Meanwhile, on the other side... (Itanium)
MacSurfer linked to a short, sweet article in the New York Times Online on Intel's defense of the Itanium at an analyst meeting.
The highlight of the article (emphases mine):
Ouch.
So: Intel is committing to Itanium. They expect to increase sales volume by a truly staggering factor. Their president is paraphrasing Steve Jobs, and the whole line is being dismissed as "a fiction." Contrast that with Apple's last analyst meeting!
Meanwhile, AMD has announced that they're building a new fab in Dresden, to be completed in 2006. The article doesn't even mention PowerPC, but if Intel's president is borrowing Steve's lines then you know Intel's watching. IBM, did, after all, sell more than 100,000 PowerPC 970s last quarter.
The highlight of the article (emphases mine):
Quote:
"At the risk of getting myself in a lot of trouble, I'm going to declare this the year of Itanium," Paul Otellini, Intel's president, told a gathering of analysts in New York.
Mr. Otellini said that more than 100,000 Itanium microprocessors would be shipped this year, a volume that is comparable to that of individual lines of microprocessors used in server computers running the Unix operating system. Mr. Otellini added that 40 manufacturers now sell Itanium-based server computers and workstations, and that there are now 1,000 software applications that run on the Itanium chip.
But industry analysts say Itanium has a long way to go, probably three or four years, before it becomes used in the mainstream of corporate data centers. "Until or unless Intel offers an alternative, they have to push Itanium," observed Rick Whittington, an analyst for American Technology Research. "But what Paul Otellini is saying is a fiction. The customers are not voting his ballot."
"At the risk of getting myself in a lot of trouble, I'm going to declare this the year of Itanium," Paul Otellini, Intel's president, told a gathering of analysts in New York.
Mr. Otellini said that more than 100,000 Itanium microprocessors would be shipped this year, a volume that is comparable to that of individual lines of microprocessors used in server computers running the Unix operating system. Mr. Otellini added that 40 manufacturers now sell Itanium-based server computers and workstations, and that there are now 1,000 software applications that run on the Itanium chip.
But industry analysts say Itanium has a long way to go, probably three or four years, before it becomes used in the mainstream of corporate data centers. "Until or unless Intel offers an alternative, they have to push Itanium," observed Rick Whittington, an analyst for American Technology Research. "But what Paul Otellini is saying is a fiction. The customers are not voting his ballot."
Ouch.
So: Intel is committing to Itanium. They expect to increase sales volume by a truly staggering factor. Their president is paraphrasing Steve Jobs, and the whole line is being dismissed as "a fiction." Contrast that with Apple's last analyst meeting!
Meanwhile, AMD has announced that they're building a new fab in Dresden, to be completed in 2006. The article doesn't even mention PowerPC, but if Intel's president is borrowing Steve's lines then you know Intel's watching. IBM, did, after all, sell more than 100,000 PowerPC 970s last quarter.
Comments
Originally posted by Amorph
MacSurfer linked to a short, sweet article in the New York Times Online on Intel's defense of the Itanium at an analyst meeting.
The highlight of the article (emphases mine):
Ouch.
So: Intel is committing to Itanium. They expect to increase sales volume by a truly staggering factor. Their president is paraphrasing Steve Jobs, and the whole line is being dismissed as "a fiction." Contrast that with Apple's last analyst meeting!
Meanwhile, AMD has announced that they're building a new fab in Dresden, to be completed in 2006. The article doesn't even mention PowerPC, but if Intel's president is borrowing Steve's lines then you know Intel's watching. IBM, did, after all, sell more than 100,000 PowerPC 970s last quarter.
I can tell you first hand that I have no interest in Itanium at this point. We are looking at IBM or AMD for our next round of 64 bit Linux machines. The Itanium isn't even on our radar screen right now! We are more interested in what Apple might offer in the next 12 months then the Itanium.
6,500 for G5.
Yeah, I know, most apps aren't optimized, but at least they run at full speed, not so with the Itanic.
And why use the Itanic when better alternatives like the Opteron are available?
The only bits and pieces that I know is that Intel once upon a time once positioned this chip as the next chip after the Pentium, and that it was supposed to be 64-bit...
Originally posted by rampancy
Could someone give me a short history of the Itanium?
Part of the Itanium's problem is that it had a very long history before it ever saw the light of day. The IA64 project began in 1994, the same year the first PowerPC rolled out.
Basically, Itanium is a cooperation between Hewlett-Packard, Intel and Microsoft (with Microsoft not doing all that much except cribbing from HP's compiler team). The idea was in fact to replace the creaky x86 architecture with something new and slick and modern.
The architecture they settled on was an HP idea called VLIW, for Very Large Instruction Word. The idea was that you could have the basic unit the CPU handled be 128 bits wide and packed full of instructions and data, which the CPU would execute in parallel, resulting in absolutely massive performance as it devoured programs in much the way AltiVec devours data. This raised an immediate issue: Since the instructions arrive packed, how does the CPU cope with dependencies (e.g.., an instruction whose result depends on the value of an immediately previous instruction, and which therefore can't be executed parallel to it) and branches (if a, do b; else, do c) - branch misses (where the CPU guesses whether a is true or false in order to speed execution, and guesses wrong) are especially painful in this design. So, what to do? The team looked at the immense talent of HP's and Intel's compiler engineers, and decided to punt the problem to the compiler.
So, the compiler basically runs the chip: It decides which instructions and which data end up packed into which instruction word; it identifies and groups independent instructions into predicates, setting flags so that the CPU knows what it can and can't run in parallel. It tells the CPU which way to go on a branch, and it can even tell the CPU to execute more than one possible outcome at once. It even manages the caches! The CPU, then, has a minimum of cleverness, and so it will be easy to design, scalable, and ridiculously fast. It would out-RISC RISC in chasing complexity out of the CPU and into software, where it belonged.
Well, that was the theory.
4 years later, Intel convinces SGI to drop their investment in MIPS processors, because in 1999 this great new 64-bit line will come out and blow everything else away. SGI spins off MIPS, and prepares to move to IA64. 1999 comes... and goes. 2000 comes... and goes (but not before Intel has used this carrot and the stick of a patent suit to get Compaq to abandon Alpha for IA64 - which, remember, still hasn't appeared). SGI is forced to use Intel's other, 32-bit processors in the meantime, and they begin going into a nose dive. Finally, in 2001, the first IA64 chip appears. It's huge. It's hot. It costs a fortune. It runs at 800MHz. Its performance is pitiful, all things considered. SGI and Compaq wonder why they abandoned a perfectly good platform (MIPS) and a fire-breathing monster (Alpha) for this? Intel tells everyone that, well, this wasn't the real thing. Wait until next year. Next year comes, and goes.
Itanium finally appears. It actually can crunch numbers with the big boys this time, although it's still ridiculously hot and expensive considering its design goals. Again, people are left to wonder why two perfectly good 64 bit platforms had to die for this thing. The reaction of the competition could not have been inspiring: Sun, smug as ever, reaffirmed its commitment to its own 64-bit SPARC architecture; AMD drove the Opteron and the Athlon 64 through the huge hole left by Intel's unsuccessful attempt to drive the market away from x86; and IBM, smirking down from its perch on the mighty POWER4, got ready to roll out the PPC970 - and a faster POWER4. Oh, and several powerful customers demanded the release of the last, baddest Alpha after Compaq had formally killed it. So the last member of the near-legendary chip family comes out for one last round and promptly runs away with the field.
And so here we are: Intel, after getting most of the competition out of the way with FUD, flops, and suddenly finds its flop surrounded by fierce competition. Having invested nearly a decade and untold millions into the project, they and HP are putting on a brave face; but the Itanium is currently being handily outsold (in unit volume) by the Xserve.
The performance achieved by the latest IA-64 design is largely due to brute force and really massive on-chip caches. The transistor count of this beast is pretty impressive (>220 million)... equivalent to four PowerPC 970s. IBM's new POWER5 will have roughly the same count, but that will include two SMT cores.
Originally posted by Amorph
...several powerful customers demanded the release of the last, baddest Alpha after Compaq had formally killed it. So the last member of the near-legendary chip family comes out for one last round and promptly runs away with the field.
A fascinating story. Really, someone should take all of these stories and put them into a book, like what Owen Linzmayer did with Apple Confidential; but I digress.
Anyway, I'm curious about your comment on the Alpha being dead, because I read on another website that the Alpha didn't really die. It just became the Athlon, since AMD bought out the patents and the whole Athlon design team moved to AMD...and as a result, the Athlon is little more than an Alpha with an x86 front end bolted on...
[edit: formatting...]
So if VLIW CPUs are meant to use the compiler to deal with OoOE, caches dependencies, etc., then why is it so huge? What do they do with all those transistors?