backtop


Print 35 comment(s) - last by Sid7039.. on Jul 24 at 10:55 AM

Intel gives the Itanium 2 one last hurrah before the QuickPath makeover next year

Intel is preparing to launch the next iteration of its dual-core Itanium 2 processors, Montvale. The next-generation Montvale-based Itanium 2 replaces the current Montecito-based Itanium 2 9000-series. Intel positions the Montvale-based Itanium 2 processors towards mission critical tasks, with scalability from two to 512 processor sockets – for up to 1024 processor cores.

The new Montvale-based processors retain the Itanium 2 name, but with a different processor number. Montvale-based processors will carry the 9100-series processor number to differentiate from the Montecito-based 9000-series. The new Itanium 2 9100-series has minor upgrades over the 9000-series.

New to the Itanium 2 9100-series is Demand Based Switching, or DBS, with Enhanced Intel Speedstep Technology. DBS allows the higher-end Itanium 2 9100-series models to enter a low-power state during idle periods. The feature is only present on the Itanium 2 9140N, 9140M, 9150N and 9150M. Despite DBS being present on the higher-end Itanium 2 9100 models, all dual-core 9100-series processors have 104-watt TDPs.

Intel continues to manufacturer the Itanium 2 9100-series on a 90nm fabrication process, as with Montecito. The biggest upgrade the Itanium 2 9100-series brings to the table is a faster 667 MHz front-side bus on three of the new models. Intel’s previous Itanium 2 9000-series were limited to 533 MHz front-side bus, at maximum.

Intel Dual-Core Itanium 2
Processor
Number
Clock
Speed

FSBiL3
Cache
  Q4'07 Launch
Price

9150M
1.66
667 MHz
24MB$3692

9150N
1.60
533/400 MHz
24MB$3692
9140M
1.66
667 MHz
18MB$1980
9140N
1.60
533/400 MHz
18MB$1980
9130M
1.66
667 MHz
8MB$1552
9120N
1.42
533/400 MHz
12MB$910

Intel offers the Itanium 2 9100-series in five different models, as long as its 1.66 GHz or 1.6 GHz. However, Intel offers the Itanium 2 9100-series in different iL3 cache sizes to designate different model numbers. Three iL3 cache sizes are available – 24MB, 18MB and 8MB. Intel also designates Itanium 2 9100-series processors with an M suffix for 667 MHz front-side bus and N suffix for 533/400 MHz front-side buses.

The ugly duckling of the Itanium 2 9100-series lineup is the 9120N. This model is the only Itanium 2 9100-series without a 1.66/1.60 GHz clock speed. It has a lower 1.42 GHz clock speed with a 533/400 MHz front-side bus. The Itanium 2 9120N is the only dual-core model with a 12MB iL3 cache. Despite the lower clock speed, the Itanium 2 9120N has the same 104-watt TDP as the higher-clocked models.

All dual-core Itanium 2 models, with the exception of the 9130M, can execute up to four threads.

Intel Single-Core Itanium 2
Processor
Number
Clock
Speed

FSBiL3
Cache
  Q4'07 Launch
Price

9110N
1.60
533/400 MHz
12MB$696

Intel also has one single-core Itanium 2 9100-series in the lineup. The Itanium 2 9110N has a 1.60 GHz clock speed and 12MB of iL3 cache. It also has a 533/400 MHz front-side bus. The single-core Itanium 2 has a 75-watt TDP rating. Intel prices the Itanium 2 9110N at $696.

Intel does not have any new chipsets in the pipeline for the Itanium 2 9100-series. The Itanium 2 9100-series will continue to use the Intel 8870 with its DDR-200 memory controller. The Intel E8870 only supports 400 MHz front-side bus processors. Third-party chipsets are required to take advantage of faster 667 MHz and 533 MHz front-side buses.

Intel’s next Itanium, the follow up to Montvale, will not arrive until 2H’08. Tukwila will be a quad-core processor with Intel’s QuickPath, formerly known as common system interface, technology. Tukwila will have a 65nm fabrication process and will drop into a new Boxboro MC platform.

