Slashdot Mirror
SGI to Scale Linux Across 1024 CPUs
im333mfg writes "ComputerWorld has an article up about an upcoming SGI Machine, being built for the National Center for Supercomputing Applications, "that will run a single Linux operating system image across 1,024 Intel Corp. Itanium 2 processors and 3TB of shared memory.""
How about a beowulf cluster of these?
I didn't know Longhorn was far enough along to get a test machine up and running.
so does this mean KDE and Openoffice will finally run at decent speed?
Join Team Mozilla #38050 Folding@home
... you insensitive clods
Please, the parent is Funny not Offtopic...
If I look at your comments' moderation... nothing to be proud of!
I wouldn't trust their data. I am just finishing up a project for a Pocket PC, which happens to run Windows CE. I use Microsoft Embedded Visual C++ as the compiler for the project, and at one point I was looking into using C++ style exceptions to make my code a bit cleaner. In the help file included with Embedded Visual C++, Microsoft claims that C++ syle exceptions are supported, and they give examples of how to do it. These claims are also available online here , and where they explain how to use C++ or structured style exception handling, they recommend using C++ style exceptions for portability. However, when I tried to compile my application, I first got the warning that I needed to compile with a flag to enable the stack unwinding semantics. So I did that, and tried to recompile, but this time it failed in the linking stage with an error like "error LNK2001: unresolved external symbol "const type_info::`vftable'" (??_7type_info@@6B@)". After hours and hours of trying to figure out if I had missed a command line option for the linker or forgot to #include something, I stumbled across an online discussion where someone else was having the same problems. Someone else responded with an answer, pointing to a Knowledge Base article, which said that C++ style exceptions are NOT supported in Embedded Visual C++, and that the functionality "is by design."
As you may guess, this is very frustrating, to be told in all documentation that something is supported, and to have the compiler act like it's supported, but for it to not be supported. It also goes to show that you cannot trust those bastards over at Microsoft when they say that something is supported. Oh, yeah, and you may notice that the invalid documentation has been up since 2000, and the KB article was put up in 2003. This means that the invalid documentation has been up for 4 years now, and only a year ago do they admit that it's not supported (even though it's STILL in the documentation for the product that it IS supported).
I am clearly fatter than you.
Dude, 2002 just called and they want their joke back!
:)
Seriously though, your post seems to represent some pretty common netizen perceptions which could use a bit of clearing up. So here goes!
If you take a look, you'll see that there are about 20,600 itanium 2 CPUs just in machines on the latest top500 list, and that's not including vendor machines. (see http://www.top500.org/list/2004/06/ )
If that doesn't sound like much, let's put it in perspective:
Opteron: ~17,300
POWER4: >25,000 (got tired adding!)
Alpha: (bless its soul) 18,528
The most common response to figures like this is:
"Oh, you can't compare those numbers yet because Opteron's haven't been shipping for very long!"
Well, the fact is, we can. From the CPU clock speeds, we can work out that 92% of the 20,600 figure above are for "Madison" (1.3GHz+) Itanium 2 CPUs, which started shipping after Opterons did. (In contrast, there are only a few systems using the newest, 2.2GHz Opteron CPUs) Taking this into account (i.e. counting only machines manufactured in the last year), Itanium 2 has clearly outsold Opteron (and is selling in similar quantities to POWER4+) in the high-end (top500) space. Of course, it looks like Opteron is on a steeper take-up curve, but that's only to be expected given the CPUs generally go for 1/3 to 1/8th of the price of an Itanium 2.
All this is in contrast to what junk rags like The Inquirer ( http://www.theinquirer.net ) will have you believe: for example, they typically take rumour or speculation as fact, for example in their article "Has AMD's Opteron outsold Intel's Itanium in just three months?" ( http://www.theinquirer.net/?article=10668 )
Look at the date on that article. It's amazing to see how they counted as a sale someone's intention to build a machine which still doesn't exist yet, and more to the point, will use CPUs that AMD haven't gotten any money, and probably haven't even built yet. Basically, they made something up to make a news story. If they had gone "Newsflash: AMD might sell 10000 CPUs to Sandia in a couple of years! Watch out Intel!" it might not have seemed so interesting, don't you think?
By waiting, Sandia will of course get better CPUs when they come around to building their machine. Nothing wrong with that. They're probably already happy they didn't spend their money on, say, 1.6GHz Opterons!
But there is a problem with concluding that Itanium isn't selling well. For its intended market, what you and I would call ultra-high-end systems (but what IBM or HP might call mid- to high-end UNIX servers), Itanium has been performing rather well given its more or less traditional big-iron pricing.