Itanium 2

[_N_]

Well, looks like they learnt from mistakes.

Intel Disclosed Itanium2 Details

Today Intel announced that their new development, 64bit Itanium2 processor intended for high-performance server solutions will run nearly twice as fast as its predecessor, the first Itanium CPU. The newcomer is expected to be able to compete with solutions from Sun Microsystems and IBM.

According to the info disclosed by Intel, Itanium2 performance is 1.7-2.6 times faster than that of Itanium 800MHz with 4MB L3 cache. Also according to this data, Itanium2 is about 1.5 times faster than Sun UltraSparc III 1050MHz during transaction of large data packs.

As for the technical details, Itanium2 will feature 3MB L3 cache and will work at 1GHz core clock frequency. The frequency of MacKinley system bus will make 400MHz and its width will be 128bit.

The memory subsystem performance will make from 6.4GB/sec to 12.8GB/sec depending on the chipset. E8870 from Intel will be able to transfer up to 6.4GB/sec (compared with 4.2GB/sec by i460GX), while more progressive zxl analogues from Hewlett-Packard will be able to provide 8.6GB/sec for dual-processor servers and up to 12.8GB/sec for four-way Itanium2 servers.

[Py7h0n]

Yeah this time it just might work ...

[Mr ABIT]

what about amd's hammer?

[_N_]

I think that the hammer is on a different playing field to the Itanium, the Itanium seems to be server only, whereas the hammer is more of a high end workstation/server end, like the current Athlon MP's are now.

[Viper_NZ]

Last I heard they had sold VERY few Itaniums. And going on the time delay between Itanium and Itantium 2 Moores Law sorta works out for the performance increase, even if the clock rate was minor.

Great thing about those chips is that when compilers for them improve, so does peroformance...

[mird-OC]

yeah well the first itanium certainly was a flop, and probably the most embarassing event in intel's history. i'd say this one should fair quite a bit better. i like the sound of that memory subsystem.

but yeah, it's still way out there as far as it ever being an option for us desktop users.

[Mr ABIT]

so due to the hammer not being as "high end" as intels itanium, you think it will have more success.. or will people just stick to regular 32-bit pc's? on the otherside of the coin, u think the 64bit desktop thing will ever really take off? cos it will mean that ALL the software developers would havto redisgn everything to take advantage of the 64bit proc wouldnt it? or have i got it all wrong :D

[swiftynz]

the idea of AMD's hammer is that it will be both 64 and 32 bit. therefore it can run all the old 32 bit applications while taking advantage of new applications with 64-bit instructions aswell.
i reckon this chip will boost AMD's marketshare in the high-end pc market quite considerably!

[_N_]

well, I think that the hammer will replace the athlon, so what I mean by that is that if you look under systems specs thread, in a years time, more than half the people who own athlons now will own a hammer.

The hammer will become the next top of the range desktop CPU, and the duron will get replaced by the athlon and Intel won't be able to compete unless they do something drastic.

There is need for 64bit appz. It will mainly help servers, but the desktop users will get something out of it too,.. eventaully, it will take Microsoft 5 years to make the move to 64bit.

[Roman]

Its a shame Intel discontinued their Pentium III Tualatin series

very powerful chips, they would be real thorn in the P4's side if Intel hadn't killed them...

[KingJackal]

Originally posted by Mr ABIT
so due to the hammer not being as "high end" as intels itanium, you think it will have more success.. or will people just stick to regular 32-bit pc's? on the otherside of the coin, u think the 64bit desktop thing will ever really take off? cos it will mean that ALL the software developers would havto redisgn everything to take advantage of the 64bit proc wouldnt it? or have i got it all wrong :D

You can't address more than 4GB of memory on a 32-bit processor. Which sucks.

4GB ain't a lot - that's only 2 or 3 years away for most of us ( bear in mind that's physical RAM + virtual memory size ).

AMD's Hammer does mean they can run current x86 code. But their 64bit instruction set also shares a lot of the same weaknesses as x86, cos it's based on x86. With Itanium, Intel had a clean slate, so they could design an ISA tweaked for future apps - so an Itanium SHOULD ( compilers willing ;) ) be able to drag off a Hammer if they both use optimised apps....

But then, at the prices the Hammer will be selling for, you'd certainely hope so! :D

[Solid Snake]

I think Athlons will stick around for a wee bit. Durons now are being phased out, but they will be around on people's motherboards for a wee while yet.

[utopian201]

isnt it kinda like a definite?
like from 16 -> 32 -> 64 bit computing, all it takes is time....??

[Deviant]

Itanium 2 should be almost 2 times faster, cause it will have double the clock speed. I read somewhere that the benchmarks Intel used work were designed for it, and designed for the cache size. Until someone else test it, I'm not going to believe al the Intel propaganda.

Also Opteron will be going against Itanium2, Athlon XP+(Sledgehammer) will go against P4, Athlon XP (probably renamed to Duron) up against Celeron.

