Sign in with
Sign up | Sign in
Your question

Itanium and x86 convergence

Last response: in CPUs
Share
February 27, 2007 12:43:23 AM

http://www.dailytech.com/article.aspx?newsid=6236

This is great news. Hopefully one day, we will do away with x86 all together. The EPIC arch. can be much better parallelism than x86. Also, on a different note, it looks like we will see CSI this year. Curious to see how it is implemented.
February 27, 2007 12:48:56 AM

CSI will come with Socket B and Socket H in 2008 or later...
But I think x86 will not be displaced in the near future unless there is an architecture that can take advantages of a better instruction set and run x86 well.
February 27, 2007 12:57:46 AM

Cool, Itaniums might become a little more mainstream.

Where do you buy Intaniums anyway?
Related resources
Can't find your answer ? Ask !
February 27, 2007 1:06:23 AM

Quote:
http://www.dailytech.com/article.aspx?newsid=6236

This is great news. Hopefully one day, we will do away with x86 all together. The EPIC arch. can be much better parallelism than x86. Also, on a different note, it looks like we will see CSI this year. Curious to see how it is implemented.


Quote:
When asked how its Core and Itanium architectures will intertwine, Gelsinger responded “The first realization of that is Tukwila [quad-core Itanium] in late 2008, the next step in the product family, where we move to common system architecture elements, as well as full alignment on design tools and process.”
February 28, 2007 8:38:46 AM

Quote:
Make sure you read..... before seletively quoting the text for a point you cannot make. The late 2008 refers to the next Itanium iteration.

However, I would be surprised if we see CSI appear this year, I thought (and it makes more sense) to see it in the Nehalem family.


Baron is right.

http://www.xbitlabs.com/news/cpu/display/20070227190938...
Quote:
Intel said that changes in plans were conditioned by an intention to offer higher performance. Now the company plans to offer Xeon MP platform with quad independent bus instead of CSI in 2007.
February 28, 2007 9:07:25 AM

You think there will ever be hope of a "hybrid x86-64 and IA64 compatible processor"?
February 28, 2007 10:14:37 AM

Quote:
You think there will ever be hope of a "hybrid x86-64 and IA64 compatible processor"?


I think the die space will be wasted...
February 28, 2007 11:43:50 AM

AMD already sank the t-Itanic with AMD64.

Just look at how slow the software/driver adoption of AMD64 and EMT64 has been. Luckily, we can run x86 operating systems drivers and software on it natively, so this isnt the end of the world.

Were Intel to try to push Itanic into the mainstream, we'd have 4-5 years where our systems were running all of their software under an emulator, with all the overheads that entails, and where people were totally unable to find hardware drivers.

Not to mention you'd need to buy a new copy of your OS, as neither Win x86 or Win x64 will run on Itanium.

Yes, x86 is a headache in some ways, and a fully RISC archetecture would be nice, but its too late, and compatibility is too important.
February 28, 2007 12:14:32 PM

Quote:
Were Intel to try to push Itanic into the mainstream, we'd have 4-5 years where our systems were running all of their software under an emulator, with all the overheads that entails, and where people were totally unable to find hardware drivers.


Not entirely true. There are many "just-in-time" compiler technologies in use out there already that GREATLY reduce run-times of emulated code (JAVA is the biggest example, but Mac's use something similar, too). The first time you run a new application on the computer it would essentially reconstruct the executable and make a "native" version of it and then use that native version from that point forward. It's tricky stuff but has been done before and could be used in this case.

The OS issue is a far bigger one, though, and that would require Microsoft to make a native version for Itanium. Unless Intel made it possible to "virtualize" an x86 environment and run the OS on that. That would be interesting...
February 28, 2007 1:15:49 PM

Quote:

However, I would be surprised if we see CSI appear this year, I thought (and it makes more sense) to see it in the Nehalem family.


That is interesting. I had always assumed that Nehalem would be the first with CSI. I guess they decided to pull in CSI and the IMC so they could get something out.
February 28, 2007 1:17:15 PM

Quote:

However, I would be surprised if we see CSI appear this year, I thought (and it makes more sense) to see it in the Nehalem family.


That is interesting. I had always assumed that Nehalem would be the first with CSI. I guess they decided to pull in CSI and the IMC so they could get something out.

Ok, looks like the article was wrong:

Quote:
Update 2/27/2007: The reference we had to Tigerton supporting CSI in 2007 was incorrect and has been removed.
February 28, 2007 1:18:11 PM

Quote:

However, I would be surprised if we see CSI appear this year, I thought (and it makes more sense) to see it in the Nehalem family.


That is interesting. I had always assumed that Nehalem would be the first with CSI. I guess they decided to pull in CSI and the IMC so they could get something out.

Nehalem is due 2008 :wink:
February 28, 2007 5:30:42 PM

Quote:
AMD already sank the t-Itanic with AMD64.

Just look at how slow the software/driver adoption of AMD64 and EMT64 has been. Luckily, we can run x86 operating systems drivers and software on it natively, so this isnt the end of the world.

Were Intel to try to push Itanic into the mainstream, we'd have 4-5 years where our systems were running all of their software under an emulator, with all the overheads that entails, and where people were totally unable to find hardware drivers.

Not to mention you'd need to buy a new copy of your OS, as neither Win x86 or Win x64 will run on Itanium.