Expect Intel to debut the Itanium 2 9100-series next quarter, after the current Itanium 2 9000-series is EOL in August.



Comments     Threshold


This article is over a month old, voting and posting comments is disabled

I don't get it
By firewolfsm on 7/21/2007 1:56:03 PM , Rating: 1
What makes this processor worth $3600? Is there anything it can do that a Quad Xeon can't? I realize there's a lot of cache but a Quad Penryn has 12MB, which is close.




RE: I don't get it
By Dactyl on 7/21/2007 5:30:03 PM , Rating: 4
Yes, it can be ganged up with 511 of its brothers. You can't do that with a Xeon DP.

Intel positions the Montvale-based Itanium 2 processors towards mission critical tasks, with scalability from two to 512 processor sockets – for up to 1024 processor cores.

I wish I had enough money to buy 500 Itanium CPUs.

I wouldn't actually buy them, I just wish I had that much money.


RE: I don't get it
By HotFoot on 7/21/2007 9:02:20 PM , Rating: 2
But is that it? These chips are specifically designed for server-type workloads and don't carry a lot of the baggage that x86 architecture does. However, I really have to wonder about the price/performance picture even for huge processing requirements. Chances are if you're writing software that can run on 512 processors in one system, you can write software that will distribute that workload amongst a cluster of Xeon or Opteron servers.

I don't recall ever seeing a benchmark comparing Xeon/Opteron/Itanium systems. I only ever find Xeon/Opteron comparisons.


RE: I don't get it
By Lord Evermore on 7/21/2007 11:34:16 PM , Rating: 2
Some reasons: Very very simple design due to things like instruction level parallelism, VLIW, etc. Allows for such small die sizes that they can put all that cache on it. Performance in some cases is far better clock per clock than an Opteron or Xeon (especially NetBurst Xeons), up to six instructions per clock, plus multithreading. Itanium is just a better option for some uses. There's more difference than just choosing between an Athlon X2 or a Core 2 Duo, which is often just based on price. Other requirements and abilities come into play for the Itanium. Floating point is blazing fast, scientific computing makes use of that. For general server workloads though I don't think it's necessarily that much faster than others.

http://en.wikipedia.org/wiki/Itanium

http://www.psds.com/products/servers/benchmark.htm
http://redmondmag.com/news/article.asp?EditorialsI...
http://www.databasejournal.com/features/oracle/art...
http://www.cbronline.com/article_news.asp?guid=B9C...

For datacenters, I assume there is the big factor that a cluster of servers would use more power and produce more heat than a multi-processor Itanium system. I don't know what the breakdown of costs are, maybe it's cheaper to do a multi-processor Itanium than an equivalent multi-processor Opteron or Xeon system, when you hit a certain number of CPUs allowed.


RE: I don't get it
By Viditor on 7/22/2007 1:22:23 PM , Rating: 3
quote:
I assume there is the big factor that a cluster of servers would use more power and produce more heat than a multi-processor Itanium system. I don't know what the breakdown of costs are, maybe it's cheaper to do a multi-processor Itanium than an equivalent multi-processor Opteron or Xeon system, when you hit a certain number of CPUs allowed

It's not actually...
Itanium aren't very good for datacenters, but they are great for scientific niche applications.
For datacenters and supercomputers, Opterons are significantly better and they can utilize far more CPUs using NUMA (Red Storm for example has over 10,000 Dual Core Opterons in it)

Of the Top Ten Supercomputers,
Number 1 is BlueGene using IBM's chips
Numbers 2 and 3 are Crays that use Opterons
The highest rated Intel based machines are number 8 using Woodcrest chips, and Number 10 which uses Itanium 2s

Where Itanium really shines is in the 32 socket scientific system that require very intensive FP (clock for clock, Itanium blows everyone away in FP, though this may get turned around if AMD's predictions on K10 prove true).


