new Itanium 2 spec:scores 1408/2553 for $850!

[Mephistopheles]

From Paul DeMone:

1.6 GHz, 9M L3, 400Mhz FSB
- 122 W TDP, 1590 SPECint_base2k, 2712 SPECfp_base2k
- list price $4226, qty 1000

1.6 GHz, 3M L3, 400Mhz FSB
- 99 W TDP, 1408 SPECint_base2k, 2553 SPECfp_base2k
- list price $851, qty 1000

More interesting is the advancement to 533Mhz FSB. Also, A 2500+ score of SPECfp_base is quite an excellent score for a processor that costs about $850, wouldn't you think? Plus, quite surprisingly, the $850 processor has quite a respectable amount (like 90%!) of the 1.6 9M model's performance!!! (costs 5 times less).

Looks like Intel is slowly getting more price-conscious with Itanium 2. I wonder if it will be enough?...

I mean, if dual-processor machines with 1.6Ghz I2s would be available in the workstation market for $850 for each processor, that's progress, but Opteron still is that much more of a practical, conscious purchase, regardless of the fact that it scores like maybe 40% less in floating-point applications. (consider price/performance and Opteron <i>still</i> has the advantage...)

So Intel is going to have to do more than that, but it's looking quite good anyway.

<i>Edit: On another note, and to Itanium's credit, the 1.6Ghz 9M model can apparently score ~2700, which matches the 1.9Ghz Power5 servers from IBM quite nicely and squarely. And the IBM servers have huge amounts of cache as well. Looks like Itanium just got a bit stronger, indeed. Not enough to truly change the politics involved, but...</i>

<P ID="edit"><FONT SIZE=-1><EM>Edited by Mephistopheles on 11/09/04 01:21 PM.</EM></FONT></P>

 

[Unseen]

NASA bought 10k, 1.5 ghz Itanium 2 for thier supercomputer. Which now the 2nd most powerful supercomp out there. Seems like some ppl are still buying them, they just might have some future left in them.

 

[Mephistopheles]

A dual 1.6Ghz, 3MB Itanium system with 533Mhz FSB is actually not as unattractive as it could be... and the 533Mhz FSB (8.5GB/s) more than makes up for the 1/3rd cache size. I'm not sure that that processor is the one that scores 1400/2550, but anyway, things are getting better in itanium land.

 

[Mephistopheles]

More information:

The 1.6Ghz, 3MB dual-processor version <b>with 400Mhz FSB</b> scores 1405/2553 for $850 indeed. This is actually cheaper than any P4EE processor from Intel - and it will most certainly trounce any P4, even heavily overclocked, in floating point (though not in int).

The 1.6Ghz, 3MB dual-processor version <b>with 533Mhz FSB</b>... well, I couldn't find the scores for that yet, but it's pretty reasonable to assume that this $1150 cpu can get a score of, say, 1500-1600 in int and 2700+ or 2800 in FP, which probably makes the $4200 9M model obsolete. In any case, the 533Mhz FSB will probably be a good addition to the lineup.

 

[raretech]

That price point should make them pretty attractive for scientific/3d workstations, that is as far as workstations go. I think we'll be seeing these suckers popping up in render farms and clusters for a while to come. Good news.

<i>Nemo me impune lacesset</i>

 

[trooper11]

ok this is where i have questions. Your nto trying to say that itanium would outperform the p4 in the areas that the p4 is used for are you? we arent talking about the xeon, but the p4. xeons are more used for workstation then p4s. where are the real world numbers for the itanium? i really want to see what it can do other then high end server applications. thats how itanium will survive and grow, by being an option in more then just high end super computers.

the price point is there now, so where are the numbers? i would think IA64 would now become the biggest obsticle to adoption, since its now coming into direct competition with x86-64. but i can see intle is really going to do whatever it takes to force its adoption. first it sets out plans to converge the itanium and xeon platforms, now it cuts prices on itanium. it looks like intel might try to get rid of xeons all together except for perhaps lower end then it is now, but again, IA64 could block that. but ill wait to see numbers.

 

[Mephistopheles]

Well, Itanium is a very strong performer in native code, but there's no way it will do desktop very soon. At best, once we get to use more parallel code because of multicore, we'll get to the point where the explicitly parallel architecture (EPIC) of itanium will be less of a stretch.

If there were, for instance, IA64 versions of renderers like 3dsmax, I'd bet all my money on Itanium anytime. Neither Opteron nor P4/Xeon can hold a candle to the raw floating-point performance of Itanium...

 

[trooper11]

of course, thats my point exactly,t here isnt such software, so you cant really say its competeing against p4s at all at this point. its only the first step. they sitll have to get a few more things to fall in place first.

 

[Mephistopheles]

Noone said that, anyway.

But it is most definitely a very, very good performer.

 

[trooper11]

i was refering to saying its a great performer. you have to always follow that with something like 'in servers' or 'high end areas' ok maybe im being nit picky, im just saying you drop doom3 on it and its not as great a performer. granted it would need to be ported lol, but you get what i mean.

 

[darko21]

<<Well, Itanium is a very strong performer in native code,>>

So is apple in it's native code.

<<but there's no way it will do desktop very soon.>>

Sure it will right behind sparc or ibm powerpc on the consumer desktop.

<<At best, once we get to use more parallel code because of multicore, we'll get to the point where the explicitly parallel architecture (EPIC) of itanium will be less of a stretch.>>

The "once we get to use more parallel code" is valid but do you really think ((Public software will be ia64?) cause I think it will be x86 32 and 64. So parallel code will come about for dual core although it will take a while, and 90% of thoses will be dual x86-64bit, 8% for dual core macs, and what about ia64, what about sparc, what about ibm powerpc?