PIII core had to go as it was at the end of it's era, and Intel couldn't get the clock speed up any more. Athlon XP would have kill it and made it look bad, but it was a great chip in it's day.

[mird-OC]

Originally posted by Deviant
PIII core had to go as it was at the end of it's era, and Intel couldn't get the clock speed up any more. Athlon XP would have kill it and made it look bad, but it was a great chip in it's day.
it seems the P3 had a lot more life in it than Intel ever expected, reaching much higher clock speeds than it was ever really designed for. the only thing letting it down was the memory bus... other than that it could compete with Athlons very easily, performing as good or better clock for clock.

[GriffiN]

Intel is counting on the new processor to vault its products into higher-priced servers and take market share from Sun Microsystems.
By Aaron Ricadela - June 14, 2002



Four-way servers that include Intel's next-generation, 64-bit Itanium 2 processor will sell for about $41,000, Intel says. It's an early indicator of Itanium 2 pricing--still undisclosed--and demonstrates how Intel is counting on the new processor to vault its products into higher-priced servers and take market share from Sun Microsystems. Itanium 2 deliveries are expected in July, Intel sources say.
The Itanium 2 server price estimate, disclosed at a technical presentation at Intel headquarters in Santa Clara, Calif., Friday, is for a four-processor system running at 1 GHz, with 32 Gbytes of RAM, and a 3-Mbyte on-die memory cache. Intel released the performance specs for Itanium 2 late last month.

The average selling price of Intel-based servers is much lower. Market researcher International Data Corp. says Intel-based systems account for 88.4% of the unit volume of servers delivered to customers. Most of those servers feature 32-bit Xeon chips, and most Xeon-based systems sell for less than $25,000, Intel says.

The chipmaker is in the midst of a product transition to 64-bit processors, which deliver better transactional performance through a combination of more complex computational ability, the ability to store a larger amount of data near the CPU, and the ability to address much more memory than 32-bit parts. Early versions of IBM DB2 Universal Database, Microsoft SQL Server, Oracle 9i, and other software are available on Itanium, the second product in Intel's 64-bit line. Customers, including Pacific Northwest National Laboratory, the National Center for Supercomputing Applications, Daimler Chrysler, and Reuters, are testing Itanium 2 systems, Intel says.

But 32-bit chips continue to supply the bulk of Intel's sales. Improvements in the Xeon chip are also on tap: It's expected to run faster than 3 GHz next year, compared with a top clock speed today of 2.4 GHz.

--------------------------------------------------------------------------------------
Thats the whole article but here's the link anyway: http://www.informationweek.com/story/IWK20020614S0022


Does anyone know what the price of the first Itaniums' went for?

[Tiggerz]

Originally posted by KingJackal


You can't address more than 4GB of memory on a 32-bit processor. Which sucks.

4GB ain't a lot - that's only 2 or 3 years away for most of us ( bear in mind that's physical RAM + virtual memory size ).

AMD's Hammer does mean they can run current x86 code. But their 64bit instruction set also shares a lot of the same weaknesses as x86, cos it's based on x86. With Itanium, Intel had a clean slate, so they could design an ISA tweaked for future apps - so an Itanium SHOULD ( compilers willing ;) ) be able to drag off a Hammer if they both use optimised apps....

But then, at the prices the Hammer will be selling for, you'd certainely hope so! :D

Umm. yes you can address more than 4 gig on a 32 bit processor. Just not all at the same time..

Itanium runs x86 code.

above press release for itanium 2 was made about 3 weeks ago.. programmers manual came out at the same time.

With itanium one thing to remember is that the actual clock speed as very little to do with the performance of the cpu..

Also, itanium will appear in desktop machines.. In fact does appear in desktop machines.. Although, you pay a lot for it.. Its intended audience is workstation and server machines.

[KingJackal]

Originally posted by Tiggerz
Umm. yes you can address more than 4 gig on a 32 bit processor. Just not all at the same time..


Just let me be simple so they can understand, aiiiggghhht?! :mad:

:p

OK, so segmentation means we can virtually address as much as we want - but that aside, we can't currently use any atomic data structure larger than 2^(address bits). And short of any hardware segmentation implementation ( which I don't see ), none of that is happening as far as addressing more PHYSICAL memory is concerned.

OK - so I see how segmentation could theoretically could allow us the maximum number of pages for not only methods and singlar data structures, but also parts of data structures like arrays - but I DON'T see that implemented in anything I'm aware of.

....having said that, I'm not in the industry, so I could be missing a few things.... :rolleyes: :(

[mird-OC]

i remember programming back in the days of using 16-bit programming languages. yeah sure you could address more than 64KB of memory but it sure was a pain in the ass, especially when dealing with data structures that were always going to be larger than 64KB. the same problem would apply with 32-bit programming languages and data structures bigger than 4GB, but in most cases that's probably not gonna happen too often :)