RE: I don't get it
By Calin on 7/23/2007 9:16:19 AM , Rating: 2
Itanium has a better FP performance. Its x86-simulation engine was able to run x86 code (very slow), and now I think the x86 code is run with a software interpreter.
Itanium is good only when you can feed it optimised code - and common servers don't get this. However, scientific workload is almost always recompiled for the platform it runs on (and always compiled for the processor it runs on). As such, x86 compatibility is not needed, and the FP performance of Itanium is better than mostly anything else


RE: I don't get it
By Viditor on 7/22/2007 5:44:58 AM , Rating: 3
quote:
I don't recall ever seeing a benchmark comparing Xeon/Opteron/Itanium systems. I only ever find Xeon/Opteron comparisons

The main place you'll find them is at http://www.spec.org/
Most software benchmark suites are for one or the other uA.


RE: I don't get it
By Viditor on 7/22/2007 5:37:59 AM , Rating: 2
quote:
Is there anything it can do that a Quad Xeon can't?

FP calculations...Xeons are very, very poor at FP while Itaniums are even better than the Opterons.


RE: I don't get it
By Shintai on 7/22/2007 5:42:14 AM , Rating: 1
You talking about P4 FPU again fanboy?


RE: I don't get it
By Viditor on 7/22/2007 6:11:36 AM , Rating: 4
quote:
You talking about P4 FPU again

No, troller toad...:)
Go to www.spec.org and run a search on 4 core SPECfp 2006 Rates for Xeon, Itanium 2, and Opteron...


RE: I don't get it
By Shintai on 7/23/2007 4:32:38 AM , Rating: 2
Then you should know specfp_rate is more or less a bandwidth test. Ofcourse all real world reviews kinda ruins that. Since the K8 would then be about 20% faster than Core 2. Yet it aint! And if you look a note on how dualcore to quadcore scales. You would see the same problematic picture. But thats how SYNTHETIC benches are.


RE: I don't get it
By Viditor on 7/23/2007 8:56:03 AM , Rating: 2
quote:
Then you should know specfp_rate is more or less a bandwidth test

It's actually the only test I know of that runs across all 3 platforms specifically in FP (which is why I chose it).
quote:
Since the K8 would then be about 20% faster than Core 2

You mean in FP only, don't you? Or are you trying to somehow equate FP calculations with real-world performance on something like a game machine??


RE: I don't get it
By Shintai on 7/22/07, Rating: 0
RE: I don't get it
By Sid7039 on 7/22/2007 3:19:40 PM , Rating: 2
Please look there, at the first row of the table:
http://www.tpc.org/tpcc/results/tpcc_perf_results....

It's not a good idea to compare x86 with Itanium - it's completely different market segment. Comparing them is like comparing eg. a car with a boat.

Regards.


RE: I don't get it
By colonelclaw on 7/23/2007 8:27:19 AM , Rating: 2
so in that case who would buy an itanium system? what kind of company would be a typical customer who would always go for itanium over opteron/xeon?


RE: I don't get it
By Sid7039 on 7/24/2007 10:55:47 AM , Rating: 2
Often the same corporations which use IBM Power, SUN Sparc and HP PA-RISC processors.
I have an experience with mobile phone company and they use... everything: IBM machines, SUN machines, Itanium linux/windows machines, etc, x86 machines.. The same I saw in a large bank.

Here is (a little old) information, that shows who buy Itanium servers:
http://h41225.www4.hp.com/integrity/fr/fr/customer... (link shows that this informaiton is in French but it is in English)

As you can see at tpc.org benchmark, typical customer is a rich company that needs massive transactional and database power (hundreads of processors, TBs of RAM) or who use Itanium as virtualization platform (buy one big machine and manage it instead of buying 20 less powerful ones)


"We don't know how to make a $500 computer that's not a piece of junk." -- Apple CEO Steve Jobs

Related Articles
Intel Responds to AMD's "Torrenza"
September 27, 2006, 9:30 PM
The Count of "Montecito"
July 16, 2006, 3:39 PM













botimage
Copyright 2014 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki