Some charts at the link.
--
Intel and Alpha convergence revealed
Exclusive: How the move will go
By Mike Magee, 30/06/01 10:41:08 BST
"Compaq is totally committed to Alpha as its high-performance e-business platform. It will keep and extend its performance lead over other 64-bit architectures, in particular IA-64" - Jesse Lipcon, now ex-Compaq guy, quoted on Alphapowered.com
AN ENGINEER close to Compaq's plans has delivered a presentation to the INQUIRER which indicates the direction Intel will take over the next 10 years as it moves Alpha technology into its existing Itanium roadmaps.
Although, he stresses, not all future options for the convergence of the chip technologies are yet complete, this is the broad direction Intel will take to integrate Alpha.
The long term technology and joint marketing agreement between La Intella and Compaq is already under way, with the full port of OpenVMS, Tru64 Unix, NSK (non stop kernel) and NSK middleware to the Itanic platform, he explained.
There will be "enhanced" server roadmaps for Himalaya, OpenVMS, Linux, Tru64 Unix and Windows up to 2003, likely to include a MIPS based Himalaya speedbump, the long-awaited 1GHz EV68 and the EV7-79 Marvel programme. Blazer will be Compaq's first Itanic Proliant four way system in this (Q3) quarter.
This is, in the early "roses, roses" days of the agreement, is how both firms see things. Time will tell whether these fall into the Burns 'best laid plans of mice and men' category or not.
Update
Compaq watcher Terry Shannon, who edits Shannon knows Compaq, had this to say after he saw our original story which we posted early today:
It appears that the architectural transition does not begin until mid-2004, apparently in the post-Madison IPF timeframe. This would make sense, since McKinley is a done deal, and Madison is well along in the development cycle. I wouldn't expect to see many Alpha artefacts materialise in IPF chips until Deerfield or Deerfield +1.
Taking a close look at the current Alpha roadmap, it seems that a 1.25GHz speed bump will be added to the midrange ES-Series lineup sometime next year. 1GHz ES45s are shipping now to HPTC customers who have ordered mass quantities of the systems, general availability is slated for 4FQ01. There's also an 833MHz speedup for the low-end DS-Series uni and dual processors, this may get announced on July 16 together with the 1GHz GS-Series and a new release of Tru64 UNIX. A 944MHz upgrade awaits the DS-Series early next year, the 1GHz upgrade has been pushed out until late 2002 for these systems.
A proposed 1.25-1.4GHz upgrade to the GS-Series is off the enterprise systems roadmap. Since the upgrade may have involved another multiple-organ transplant (another round of QBBs and FireBoxes) to support a proposed 16MB L3 cache, Compaq apparently decided not to bother. The Next New Thing for the high end will be the 1.2GHz EV7-based GS-Series Marvel in early 2003. Marvel and the ES-Series then get a 1.6GHz EV79 upgrade in 2004. (Thanks to the modular design of the Marvel platform, the ES and GS Series will differ primarily in CPU capacity; you should be able to start with a 2P ES-Series and take it all the way up to a 64P Marvel.)
Moving right along in the slide show, we see InfiniBand entering the Alpha scheme of things in mid-2003, and the IPF world in mid-2004. Since InfiniBand will form the I/O, system interconnect, networking, and clustering fabric for the post-Marvel system, 2005 seems like a reasonable timeframe for the first post-Alpha large-scale enterprise system. And lo and behold, the next-generation system will bear more than a passing resemblance to the QuickBlade architecture, as shown on the last slide.
Since the next-generation fabric-based system was designed to be architecture-independent right from the get-go, the reorganised AlphaServer Platform Team and lead architect David Fenwick should have no problem adapting to the Alpha-IPF CPU swap. This post-Marvel system has been under development for more than a year now, and initially targetted EV9 and IPF processors. The only technical "knothole" I foresee is delivering glueless SMP support (a la EV7) to IPF chips; the Blade strategy may render this a moot point.
The Himalaya NSK platform is omitted from the roadmap, but the NSK folks shouldn't have a difficult time coping with the CPU change. Five years ago, Tandem opted to base future Himalayas on IA-64 CPUs. Some initial work was done with Intel, and when Compaq three years ago announced that Himalayas would become Alpha-inside, Intel--perhaps with no small amount of foresight--decided not drop the Himalaya-specific chip attributes (deterministic behavior, etc) from the IA-64 design.
Accordingly, the Himalaya designers will take their original IA-64 plans down from the shelf, dust them off, and implement the processor-specific millicode necessary for the migration. The net result is that post-MIPS Himalayas will show up a little later than expected (2005 instead of 2004), but replete with the same code translation and optimisation technology that smoothed the path from CISC to MIPS RISC-based Himalaya S-Series systems.
Also omitted from the roadmap but critically important to Compaq's architectural "harmonic convergence" is marchitecture. Compaq has but one chance to get this right, and if the firm displays the same marketing prowess (or lack thereof) that Digital used to snatch defeat from the jaws of victory with the Alpha program, the outlook is bleak. To its credit, Compaq seems cognisant of the gravity of this strategic shift, and thus far seems to be acting accordingly. If the folks in Compaq's marketing war room can see the forest for the "Trees" and pursue an aggressive and proactive marketing strategy coupled with a credible and compelling message, the firm should be able to accomplish the transition. But time will tell. µ
theinquirer.net |
