Slashdot Mirror
HP Introduces Final Processor in PA-RISC Family
The HP Way writes "According to an article on InformationWeek, HP announced the immediate availability of the 800 MHz, 1.0 GHz, and 1.1 GHz dual-core PA-8900 with 64MB on die L2 cache, the last member of the PA-RISC family of microprocessors. Customers with Superdome chassis can install Itanium 2 CPUs alongside PA-8900 processors."
There has been some speculation that the new computers from Apple which use an Intel processor will use an Itanium 2 CPU, which HP has used to replace the PA-RISC as their main workstation and server CPU. This indeed raises a very interesting question: if the new Macs do indeed use Itanium chips, would we one day see HP-UX running on a Mac? It is a real possibility.
Cyric Zndovzny at your service.
The high-reliability customers are not going to like this. Those machines run important stuff - 911, NASDAQ, power grids, VISA.
My hope is that IBM does not make the same mistake as HP, but instead continues with their AiX/PPC combination on workstations and servers. We need variety in the UNIX market to result in innovation and improvements. With IRIX and SGI gone, Compaq and Tru64 gone, and soon possibly HP and HP-UX (there are doubts that the Itanium can fully replace the PA-RISC), the major UNIX vendors left are Sun and IBM. Frankly, that may not be enough to provide a sufficient level of innovation.
Cyric Zndovzny at your service.
Indeed, HP has ported HP-UX to the Itanium. But there are doubts about the veracity of the Itanium as a stable, high-performance, fault tolerant, uptime-guaranteed platform. Part of going with HP-UX was knowing that you were running on PA-RISC, and you know you could trust your system. Now that has been taken away. It's as if HP-UX has been partially neutered.
Cyric Zndovzny at your service.
In the server market, only 2 RISC chips remain. They are the PowerPC by IBM and SPARC64 by Fujitsu (not UltraSPARC)[1]. Unfortunately for both chips, they do not enjoy the economies of scale that x86 enjoys (especially with the lack of future PowerPC Macs in the future), and development costs will soon become too great to support them. PowerPC may, barely, survive because IBM sells enough highend systems to support PowerPC R&D.
At this juncture, looking back 16 years ago, we can see whether the RISC movement was really hype. Remember Hennessy and Patterson from Stanford and Berkeley, respectively? They were foreseeing the end of x86 because of this new RISC "technology".
Yet, RISC was more marketing than technology. Remember branch slot delays? Remember uniform instruction widths? Remember instruction scheduling for load slot delays? Remember, in particular, that sentence in their famed textbook, where they claim that computer architecture will move beyond mere art and enter the realm of a quantitive hard science?
Well, history has shown that computer architecture remains more art than science. There is science, but it is only at the level of arithmetic for calculating cycles per instruction (CPI).
The supporters of RISC point to the RISC engine underneath the translation machine in the Pentium III. Nonetheless, the point that Hennessy and Patterson repeatedly made was that the "bad" x86 instruction set requires the translation layer and that, therefore, the translation layer would severely damage performance. Well, have Hennessy and Patterson looked at the latest numbers for integer performance on a Pentium 4?
Not surprisingly, Patterson has moved onto a new marketing job as head (?) of the ACM. He is now arguing that we desperately need to open the gates to H-1B engineers because America has a desperate shortage of good programmers. (Professor Matloff aggressively countered this marketing by using hard statistics and called Patterson a liar on CNet.) As for Hennessy, he became president of Stanford University. That job is also focused on marketing. I congratulate them both on their success. They were able to parlay their previous job of marketing RISC into signficant career advancements.
sidenote
--------
[1] Not ironically, the only two surviving RISC chips in the server or desktop markets was designed by native engineers, not H-1B engineers. As a matter of policy, IBM does not hire H-1Bs unless they have a Ph.D. and a critical skill. Fujitsu just, flat, does not hire foreign engineers; like other Japanese companies, Fujitsu prefers native engineers.
Ack, the flames, THEY BURN!
But seriously, there are far too many architectures around to keep running. Fine, perhaps the elegant ones with technical superiority didn't triumph over the cruder general purpose, but I can't imagine being a developer still trying to support a dozen processors. There is market room for at least 3, and possibly 4 architectures out there, and the fewer there are, the more software choice there is for each as developers are forced to move to successful platforms.
People who think they know everything really piss off those of us that actually do.
IBM are doing alright at the very high end, but the formerly Risc middle is moving towards AMD64 (including AMD and Intel clones), and most systems vendors haven't got all of the other business IBM have to support their chip development.
For all the criticism Itanium gets (I don't like its instruction set at all), Itanium2 is very fast and very scalable, and has had quite a lot of success at the high end. In the long run, given Intel's manufacturing capabilities, it might still have been the right choice for HP.
The point here is that the worst of all the CPU designs out there is the Intel one. Alpha, MIPS, SPARC, PA-RISC, POWER, PPC are all better designs. The reason they never really made the desktop is because they aren't Intel. This is the same rationale that lead to Windows, Word DOCs, etc being "the way".
It comes down to managers that don't know a damned thing about the tech, but making all the decisions on it. These other architectures had more growth potential, higher performance, and better overall design than any Intel chip released in the x86 line. The downside was mostly in channel cost. Since they weren't already abundant, they were expensive. If they were mass produced, they wouldn't be any different in cost than the x86 market is.