I guess what I am trying to say is, who cares if some obscure increadibly expensive cpu (Itanium) that runs only on expensive custom OS and software can crunch #'s on a synthetic benchmark and looks imperessive. Yes Itanium can run x86 in virtual mode allbeit slowly, yet so can the mac. The mac is no threat to x86 and neither is intanium.

Mephistopheles I'm curious why you never pump apple? Surly you can agree the Macintosh 64bit will greatly out sell the Itanium, at least in volume and for good reason lots of great software and resonable prices.


If I glanced at a spilt box of tooth picks on the floor, could I tell you how many are in the pile. Not a chance, But then again I don't have to buy my underware at Kmart.

 

[Mephistopheles]

Well... Itanium isn't so insanely expensive if an $850 I2 can give 1400/2550 scores, if you want heavy floating-point capabilities like scientific workloads. It's much, much better than the P4EE for that price.

Also, about apple, there are already people enough defending it.

On another note, I wonder how much time will spec.org take to publish the results for the 1.6Ghz, 3MB, 533Mhz FSB version of I2? I'm thinking the increased FSB will more than make up for the 1/3rd cache! And its price point, $1150, is not so insane as typical Itanium stuff. If supermicro were to support the 533Mhz FSB in their motherboards, we might be seeing more Itanium platforms around. Sadly, though, Itanium is stuck with quad-channel DDR200 instead of dual-channel DDR400 or dual-channel DDR2-533 to feed its bus...

 

[Mephistopheles]

On yet another note: the fully fledged MPUs will be given a <b>667Mhz FSB (10.6GB/s) by 1Q05.</b> This is in line with expectations that Montecito will actually feature dual 667Mhz FSBs, meaning a total system bandwidth in excess of 21GB/s, nothing to sneeze at. In any case, the 667Mhz FSB is said to be the magic FSB that the Xeons too will share at some point or another...

The big advances we've been seeing in IA64 are most likely credited, of course, to Opteron's strength. We're now seeing 12x133 = 1600Mhz Itanium 2s and we'll probably be seeing, within the next 6 months, something like maybe 10x166 = 1667Mhz or 11x166 = 1833Mhz Itanium 2s. This last one will probably break the 3000 spec_fp barrier by early 2005.

These beasts are very interesting for me, for instance - I'm a physicist and I'm very interested in very high FP-throughput. I know of some massively parallelizeable computer simulations that could, in theory, benefit from IA64. <P ID="edit"><FONT SIZE=-1><EM>Edited by Mephistopheles on 11/10/04 01:54 PM.</EM></FONT></P>

 

[trooper11]

no one is denying the power of the itanium in hpc or other calculation type work. im just say the price drop alone gains itanium 0 market penetration. this wont suddenly cuase it to be used in workstations and take over where xeons are or where p4's are. now in the areas itanium is on currently, this will make it an excellent option.

one thing that was strange though is Microsoft stating that there would be no IA64 port of its upcoming special HPC OS:

<A HREF="http://www.infoworld.com/article/04/11/09/HNskipitanium_1.html" target="_new">http://www.infoworld.com/article/04/11/09/HNskipitanium_1.html</A>

now one could argue itaniums wouldnt be used in the area the OS is aimed for, but Id argue that the new pricing of the itanium would definitley make it an option for hpc's. but instead, MS says it will only support x86-64, with IA64 coming at a "later, undefined date"

 

[trooper11]

there was something else that came to me that we havent considered.

with th adoption of x86-64 and windows 64bit coming next year, apps will inevitablly be ported to it. where does this leave the itanium? It can emulate 32bit, but not x86-64. will intle just forgo any emulation and just force all to port to IA64 or will they add emulation for x86-64? I know intel doesnt want to do that, maybe they will just ignore it all together.

 

[Mephistopheles]

Yes, I thought about that too... If Intel actually managed to emulate x86-64 <i>faster</i> than they can emulate 32-bit code, then they'd have an advantage for the years to come.

But something tells me that montecito will be a very big thing to come. It is already undergoing testing for a while now, and it represents such a big performance increase that even the emulation will probably get a healthy boost from it. Just consider that Montecito is built for speeds in excess of 2Ghz (estimates put it at 2.5Ghz) and it has the highest IPC of any architecture, and you've got yourself a performance lead...

 

[RichPLS]

your not likely to port software, and not get a hit on performance.

_____________________________________________
<font color=red> And the sign says "You got to have a membership card to get inside" Huh
So I got me a pen and paper And I made up my own little sign </font color=red>

 

[trooper11]

well ok so i should rephrase, not porting to x86-64, but releasing versions of thier apps designed for 64bit

 

[slvr_phoenix]

your not likely to port software, and not get a hit on performance.

Any port in a storm?

<pre><b><font color=red>"Build a man a fire and he's warm for the rest of the evening.
Set a man on fire and he's warm for the rest of his life." - Steve Taylor</font color=red></b></pre><p>

 

[RichPLS]

during that storm and in that port, it is considered a premium to get hits on, and hopefully you are up to performing.

_____________________________________________
<font color=red> And the sign says "You got to have a membership card to get inside" Huh
So I got me a pen and paper And I made up my own little sign </font color=red>

 

[juin]

Everything is inligne of expectation.

The good news but wont change the fact that Power5 have leap frog it.That still the Mackinley in there that is 2 year old.Until Montecito is out the performance delta wont change much with Power5.It may keep the new sparc 2.13 from taking any advantage over IPF.May help HP to convince customer to move to IA-64.A 25% to 100% increase getting interesting.

i need to change useur name.