BenchMarks of Itanium on x86

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
G

Guest

Guest
.... Fugger,

.... Here's the URL and/or link if I can get it to work, haven't done a link in Tom's in years, see if I remember how.

.... I thought images and links were banned years ago here, but I saw one yesterday so it must just be images that were banned.

.. Link:

......... <a href=http://athena.tweakers.net/reviews.dsp?Document=191>iTanic ia32 Benchmarks</a>

.. URL:

.. http://athena.tweakers.net/reviews.dsp?Document=191

.... Not sure why intel bothered wiring ia32 capability onto the core/die of the Merced-iTanic 667 if its only going to perform like a P 75 on 32bit Windows apps. It would have been cheaper, simpler, and easier to just add a Celeron 600 or Duron 800 riser card to the machine using the iTanic CPU if and only if x86-32 capability was needed and it would have provided much better performance.

.... iTanic's other 'native' tongue is the new, proprietary ia64 which is incompatible with ia32, WindowsXX(XX), x86-32, x86-16, x86-64, Alpha's Tru64, IBM's UNIX, etc.... or any other existing code unless its ported over to it so the iTanic can slowly run it.

.... By the time the 1997 Merced-iTanium is finally launched in 2001, or whenever, if ever, there may be ia64 code apps. waiting for it. Or maybe everyone will have tired of waiting by then and simply gone with an AMD 'Hammer that runs extremely fast on all new and existing x86-64-32-16bit code that we are already using.

.... Transmeta prolly could have made a faster ia64 & ia32 CPU by now.

............................................................................................................. -Buzz+
 
G

Guest

Guest
1) look for the itanium on intel's site... you won't find it with their server products... not desktops. 2) look at their product road map for 64bit chips.

http://developer.intel.com/design/ia-64/IA64Roadmap/sld029.htm

it doesn't have the text i was looking for, but i remember reading it and makes sense from the diagram that the deerfield processor will be the first intel 64bit processor that is targeted directly at the "home" user. that's a long way off to deem ia-64 as a failure for desktop processors even if amd comes out with a successful one before them. 3) there is no such thing as a 128bit cpu especially in a desktop. i don't know what apple is trying to pull with that one. i imagine they're doing something like the p4 having a 400mhz fsb. from an article about the sledgehammer--

"Perhaps when architects in 2025 begin to debate the move to 128-bit computing, they won't be so quick to reject extensions to the x86-64 features AMD laid out today"

buzz... thanx for the informative info. not so clear on it though. ia-64 is different than the other 64bit systems out there, but isn't alpha different than sun's version? also linux and unix are already to go on the itanium i believe. so would the existing products for these work automatically or would they need to be tweeked and recompiled. either way though seems like there's been plenty of time for both and not much of a hassle?

in the end the hammer may be a better buy as far as price/performance goes as well as well roundedness (for the time being), but i'm not holding my breath for it either. there's not a whole lot of info on it at the moment. i'm not saying the itanium will work out but i agree with slvr_phoenix amd fat_chucky- amd doesn't seem to be looking too far ahead. what happens in a couple years when the advantages of epic and vlims are exploited with an ironed out processor. don't know if it will work but glad to see that they are looking for new ways to increase performance-- now or in the future. maybe amd will be more on the ball than intel (hard not to be... wasn't the itanic due out over a year ago?... at 800mhz and behond.) but then amd has had it's own problems getting stuff out on time. anyway sorry to prolong this it's kind of funny debating a processor that isn't due out for at least a month and isn't a pc anyway, against a processor that won't be out for, at best, almost a year. by that time won't the mckinley be almost out or maybe even madison or deerfield? shouldn't we debate mickinley vs. the hammer? i hear it's gonna kick some @$$.
 

FUGGER

Distinguished
Dec 31, 2007
2,490
0
19,780
My god, most of you people are so clueless.

First off, Itanium is not a desiel engine. dont waste your time with 32bit apps or OS.

There is alot of 64bit applications waiting for itanium already. just because you dont use them doesnt mean they are not ready.

95% chance that none of us will use a itanium in the next 2 years. 100% chance that no one needs itanium at home (that includes you seti@home and quake players). If you buy your itanium for 32 bit applications you are a fool!

Refrain from specultaion of the hammer vs Itanuim untill both are on the market. If its true that hammer is just smp on a chip 2x32, I wouldnt bet on AMD winning the 64bit race.
 

AmdMELTDOWN

Distinguished
Dec 31, 2007
2,000
0
19,780
"Theoretically it should run 2 times faster"

OMFL!

hehe, theoretically you should be an intelligent person Antipop,...theoretically of course!.

;-)

- Amd Helpdesk -
 
G

Guest

Guest
I had a responsibility (and privilege) to port my
company's server software to Itanium.

That's why I'm one of the few who really worked on
the real Itanium boxes.

I've got some lectures from Intel on this matter as well.

One of the best clauses in Intel "IA64" book is -
"if our idea would be only to create 64-bit CPU,
it would be done few years by now by adding
64-bit registers and new address modes." // this is not a quote

They also said that an idea to start
"with a clean sheet of paper" without those x86-legacy
weights on their legs brought another thing -
yes, Intel could design yet another RISC CPU as we
know them.

But according to Intel, the gain,
although measurable, was not high enough.

So they came up with something completely new.

It would be a mistake to evaluate all CPUs in
terms of Quake3 frames-per-second.

After all, some people buy Sun servers - for some reasons :)

The reason my company wants Itanium is VERY simple -
we have to handle tens of GB of data,
and having only some 2-3 GB of address space
doesn't help in it.

So next time you'll beef up some 3+ GB of RAM
into your box and you will consider an upgrade :) -
beware - 32-bit CPUs as P3, P4 and Athlon can
handle only up to 4 GB, and OSes likely will
limit this range even more to 2 GB, sometims - 3 GB.
 

smn198

Distinguished
Dec 31, 2007
179
0
18,680
Whistler, windows 2002 or what ever the MS marketing people decide to call it supports the Itanium chip now in it's current beta

<A HREF="http://www.anandtech.com/mysystemrig.html?id=1686" target="_new">System spec.</A> Ideas appreciated.
 

Tormented

Distinguished
Dec 31, 2007
263
0
18,780
Reminds me of Pentium Pro, and it was used in servers
it runs 16 bit code badly but flyies on 32 bit!
So Itanium will be as Pentium Pro was.

<b>-----------------------</b>
-<font color=red><b>R.K.</b></font color=red>
 

8procstooslow

Distinguished
Dec 31, 2007
40
0
18,530
32 bit performance on 64bit chips is very important in the corporate database market. I work for the largest ERP vendor out there (not talking oracle), with many customers migrating to 64bit versions of our software in order to take advantage of 64bit addressing (i.e. over 4gb of memory).
Most of these customer only have a 4 day downtime window a year. You cannot migrate large database 500GB-1TB to a new hardware plattform and upgrade the software in a weekend.
So generally you change the hardware one weekend, let it bed itself in and then upgrade the software at a later point.
If I had to do that with the latest intel IA-64 bit chip I would be shot if the software ran 10 time slower.
Also on high end servers you run more than one application (i.e. interfaces, data migration tools, etc) and not all of these will be 64bit when you migrate your main application.
Also 64bit will be important for PC users in the future (look at the new games consoles, they all use 64bit processors), just as the 32bit Ppro ended up on our desktops in the form of the PII. Will you want your favorite game running 10 times slower?
 

SoulReaper

Distinguished
Dec 31, 2007
194
0
18,680
Another GREAT product by the GREAT intel....NOT!!! See this is why i stick with AMD, cause they atleast put out products that work right.

--SR
 
G

Guest

Guest
.... Yes, the PentiumPro was only a little slower in 16bit than 32bit, but iTanic, although hardwired for both ia64 and ia32, appears to be running ia32 much much slower, which again begs the question why even bother to include ia32 capability on die? Itanic machines would be very high priced and adding a ia32 riser card with the latest ia32 CPU wouldn't boost the price much percentagewise.

.... Actually, iTanic isn't considered a clean paper fresh design but simply another rehack of a design intel attempted 11 years ago and failed at. Compaq-DEC-Alpha is laughing at intel for attempting to resurrect that failed non-OOO concept once again. For full details of that earlier failed effort and hype and how it relates to iTanic's design see www.realworldtech.com/insider/i860-ia64-1.cfm

.... BTW, how do you put links into this BBS now?
 
G

Guest

Guest
The thing is Intel have made a brave choice to scrap almost all backward compatability in the Itanium(no matter if this is right or wrong), so it can't be any kind of comparison unless you run software that is compiled for it's infastructure.

one of the first UK T-Bird users....
 

slvr_phoenix

Splendid
Dec 31, 2007
6,223
1
25,780
"32 bit performance on 64bit chips is very important in the corporate database market. I work for the largest ERP vendor out there (not talking oracle), with many customers migrating to 64bit versions of our software in order to take advantage of 64bit addressing (i.e. over 4gb of memory)."

I'm not so sure how much sense this makes. I mean why not just get 64bit software? I try my best not to ever run 16bit applications on my 32 bit computer. So why would other people want to run 32bit applications on a 64 bit server?

"Most of these customer only have a 4 day downtime window a year. You cannot migrate large database 500GB-1TB to a new hardware plattform and upgrade the software in a weekend.
So generally you change the hardware one weekend, let it bed itself in and then upgrade the software at a later point.
If I had to do that with the latest intel IA-64 bit chip I would be shot if the software ran 10 time slower."

According to a friend of mine who has been a network admin for decades and done just such types of things, he says only an idiot would try to do it that way. They way he's always done it is to simply leave the old server running for the users while you get the new server running without user access. Once the new server is all up and running fine, you then take a weekend of downtime to remove access from the old server and switch it to the new server. And low and behold, in usually just a matter of hours the users can go from using the old servers to the new servers with a seamless integration. Then you just remove the old servers. And to do it any other way is only asking for problems.

"Also on high end servers you run more than one application (i.e. interfaces, data migration tools, etc) and not all of these will be 64bit when you migrate your main application."
But they really should be. It doesn't take much to recompile for a higher bit platform. So if you can't get all of your software as 64 bit, then you really need to consider consulting the software vendor and find out just why their product hasn't been upgraded.

"Also 64bit will be important for PC users in the future (look at the new games consoles, they all use 64bit processors), just as the 32bit Ppro ended up on our desktops in the form of the PII. Will you want your favorite game running 10 times slower?"

We are only talking SERVERS at the moment. Desktops won't be pushed to go to 64 bit for a few years still. And if they run 32 bit applications ten times slower two years from now, then someone has really miserably failed at chip design. But to reitterate the point: THE ITANIUM IS FOR SERVERS ONLY. So to argue about it as though it were meant for anything else only shows how you really have no valid points to argue against it with.

- Sanity is purely based on point-of-view.
 
G

Guest

Guest
>>.... Yes, the PentiumPro was only a little slower in 16bit than 32bit, but iTanic, although hardwired for both ia64 and ia32, appears to be running ia32 much much slower,

That's why I used Itanium's IA-32 support only for things like Winzip :)

Do you think otherwise?
Let's explain...

I had an ample experience with Alpha, MIPS, Sun, i860, PA7100 and some now-unheard-of breeds.

My opinion: an ability to run IA-32 code is a HUGE blessing!
I learnt it on the Alpha, which has been able to run normal
regular applications for Windows, such as WinZip and even
16-bit Windows applications.

So there are some applications
(let's name them legacy applications) who are very
useful to use, such as utilities, drawing apps, etc.

Many of them are unsupported for years or somehow there
is no hope of a port in any way - ever, forever.

Meanwhile it would take a serious dose of insanity to
use such a "grace" of legacy / cross-platform support
in any REAL PRODUCTION system where performance is of
ANY importance.

As of making fun of i860...
You are probably never seen it, aren't you?

In 1991 I had a chance to use it for gas / fluid dynamics
computation, and in floating point it was light years
ahead of, say, i386+i387 or i386+Cyrix.

More - there was a talk of PC industry going to i860
(as Apple did with Motorola 68K->PowerPC transition),
but Compaq issued i386 servers, and this chance was
gone as industry chose path of least resistance.

Later on I had an experience with high-availability
server based on i860.

So it was not a failure, it was an offshoot :)
Later Intel invented very pure i960 RISC chip -
it can be found in many network cards, modems, etc.

So now one company (Intel) tries to make one huge leap
for a humankind with Itanium, while another (AMD) keep
beating this old horsie of 8-bit 8086 to the 64-bit heights
(path of least resistance).
Intel might fail, but they might prevail.
I honor their bold move.
BTW remember Apple with 68K->PowerPC dare move.

So who is trying to innovate and who is not?
 

8procstooslow

Distinguished
Dec 31, 2007
40
0
18,530
"According to a friend of mine who has been a network admin for decades and done just such types of things, he says only an idiot would try to do it that way. They way he's always done it is to simply leave the old server running for the users while you get the new server running without user access. Once the new server is all up and running fine, you then take a weekend of downtime to remove access from the old server and switch it to the new server. And low and behold, in usually just a matter of hours the users can go from using the old servers to the new servers with a seamless integration. Then you just remove the old servers. And to do it any other way is only asking for problems."

Unfortunatly you cannot do this with online databases with thousands of interrelated transactions per day, you have to consider data integrity. Example. User enters an order for a customer, during that week the product is delivered to the customer (product release order is created and delivery confirmation entered) at some point an invoice will be created in the system. If you suddenly switched to a new server that has been off line for a week and ran an invoice creation run the system would not know about that particular order. Most large companies keep orders open for up to a year (talking business to business orders here) so potentially you would have to run the two systems in parallel for up to a year to do what your friend says.
a) the users will not be happy having to look at two systems
b) the admistration overhead is a nightmare
c) the hardware costs are high (two sets of 1TB disk subsystems?????).
Obviously the new hardware will be tested before with the application concerned, with a limited size database.
Also remember in some cases an upgrade to the software has to be done from 32bit to 64bit and that can take a while for a 1TB+ database (data conversions, etc). In the corporate world systems are not upgraded for years and software houses will not recompile/rewite an old version to run on 64bit but will force you to upgrade to the latest version. These upgrade projects cand take months if not years and cost millions of dollers depending on the size of the organization (30000+ online users)

In terms of intefaces and other software being recompiled to 64bit. This is not always possible. The companies may have gone bust or no longer support that operating system.
 

mpjesse

Splendid
According to benchmarks, the Itanium is very efficient in it's use of IA-64 compiled programs. But, the only benchmarks i've seen are comparisons to 32bit chips. I wish someone would post a benchmark comparing the Itanium to an Alpha or SPARC.

-MP JEsse