Itanium?

[3dilettante]

I think the latest Itanium version has separate integer and fp multipliers now.

The biggest issue I have with the itanium is more of a philosophical one, in that its heavy dependence on compiler optimization and profiling can make the chip too inflexible for an instruction stream that isn't spoonfed to it.

I do like the attempt to get some regularity in the ISA, rather than have a half dozen predecode and decode stages in the chip like x86 requires in order to figure out if a byte stream is any of several possible instructions.

It may turn out that future chips will be more like a more lithe Itanium than a buff Opteron, or a lot of tiny Itaniums. There are serious roadblocks on the way, and while ISA isn't the primary problem, it can be a serious bottleneck.

 

[T2k]

One thing is sure: Itanium is almost dead already. Intel sunk 1x billion and produced nothing but a bitter reality: nobody needs this piece of crap, except couple of out-of-world scientists, that's all.

It was a big, BIG, VERY BIG failure. Intel, as it is now, not gonna function anymore after a deadly serious like this: Itanium, Itanium2 are REAL burden, Prescott also a wasted non-performer, a bad joke against A64 and Tejas canned already - somebody has to go...

 

[Fox5]

Is itanium really much more powerful than opteron, in actual performancec but not theoretical? Why no itanium in a new console?(I'd imagine it beats opteron in theoretical performance, but it seems just about every chip does)

The Opteron 150 your using is AMD's latest. Wait for the 2.0 GHz Itanium 2 with 9 mb cache due out at the end of the year. IA-64 is better overall to the aging x86 platform; Intel just has to iron out the kinks.

Wow, 9MB cache? That's gotta be expensive, bet you could get like an 8 cpu opteron system for that price. Anyhow, the x86 system isn't really aging, isn't it more like replacing everything in your car, but keeping the exhaust pipe?

Is itanium designed like ibm's power cpus? Those kind of suck too, but at least are competitive.

 

[T2k]

Itanium is a totally screwed up design, fully against today's general computing needs - and being like this, totally depends on it's compiler technology. Which is totally new (nota bene: it's not compatible with anything) - which means Itanium won't mean anything in the next 10 years or so.
After then? It's going to be too old anyway.

IA64 has nothing promising as of now.

Waiting for 9MB cache as the "magic weapon" is simply silly: they HAVE to put more and more memory onboard to be able to compete with AMD's integrated memory controller.

Integrated memory controller - that's what made new AMD VERY succesful.

PS: I did not measure but if anybody needs, I can do some synth comparison between our dual Opterons and dual Itanium2s.

 

[mczak]

PS: I did not measure but if anybody needs, I can do some synth comparison between our dual Opterons and dual Itanium2s.

I'd be interested in speccpu2000 numbers with a recent gcc (linux) - all published numbers use either intel's compiler or a hp compiler. Somehow I have a feeling without auto-vectorization (which gcc at least currently lacks) it is going to be very slow...

 

[T2k]

PS: I did not measure but if anybody needs, I can do some synth comparison between our dual Opterons and dual Itanium2s.

I'd be interested in speccpu2000 numbers with a recent gcc (linux) - all published numbers use either intel's compiler or a hp compiler. Somehow I have a feeling without auto-vectorization (which gcc at least currently lacks) it is going to be very slow...

SSure, no prob - but you need to wait at least a week, I'm on a very tight schedule everyday.

 

[mczak]

PS: I did not measure but if anybody needs, I can do some synth comparison between our dual Opterons and dual Itanium2s.

I'd be interested in speccpu2000 numbers with a recent gcc (linux) - all published numbers use either intel's compiler or a hp compiler. Somehow I have a feeling without auto-vectorization (which gcc at least currently lacks) it is going to be very slow...

SSure, no prob - but you need to wait at least a week, I'm on a very tight schedule everyday.

Oh thanks. It's not like I absolutely need these numbers for anything, I'm just a curious person .

 

[pcchen]

I'd be interested in speccpu2000 numbers with a recent gcc (linux) - all published numbers use either intel's compiler or a hp compiler. Somehow I have a feeling without auto-vectorization (which gcc at least currently lacks) it is going to be very slow...

AFAIK it has nothing to do with auto-vectorization... IA-64 doesn't have much vector instructions. It's just that current GCC generates pretty bad codes for Itaniums.