Intel's plan to unveil its copy of AMD's Hypertransport technology at next month's Intel Developer Forum (IDF) has been spoiled by a rather thorough analyst. Back in June, Intel confirmed some dates around its CSI interconnect for the first time, saying the technology will appear with products that ship in 2008. CSI, which will …
"A turning point for the industry"
And another nail in the coffin, if any more were needed, for Itanic.
Once Intel's AMD64-instruction-set chips have the same system interfaces as their "bigger" (but far less profitable for Intel) Itanium siblings, one in vaguely the same class as Alpha's EV7 interconnect (now more widely known as AMD Hypertransport?), that's one less USP for IA64.
So then who's going to want to buy (or even build) IA64 systems? Presumably the only customers interested will be those unfortunate ones whose vendors have locked them in to IA64-specific software, as all the rest will have the choice of x86/AMD64 instead. x86/AMD64 customers will be offered similar CSI-based hardware, but will likely pay much lower hardware prices than the Itanic addicts, and the x86/AMD64 market will still offer a far greater range of software than the unfortunate locked-in HP-UX and OpenVMS customers can choose from.
Having a common system bus *might* simplify life for the low-volume hardware assembler, but Itanic can't/won't sell in the serious volume market, and in the low-volume "niche" market (SGI?) the common system interconnect doesn't change the economics all that much.
If an Itanic Proliant emerges (again), I'll believe the tide has turned. I'll also eat my hat, especially if Proliant Itanic emerges and sells to worthwhile numbers of paying customers (rather than a handful of Intel-funded seed sites).
finally some good news
We should be dancing in the streets over this were finally
getting rid of that stupid fsb bottleneck everywhere not
just AMD. I think this will be bigger than most people
realize just my dos centavos.
One more nail...
In AMD's coffin. Which is kinda scary, since without AMD, where is Intel gonna get all it's processor design from?
Arrrggghhh, ya Intel/AMD fanboies!
Who really gives a d*mn about Itanium anyway? I certainly don't care if CSI hurts Itanium and I know my neighbours are not too worried either. Really, Intel just keeps Itanium around because it's a good internal pork barrel for research money... probably a nice tax loss too.
Or.... maybe CSI is a straw man designed to distract people when AMD drops Barcelona next months and activates it's secret plan to transform Intel into a BOLUS OF BLOOD AND SILICON GREASE and throw that bolus onto the ash heap of history! Aaahhaaahhhaahh!
Seriously, though, nifty as all this computational power is, does anyone besides eng/sci researchers and FPS game developers have a real use for it?
There is some stuff floating about the web that strongly suggests CSI will work best in 2 to 8 CPU configurations, so really it just means Intel may compete better against AMD in the 4+ CPU server market. I definitely wouldn't mind seeing the price of quad CPU servers drop like dual CPU server prices have in the past 5 years. I work with physicists, they are always calculating horribly abstruse sums and redlining the CPUs in their dualie boxes. I'd love to offer them four or eight quad core CPU boxes at a good price.
Computing power for more than just physicsits!
I would think anybody who needs image processing, audio processing and such like things would make good use of the computational power. This goes for people working in marketing, CAD, music production, sound design, and all those horrible "home design" apps people buy in places like Asda,
Convergence towards Singularity?
Reading the paper from David Kanter, and boy, is IT Real SMART or is IT Real SMART, had me thinking that Sun and ITs UltraSPARC T2, ... http://www.theregister.co.uk/2007/08/28/sun_ultrasparc_t2_processor/ .... are also playing the same Games, albeit on a different parallel level dimension ..... and probably also on all three .
AI Calling? Human Fate or Man's Destiny?
"Seriously, though, nifty as all this computational power is, does anyone besides eng/sci researchers and FPS game developers have a real use for it?"
Yes, ARGonauts are using IT for Real, Virtually. Seriously, I Kid U Not.
And their SMART Never-Ending, Open Ended Quest?
Nothing less than the Holy Grail of Command and Control aka UltiMate Powers ....... which should keep them busy enough if they are Good enough.
SO many nails, so little death....
<Yawn> And the Itanium-bashers have heralded Itanium's death yet again! How many times is that? Let's go over this just one more time - Itanium scales, Xeon does not. I'm not even going to try to explain to the psedo-technicians out there about the register count, the either-endian capability, or anything else, as - just like the unscientific drivel of the "Troofers" - they just keep on squealing. I really wish I had a dollar for every time there's been "another nail in Itanic's coffin"! In fact the coffin seems to be all nails!
And as for comparing T2 to Itanium, that's a bit like comparing four scooters roped together to an articulated lorry. And I can't wait to see what AMD come up with next, I'm sure they have no intention of throwing in the towel just yet.
"Itanium scales, Xeon does not"
Do you have any specific independent benchmarks in mind to back that up? Ideally reasonably recent ones where cost and availability comes into the picture somewhere?
How many CPUs do you want to "scale" up to (obviously if >8, hardly anybody cares, just ask Unisys how well their ES7000 multiprocessor has sold, the one that was briefly available badged as Compaq Proliant 9000 till they realised no one was interested)? Yes Superdome etc go to even more processors. So what? How many 128-CPU applications are there these days, even if server and datacentre consolidation and thinner clients (a return to virtual timesharing?) is going to be all the rage for the next few years?
Anyway are you sure you're not confusing hardware scalability with OS scalability? What scales with the high end Itanium systems is just as likely the OS (and maybe application) software, not necessarily just the chip/system setup. Pick a benchmark where the same decent x86 server hardware is used to run the same benchmark on both Windows and a real OS (Linux? Solaris?) The same hardware will typically deliver far worse scaling on Windows than with a real OS.
I thought a major point of CSI was to allow Intel's next set of AMD64 stuff to finally scale about as well as AMD's kit has been doing for the last few years, by getting rid of bus congestion? If CSI means the x86 bus is becoming just as scalable as the Itanic bus, and the apps and systems are there in volume on x86, and the few that are available on Itanium cost a fortune, it's not a difficult decision is it. *Unless* your system wants HP-UX or HP/Compaq/DEC OpenVMS of course (or whatever the HP/Compaq/Tandem NonStop OS is called these days).
Intel don't have an infinite amount of money to play with, and won't want to risk wasting too much more of it on Itanium when almost nobody except HP is interested, when Intel could invest it more profitably elsewhere (eg to prop up their WiMax efforts). After all, HP have been using those pesky Opterons in Proliants and other AMD stuff across the whole HP x86 range, what kind of brand loyalty is that?