Sign in with
Sign up | Sign in
Your question

Opteron vs. Itanic

Last response: in CPUs
April 22, 2003 10:16:13 PM

Out of all the articles I have read over the years this one I think is most disappointing. THG, please answer my questions.

Where is the real comparison? Even if I liked AMD products I certainly wouldn't buy Opteron for a regular PC. It's a server processor...why is it being compared to Workstation and Desktop Processors? Ok, why is Itanium2 briefly mentioned and not added as a member of the comparison set?

"This results in a total bandwidth of 6.4 GB/s per Hypertransport port. By comparison, the Pentium 4 with 533 MHz FSB allows a maximum data throughput of 3.97 GB/s - but not in both directions simultaneously."
Is this Theoretical Max. vs. Performance Max. I am not sure what the conclusion is here. Please explain which Intel chipset Maxs at 3.97 GB/s between Processor and MCH.

"64 Bit: For Whom, and When?" Yet again, why not tell the readers the Itanium2 System Memory capacity?

"Test setup and details:"
What memory are you using with what chipset/motherboard. Did you run DDR400 with the i875? etc...

The information provided in the article is excellent but I feel it is misplaced. This is an asynconous comparison. Please write an article comparing and contrasting the Opteron vs. Itanium2 systems.

Opinion of Opteron: POS but has some surprises. It's a fat bloated Athlon. Double the bandwidth, Double the cache, Double the Integer Units, same FPUs, stole SSE-2 and doubled it. HyperTransport is refreshing. One up for AMD. Still .13 process? That must REALLY cook. You should get a mini-wok...make some stir-fry. Seriously, what is the heat like?

Please reply soon.


The Power of MAR Compels Us All!

More about : opteron itanic

April 22, 2003 10:29:30 PM

First of all, i really doubt you understand the big picture of what makes opteron so important and how it differs from an athlon, or even an understanding of how server processors are handled and how they work, especially if you are a home consumer

As for the server processor comment, i really doubt Intel and other server processor manufactures would give tom their thousand dollar processor to benchmark, what if they get damaged, THG has to pay for it right? I dont think they want to risk the $1,400 fine they will get for damaging an itanium2 or ultrasparcIIIi. And if tom cant get the benchies, where is he going to get them from, the processor manufacturers themselves. Do you really trust the manufacturers to give accurate numbers on their own products? Plus Opteron isn't in direct competition with just Itanium2, it is in competition with just about all server processors, include xeon, ultrasparc, Power4+ and the like
April 22, 2003 11:41:18 PM

Newsys seemed to trust Tom with $760 Opterons just fine. Aceshardware had Ultraspark IIIi servers in the benchmarks. If Intel wants to show off their Itanium2, there is no better way to do it. $4000 is pocket money for them compared to the PR they could get.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
Related resources
April 23, 2003 12:54:16 AM

I've done my share of Itanium/Win64 ports over the past two years. I am waiting for MS to ship us a Beta Win64 for Opteron which I will be hitting next much to my displeasure.

HyperTransport is a very GOOD feature for the Opteron. I would expect the write-back/write-through performance is great. Here's a question "Mr. AMD Wizard" how well does it handle thread migration? How about cache coherancy?

My point in my posting is that it would be nice to have a synchronous testing stratagy rather than compare two different classes of processors. Since THG USUALLY only focus on AMD and Intel processors; UltraSPARC, PA-RISC, Alpha, PowerPC, MIPS (all 64-bit) really wouldn't be seen here. Opteron is targetted to compete in the 64-bit processor platform market not 32-bit.

It is obvious that you do not UNDERSTAND Intel's product structure. Celeron for economy, P4 for desktops, Xeon for workstations, and Itanium2 for servers. The Xeon is far from a server processor. I could turn my Pentium 233Mhz system into a DNS server but it doesn't make the P54C a server processor. The Xeon is a workstation processor. A number of vendors use it for servers but that was not Intel's design intent. AMD's product structure is based on what they can borrow next from anyone else. Break into the 64-bit market because no one is buying the AthlonMP for servers and they are one of the only ones on the block without a 64-bit product. I'd like to see Opteron's 64-bit server market share in a year and laugh when it is 1%.

THG can get their hands on an Itanium 2 system if they wanted to. "Excute me neighbor, can I borrow a cup of Itanium?"

The Power of MAR Compels Us All!
April 23, 2003 1:12:21 AM

In thread migration you mean the ability to migrate a thread from one virtual machine to another. As it is a virtual task that is really processor independent, I can't imagine any problems. As for cache coherency, I believe it has MOESI built into its crossbar memory architecture

Dichromatic for your viewing plesure...
April 23, 2003 1:21:46 AM

1% of 64-bit marketshare isnt that bad considering 64bit marketshare equals about 10-20 percent total server/mainframes
April 23, 2003 2:23:53 AM

I´d have liked to see some more Itanium 2 performance numbers too... After all, it´s Intel´s server solution, as egbet put it. The only number I´ve seen is <A HREF="" target="_new">the one at Aceshardware</A>, in which the Itanium 2 1Ghz is more than three times faster than the Opteron. Of course, then again, Itanium 2 is a hell of a lot more expensive, but I´d have liked to see it side-to-side with Opteron... It wouldn´t have had an easy time.

What´s the point of THG´s review if there are no Itanium benchmarks to compare to, after all?

Maybe we can conclude something in the workstation market - after all, the Opteron will be offered as workstation alternative too - but in this field, the Xeons are still up for a performance fight. As servers, the Xeons trail Opteron big time - but where´s Itanium??? It´s supposed to be Intel´s server processor!!!

Very frustrating... :frown:
April 23, 2003 2:30:22 AM

Hrm I wonder how the submissions are handled for the pov benchmark though. it seems there is another score faster than the Itanium which is intended to be a joke, so thers no real authentification system setup to filter joke submissions?
April 23, 2003 2:39:03 AM

That´s strange indeed...

All the more reason to see a trustworthy source of benchmarks for the Itanium 2... Is there a way to actually find out if the number Aceshardware used is intended to be a joke?

Why couldn´t there be an enlightened soul around with an Itanium 2 CPU and who is willing to benchmark it? Can´t be that hard! Couldn´t THG get one if they wanted to?...
April 23, 2003 2:42:02 AM

There isn't all that much softwware that runs on Itanium2 that most people care about. How many server benchmarks have we seen till today? It's only because Opteron came out and there's so much hype about it, that these review sites finally decided to do server benchmarks.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
April 23, 2003 2:45:34 AM

Do they even make single cpu Itanium2 platforms?
April 23, 2003 4:17:49 AM

Yes. HP has several models with 1 Itanium2 processor. They're meant for workstations, but I don't think many people use them that way because of the specialized software needed.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
April 23, 2003 6:30:57 AM

As far as 64-bit conversions are concerned the ones I've heard about so far (CA Ingres, IBM DB2, Epic Unreal) have been straightforward and quite painless so I wouldn't be wincing just yet.

As far as my understanding of cache coherancy is concerned isn't that an issue only encountered when processors share cache? Opteron's don't share any write cache elements so this shouldn't be an issue. I believe it you want up to date information on thread migration the AMD documentation does a good job describing that.

I think you have failed to understand where AMD is targeting this processor though. Opteron is targeted squarely at Xeon and Itanic solutions. Xeon on performance/price and Itanic on price/performance/compatability.

I think your last comment is blatantly false. Xeon is a server processor, if it's not please explain why there are so many of these processors sitting in my works datacenter running 2 and 4 way systems. Not Intel's intent? wtf? Then why have they had Pentium 4 and Xeon systems? Xeons are server systems no doubt about it.

With Microsoft and linux operating systems lined up to be ready from now to Q3 (MS) this year and RDMSes like DB2 (now) Ingress (real soon) and MSSQL and Oracle Q3 they will get the market share they want. Who's going to buy Xeon when opteron is faster, produced by reliable manufacturors (IBM at least) and will run 64-bit apps that people want to run in the real world (RDBMS) at a price/performance point that makes Itanium gag not only on machine cost but also on development cost? You? Ok then, hand your pink slip at the door when they get more sensible people in to replace you.
April 23, 2003 6:51:16 AM

I am starting to become dubious of this result. I have no idea what they put into POV version 3.5, but for the most part ray-tracing is just solving a bunch of linear/quadratic equations, usually in a stack based recursive manor, then throwing the results through a lighting and texture equations. Although it is very floating-point intensive, it has a fair amount of integer. (looking up textures, comparing results, maintaining geometric hierarchies, etc) The Xeon, Opteron, and Itanium vary in SPECfp2000 by no more than 20% and the Xeon and Opteron stay within each other by 10% on both SPECint and SPECfp. In all the rendering benchmarks, the high-end systems varied by less than 20%, often coming within 5% of each other. Then all of a sudden the Itanium II comes along, and bests everyone by 350%, when in other relative benchmarks it comes in at most 20% more and in fact falls short in integer by 40%. Maybe if the Itanium II ran twice as fast in other benchmarks and wiped everyone a$$ in SPECint and SPECfp, I wouldn't question this result, but the clues point conversely.

Dichromatic for your viewing plesure...
April 23, 2003 7:07:12 AM

The industry needs a economic 64-bit processor. Intel sure isn't goin' to give us that. But, in a year or so I think we'll see the price war will have warmed up.

Any rumors of Intel moving to 64-bit in the Pentium line?

The Power of MAR Compels Us All!
April 23, 2003 7:14:18 AM

The repeated rumer that keeps coming up is intel already have an X86-64 CPU nearly ready to go. No onw knows if it is true though.

'It's only when you look at an ant through a magnifying glass on a sunny day that you realise how often they burst into flames'
April 23, 2003 8:56:49 AM

Well, keep in mind that in benchmarks such as SpecInt and SpecFP, there's only a stream of FP or Int instructions. A program that has a well-mix of both integer and FP instructions along with quite a bit of load and store instructions could take advantage of Itanium2's wide parallel execution resources more so than either Opteron or the Xeon, which are limited to 3-way decode and even then, x86 code is difficult to extract parallelism out of vs IA-64. Also factor in the availability of FMAC instructions (which, according to the description you've given, could be used quite a bit in POV). This sounds like exactly the type of situation in which the Itanium architecture would shine.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.