Look at how well the PPC is doing in the console industry right now. It was obviously a better choice than the x86 based chips or it wouldn't have been done. It obviously could be manufactured for the same price or less.
Two interesting tidbits. First, look into the iAPX-432 processor. Intel intended to kill off their 8-bit CPU line because in favor of that chip. It was 32bit, could do SMP, supported hot-swappable chips, and a host of other features. The 8086 was thrown in as a quick product to hold the company until the 432 was ready. Needless to say that the 432 never became popular as a result of the x86 line.
The second tidbit is that the Itanium actually needed an instruction set translator to run existing x86 apps. This layer was developed in partnership with HP. Intel *doesn't* maintain compatibility in their chips. They were trying to kill off x86 again, because it was a dead end.
The InformationWeek article reads: "HP was the first hardware vendor to bring out a RISC chip, releasing it in 1986, and PA-RISC--which powered HP Unix-based servers for high-end application processing--served the vendor well for years, [HP's Director of Server Marketing Brian] Cox said."
Is that true? IBM announced the PC RT on January 21, 1986. The first systems shipped in March, 1986. I can't find information that HP shipped any RISC systems that early -- and some evidence it was much, much later. Or do I need to read "chip" in Cox's quote extremely literally, in which case: (1) is that even true?; (2) who the hell cares if there's a whole shipping RISC *system* from IBM in first quarter 1986?
Are people buying Pentium 4s to run DOS in Real Mode?
The frightening answer to that question is yes. There are still a plethora of programs in a variety of niche applications (machine control, point of sale, etc) that still run in real mode DOS. Many of these applications rely on hardware compatibility with the original IBM PC. That is why they still sell Pentium 4 motherboards with ISA slots.
I don't know the details of the Intel architectures, but in the case of AMD, this is because all the prefetchers and branch-prediction logic happens during the decoding, so there would be no benefit, and probably need a whole bunch of extra circuitry to make it perform as well as the x86 stuff.
Itanium's often laughed at for sucking; but in some ways Itanium was the most successful bluff every played in the tech industry. In much the same way that Reagan's Star Wars bankrupted the Soviet Union got almost every single competitor to fold.
Back at the begining of the project, Intel was nowhere in high-end & 64-bit computing. There was HP (PA-RISC), Sun (Sparc), Dec (Alpha), IBM (Power), MIPS (SGI). Intel wisely picked the partner with the stupidest management (Carly) to give up their competitive edge and announce to analysts that Intel's vision/roadmap is so AwSuM that RISC is dead and that they're going to follow the bidding of their master Intel for their 64-bit plan. Wall Street bought in to the story so much that almost everyone else with competitive chips folded their strong hands to Itanium's bluff - SGI spun off MIPS and MIPS decided to leave the hgh-end space. Compaq undervalued Alpha and let it die. Sun tried to become a software company and if it weren't for Fujitsu making modern sparcs, sparc would be dead.
Basically, with nothing but PR and Carly's stupidity, Intel wiped out over half of the high-end computing processor market.
Thankfully AMD had the vision to see through the bluff, and saw the opportunity for 64-bit computing that worked; and thankfully IBM didn't have someone like Carly around so they saw the value in retaining competitive advantaces; or the computing world would be pretty bleak place right now..
Actually, it sounds less like it's related to managers and more related to simple market factors and economics.
People started to expect commodity machines to just simply be bought, plugged in, and left.
Once the PC stared replacing other type systems, the inertia of the x86 architecture, as you point out, kept those more affordable and available, and further eroded sales of those chips.
The problem with saying if the chips had been mass-produced, they would have been cheaper puts the cart before the horse. It's not "if you build it, they will come", but "if they buy it, you can produce in volume". Mass producing some of those exotic CPUs to the extent of pentium chips would have been absurd because there wasn't a market for them.
Unfortunately, a niche market product is either sold at a bazillion dollars to cover costs or eventually phased out due to lack of sales. Sometimes both.
Granted, all of the big companies merging into one another or giving up on making their own CPUs hasn't helped much either. But HP started becoming pointless and crappy when they merged with Compaq.
Lost at C:>. Found at C.
It comes down to managers that don't know a damned thing about the tech, but making all the decisions on it. These other architectures had more growth potential, higher performance, and better overall design than any Intel chip released in the x86 line. The downside was mostly in channel cost. Since they weren't already abundant, they were expensive. If they were mass produced, they wouldn't be any different in cost than the x86 market is.
I hope you're not the CEO of some company. If you are, let me know what your company is so I can short your stock on Monday morning. It seems that those managers know more about tech and finance than you.
What did you expect managers to do? Increase costs to get a processor that runs only 1% of the world's software? How do you suppose x86 became abundant whereas other architectures did not? It's all about the software... Tech savvy people know that. IBM, HP, Digital, Sun should have all forseen the tremendous potential in x86 software and started competing with Intel by making their own x86 parts.
The ISA is such a red herring. I am paid to design x86 microprocessors and I can tell you that there's an internal instruction set that is optimized for the hardware. Intel/AMD do a translation in hardware, Transmeta does it in software. In any case, x86 is just a thin layer around what is really going on inside. That layer amounts to a tiny amount of the chip power and area budgets. A small price to pay for the privelege of being able to run 99% of the world's software.
I will agree with you on one thing. Itanium is the dumbest thing Intel has ever done. No software - no market...