Yes, x86 is a headache in some ways, and a fully RISC archetecture would be nice, but its too late, and compatibility is too important.

I think the point of the emulator is to promote the new compiler and the Itamium should be so much faster than the core2 that the emulator will not show any speed slowing. Like what happend with the Mac OS9->OS10.
If this is not done, we may have x86 forever! A scary thought.
February 28, 2007 7:46:10 PM

Quote:
Were Intel to try to push Itanic into the mainstream, we'd have 4-5 years where our systems were running all of their software under an emulator, with all the overheads that entails, and where people were totally unable to find hardware drivers.


Not entirely true. There are many "just-in-time" compiler technologies in use out there already that GREATLY reduce run-times of emulated code (JAVA is the biggest example, but Mac's use something similar, too). The first time you run a new application on the computer it would essentially reconstruct the executable and make a "native" version of it and then use that native version from that point forward. It's tricky stuff but has been done before and could be used in this case.

The OS issue is a far bigger one, though, and that would require Microsoft to make a native version for Itanium. Unless Intel made it possible to "virtualize" an x86 environment and run the OS on that. That would be interesting...

There is a Windows Server 2003 IA64 for Itanium already.

However, the top performing Itanium supposedly has slightly less FP performance than a Core 2 Duo x6800, and worse integer performance than even a decent netburst.

Take away a bit more performance (even if only a mild 20%) for running x86 native code, and why the hell would I want to adopt an Itanium system now for my x86 applications? Its a safe bet most of the rest of the consumer market (and a very large proportion of the business market) feel the same way.

Without it being able to get a foothold, it wont ever get the applications to give it that foothold, its a chicken and egg situation.

CSI is nothing to do with making us all switch to Itanium, it is simply that the curent FSB technology is approaching its limits, and it makes sense to only design one new interface, especially as it means Intel avoid having to design itanium specific boards at large costs for small volumes.
Anonymous
a b à CPUs
March 3, 2007 6:24:13 PM

Well Itanium processors can emulate x86 if you integrate the emulator in the OS you want to use it with. I beleive you can find such an emulator for Windows at the Microsoft website. Also you can find linux x86 emulators at the Intel website. Oh and I beleive XP x64 already has integrated support for Itanium.

It would be nice to see an IA-64 based architecture used for an x86-64 architecture. Itanium's architecture is a really nice one.
March 3, 2007 9:53:42 PM

Windows XP Pro x64 does not support the IA64 mArch. There are two other versions of windows that do, "Windows XP 64-bit Edition for Itanium systems, Version 2002" and "Windows XP 64-bit Edition, Version 2003". Neither of these run on x64 (AMD64 and EMT64) machines.

Yes, you can emulate x86 on IA64 via the full implementation of WOW64, very different from the version of WOW64 we see in XP Pro x64.
It has a high CPU overhead, and while IA64 is an "elegant" mArch, it is not really faster, in fact it is quite a bit slower than the top C2Ds in Integer operations. Couple this with the WOW64 overhead, and it would be hard to convince anyone to switch to IA64 systems until the software was there, just as it would be hard to convince software writers to switch before the users are there.

This chicken and egg scenario will forever prevent IA64 taking hold imho.

IA-64 is nice, mainly for one reason - x86 sucks.

x86 has always sucked, although it has shown this more in the past 10 years or so. The complexity of modern CPUs is in no small part due to the necessity of maintaining x86 compatibility. Throwing x86 out the window and starting again would be nice, but it will never happen.

IA-64 used for x86-64 makes no sense. Either it is IA64 running an emulator (slow) or IA64 with x86 and x64 instruction support, in which case we are basically back to current CPUs, as all modern CPUs can be argued to be RISC CPUs with a CISC/x86 frontend.

Adding x86 instruction support to IA64 would remove all its advantages, and leave most of its disadvantages (poor integer performance compared to Core 2, for example)
March 4, 2007 11:53:52 AM

Quote:
IA-64's speed comes from programming concepts that were totally unheard of in Intel-based PC hardware design when Merced was announced back in 1994. No longer is the responsibility of speeding things up delegated to silicon logic alone. IA-64 now allows, or rather requires, software to convey hardware usage logic directly to the CPU. IA-64 accomplishes this by redefining the instruction format into something called an EPIC design (Explicitly Parallel Instruction Computing) whereby the very nature of instruction encodings tell the CPU which parts of the chip will be used to process data. This has the serious side-effect of relegating instruction ordering, logic unit usage, and optimization techniques directly to the compiler or Assembly programmer's back. This is why you see such a tremendous variation in published benchmarks. In one benchmark it may seem like an 800MHz IA-64 performs worse than a 100MHz Pentium, only to turn around and visit another site where it seems the same 800MHz IA-64 would outperform a 10+GHz Pentium 4, if such a creature existed. This is directly due to the fact that IA-64's instruction scheduling and resource utilization are determined by the compiler at compile time and are not negotiated or altered by the processor at runtime. As many of you have probably guessed, this places a much heftier burden on the compiler than previous Intel CPUs required. Effective utilization of such a dynamic CPU is a monumental task that, I'm sure, will not be fully realized for years. Until that time IA-64 is destined to live out its life falling short of its true potential, which is nothing short of "damned impressive."


The problem is Itanium isn't "Plug and Play", if you will, and unless people are willing to spend the time and money to tame it, it will never be that good.

http://www.geek.com/procspec/features/itanium/
!