Slashdot Mirror
HP Terminates Itanium Workstations
vincecate writes "The largest Itanium system maker,
HP, has terminated its Itanium workstations.
It seems their workstation customers have spoken in favor of x64.
In related news, Intel expects to ship
over 100,000 Itaniums in all of 2004
while AMD is estimating
1.5 to 2 million AMD64 chips in Q4."
AMD sold around 100,000 Opterons in Q2 however. This should increase to 200,000 in Q3 given recent products from HP, Sun, IBM etc, especially with the increase in 4P systems.
Of course, the ASP of Itanium is a lot higher, so Intel need to sell a lot fewer Itaniums to get the same money back as AMD. On the other hand, AMD haven't sunk $billions into K8!
Check out this article: IBM mocks Itanium server sales - again, make sure you look at their very amusing graph of changing sales forecasts.
Of course, they're pledging to continue selling Itanium servers.
In the longer run, IMHO it sounds somewhat problematic, considering that all the engineers developing software will be running on x86-64. I.e. the software will first be available on x86-64, more tested etc.
So why should the customer shell out money for an Itanium server instead of an x86-64 server which has better bang-per-buck and runs the software more reliably? In the short run HP can probably contain x86-64 in low end servers, keeping high end stuff reserved for Itanic. But in the long run, they'll have to start providing higher end x86-64 gear too, or their customers will move to a competitor that will.
Exactly what I was thinking.
HP and Intel deserve this for killing off the two most powerful processor lines in history.
Back when PA-RISC and Alpha were in production, the gap between them and the next fastest CPU lines were staggering. I used to check the CPU Info Center at Berkeley every time a new one was released, just to see how badly it humiliated the competition (sadly, the CPU Info Center is no longer maintained).
The Athlon (before it was named such) uses the Alpha's bus... and the original slot-A design was compatible with both the Alpha and the Athlon, all you would need to sell a motherboard for the other one is a different BIOS. This was the selling point that convinced many motherboard manufacturers to actually make these boards. Unfortunately, only a tiny handful of companies actually marketed the resulting systems using the Alpha CPUs (mostly in Linux Journal & Linux Magazine as rackmount servers).
They could have done so much more... oh well.
My current favorites are UltraSPARC and PowerPC (with POWER close behind).
- Preferences: Solaris 10 (servers), Ubuntu (desktops), Solaris 11 (personal servers) -
"AMD chipsets support slower CPU to memory times than Intel (32bit or 64bit) counterparts. "
Bzzzt... wrong answer.
In AMD64 chips, the chipset doesn't have the memory controller - it's in the CPU.
AMD's CPU-to-DDR latency is much lower than Intel's.
I can't believe this was modded up...
A64 FX's and Opterons support dual channel ddr and have much lower latencies then intel at the same mhz (400mhz ddr X 2). Usually the FX's and Opteron's win the memory bandwidth benchmarks.
As low power AMD has a line of mobile barton core processors that use as little power as 45 and even 35 watts. They can also be placed in a destop motherboard. The xp2400 35 watt is also under $100. But there is a good chance the pentium-m uses less power but they are only found on certain laptops (can't be bought).
Hmmm... Pie...
Predication and explicit speculative loads were primarily added to the IA-64 architecture because they'd decided not to implement out-of-order execution (dynamic scheduling) and other dynamic techniques*. They aren't nearly as important on a modern superscalar processor.
(* Itanium2 doesn't even do next-line prefetching!)
Explicit speculative loads was a major mistake because in many kinds of code the compiler cannot place speculative loads far enough ahead of the actual use for it to pay off. Often the address to be loaded from is simply not known far in advance of the load (consider executing the C code "x = a->b->c"). So Itaniums spend a lot of time stalled waiting on memory accesses. That's why Intel spends so many transistors on gigantic on-chip caches, to try to reduce that pain. The architecture's pretty good for workloads with very regular and compiler-analyzable access patterns (regular number crunching, SpecFP) but it's bad for everything else (servers, user applications, irregular numeric codes).
Yes, IA-64 is a aggressive, radical, clean and somewhat novel design, so it's understandable that some geeks love it. However, it is not a good design.
If it was a good design, then with Intel engineering, 5x the transistor count, and no backward compatibility requirements, it would be absolutely crushing Opteron performance. Instead it is merely competitive.
BTW it is quite odd to consider IA-64 a small tweak over RISC chips. IA-64 is the most dissimilar of all viable architectures today.