The multicore duel begins
30 March 2005: Dueling Multicores: Intel and AMD fight for the future
At the beginning of 2005, Intel found itself having to integrate multi-core, 64 bit, and low-power technologies all at the same time, and it didn't yet have a clear roadmap to do so. So as a sort of stopgap measure, Intel introduced hyperthreading (HT) - a way to make a single core handle the workload of two cores. It did increase performance, and it was a fairly popular technology - the trouble was, it didn't lead customers where Intel wanted them to go.
Intel is now faced with the problem of whether HT truly is a stepping stone to its own approach to dual-core. Its challenges are these:
- Despite Intel's good intentions to accelerate existing single-threaded software, the company is asking software developers to utilize compiler switches to help optimize their use of hyperthreading. If performance gains average only about 15%, however, there may not be enough incentive.
- The fact that AMD uses shorter instruction pipelines - which means fewer available resources for distribution among multiple threads - had been touted by Intel as a design deficiency. But with AMD demonstrating a working x86 ("x64") dual-core Opteron, and touting benchmarks showing that Opteron outperforming Intel packages with HT turned on, AMD is characterizing AMD64 as not needing hyperthreading.
- HT does not lead consumers on a direct technological path to Itanium 2 technology; in fact, a clear technological barrier exists between HT and Itanium 2's EPIC approaches to parallelism. Conceivably, both approaches may be technically capable of co-existing - in other words, theoretically, EPIC code can run in an HT CPU. But once thread division is made at one level, by either software or hardware acting as an agent, further division by the other agent becomes far more difficult, if not altogether unnecessary. So whether pairing EPIC with HT would lead to negligible performance gains, or perhaps even performance losses, is unclear at this point. Further hindering Intel's decisions are reports that have already been published elsewhere - whether through a real Intel leak or a misunderstanding by reporters - that HT will be integrated into future versions of Itanium 2. At the time this article went to press, however, multiple Intel sources declined comment on this specific issue.
25 May 2005: Dothan over NetBurst: Is the Pentium 4 a dead end?
The Pentium M - a mobile processor, not a desktop or server CPU - became the forerunner of all of Intel's Core Microarchitecture. It solved the problem of power consumption when applied to a mobile platform, so why couldn't it do the same thing for a desktop platform? The team at Tom's Hardware Guide tackled that problem, and saw the truth perhaps before anyone else did:
Let us try to sum up the insights we have gained during the course of this little project...With the help of a simple socket adapter card and a BIOS upgrade, certain mainboards using Intel's 865/875 chipsets can be upgraded to use a Pentium M instead of a Pentium 4. Such a system offers up-to-date performance paired with low power requirements.
Additionally, we were able to raise the FSB from 133 to 160 MHz without any trouble at all. The result was that our 2.13 GHz Pentium M 770 ended up running at 2.56 GHz! At this clock speed, our two year old platform was able to beat the processor heavyweights Athlon 64 FX and Intel Pentium 4 Extreme Edition in all 3D games!
8 August 2005: Intel to trim power consumption of desktop CPUs in half by H2 2006
It was just under a year ago that Intel's real battle plan for the code-name it had leaked out over a year earlier, would finally come together. Unlike Microsoft, Intel knew that the second half of the year meant July, not December...or the following June.
Intel promised us in earlier conversations that its engineers are working on reducing the power consumption of its desktop processors. But it was unclear when this will happen. Documents seen by Tom's Hardware Guide now indicate that the new processor architecture code-named "Conroe" and scheduled for the second half of 2006 will deliver on this promise. If we believe our sources, then Intel is targeting a power consumption of about 60 to 70 watts per processor - or 30 to 35 watts per core.
