downplay (SMT) on 32-bit just to keep Itanium looking good."
At least one analyst thinks that Intel's acquisition of the Compaq engineers and technology may not be useful to the chip maker and its Itanium architecture.
"Frankly, I'd be surprised if the Alpha guys add any value here, I don't even know how many of these guys are going to transfer to Intel," said Linley Gwennap, principal analyst at the Linley Group. When Intel purchased the StrongARM technology and engineers in a similar deal with Digital Equipment Corp. in 1997, all of the engineers involved in the deal decided to walk, rather than work for Intel, Gwenapp said. He estimated the team was made up of about 20 engineers.
Even if the Compaq engineers do transfer to Intel, SMT technology might not work as well as planned with the architecture Intel has chosen for its 64-bit processors. "I'm not even sure that SMT is a good combination with the VLIW (very long instruction word) architecture they have adopted," Gwenapp said.
Although the exact numbers are not known, Intel will likely gain "several hundred" Alpha engineers in the deal, estimated Intel spokesman Bill Kircos. "There's not an exact number, it's kind of a gradual move," he said. "Some of them will be working (at Compaq) on Alpha for a year or two."
The first group of engineers transferred to Intel will be those who were working on the Alpha EV8, a Compaq spokesman said. Although the complete transfer of Alpha to Intel will not be final until 2004, aspects of the technology and certain employees will move over between now and then.
As Intel has become more talkative about SMT technology, it is somewhat downplaying the significance of Alpha engineers who will be working on SMT for Intel. "Our engineers are already looking at that, it wasn't something we weren't looking at before," Kircos said of SMT.
If Intel does plan to introduce SMT to its 32-bit processors, you can bet that the company will quickly follow with SMT on Itanium. Compaq has said it will release one more version of the Alpha processor, the EV7, before it completely transfers the technology to Intel. The following processor, the EV8, which was to offer 4-way SMT, will now be part of Itanium's future, as part of the deal between the two companies.
"I think they'll be taking SMT technology and incorporating it into their 64-bit line," Brookwood said. "Who better to do that than people who have been doing that a long time with Alpha?"
However, Intel's current engineers and those involved in the project on the next generation 64-bit chip, code named McKinley, might not welcome the Alpha crew with open arms, Gwennap said.
Currently, Santa Clara, California-based Intel has most of its key designers at its headquarters, while most of the work being done on McKinley is taking place at Hewlett-Packard Co.'s (HP's) campus in Fort Collins, Colorado. Itanium is a joint project between Intel and HP. Neither of those teams is likely to sit back and let the Alpha team take over, Gwenapp said.
When all things are said and done, end users probably won't see the first products from Intel's newly acquired engineers and technology for some time. McKinley is already sampling, and much of the work has also started on the followups, code named Deerfield and Madison, so it's not likely that the Alpha team will have much input on those products either, Kircos said.
"There is a chance, but likely it will be after those two processors -- and we haven't even provided code names for those yet," he said of the successors to Madison and Deerfield.