Slashdot Mirror
Intel/HP Release Linux SDK For IA-64
HerbieTMac writes: "Intel and HP are releasing HP's IA-64 emulator for Linux later this week. Particularly interesting is that this emulator will be free (as in beer) for download from Intel's and HP's Web sites."
NUE provides the toolchain (compiler, linker, assembler), the libraries, and execution environment necessary to develop IA-64 Linux software. A user running within NUE can compile, link and execute applications as if he or she were on an IA-64 system executing the Linux operating system. (72 MB)
ftp://download.intel.com/design/IA-64/lnx_nue.htm
A little C code test:
/root]# cat moks.c
/root]# cc -O2 -g -omoks moks.c /root]# file ./moks /root]# ./moks
:-)
--------------------------------
[root@localhost
#include
#include
int main (void)
{
unsigned long int myvar;
printf ("Size of int : %d\n", sizeof (myvar));
exit (0);
}
[root@localhost
[root@localhost
./moks: ELF 64-bit LSB executable, IA-64, version 1, dynamically linked (uses shared libs), not stripped
[root@localhost
Size of int : 8
--------------------------------
So.. It obviously works; compiler and everything
This message is printed on 100% recycled electrons.
Q: How does a Unix guru have sex? A: unzip;strip;touch;finger;mount;fsck;more;yes;umount;sleep
It is an impressive piece of technology. /nue gets the root directory though, so your home directory in the NUE is /nue/home/myname.
/proc, which is not in any way symlinked to the host computer's /proc, but still contains up to date information on everything.
You start the NUE and get your usual UNIX shell,
The whole thing runs mostly IA32 binarys, but once you try to run IA64 code, it automatically triggers the SKI IA64 emulator via binfmt_misc. So it's really fast compared to a complete emulation, but still gives you a (more or less) native environment.
I have no idea, how they run this subsystem, it even has its own
(I'm using it for documenting how SGI's GPL Fortran 90 compiler interfaces their I/O-library, so that we can use it in GNU g95, the Fortran 95 compiler that is going to be part of GCC. See http://g95.sourceforge.net for information on this project)
Merced is supposed to be a server platform mainly for enterprise and scientific applications - not for your basic desktop, which is why it costs as much as it does.
According to this page, properly optimized code will be able to execute 8-12 parallel operations/cycle. This is hardly a "waste", as you put it, of designer effort. Willamette will only beat it in terms of IA-32 code, because Merced will only emulate it.
Running native IA-64 code, unless AMD's got something up its sleeve that no one's talking about, Merced will blow Sledgehammer and Willamette out of the water instruction-wise clock for clock. Kinda the reverse philosophy of Willamette.
IMHO, I think EPIC's going to kick some major hiney, in terms of pure processing power, as soon as Intel scales up the clock speeds (coming in McKinley and beyond). It's a pretty nice concept (removing guesswork from optimization) and I have to give Intel props for sticking with their guns.
Light a fire for a man and he'll be warm for a day. Light a man on fire and he'll be warm for the rest of his life.
For anyone who wants the actual links. There's a press release here that contains links to the download pages on both the Intel and HP sites. The Intel link takes a bit of searching to find otherwise. HP puts a link to this press release prominently on their main page.
The net will not be what we demand, but what we make it. Build it well.
Can I assume that this is not a "free speach" release because it is sooooo perfect that any bugs in the emulator will also be in the processor so you will have to get around them anyway?
Never underestimate the dark side of the Source
Well with Slegdehammer and Willament expected at 2GHz next year. And Foster still to be late and expensive. PowerPC and Alpha will tail a little at 1GHz-maybe 1.5 next year, maybe more. So i would say IA-64 hasn't got a chance until Northwood 3GHz in 2002/3 and then only if the rest of the industry is sleeping. Follow link for a leaked IA-64 roadmap http://www.google.com/search?q=cache:www.theregist er.co.uk/990428-000005.html+IA-64+roadma p&hl=en
Given that the uptake of the IA-64 is likely to be slow enough anyway, the idea of making people pay a hefty amount for this tool is one that's just too stupid for words. Intel and HP need to have software available for the chip, since otherwise nobody will touch it with a bargepole despite all of Intel's marketing might.
No, the two companies are taking a huge risk by starting with a fresh, non-compatible arhcitecture, and they're going to want as much support for it as quickly as possible. This way they can get the Linux developer crowd, always eager for a new platform, to start work on an IA-64 version as quickly as possible, so that the server market will be open to the chip.
No, it's not suprising really. Not doing this would have put Intel/HP at the mercy of closed source development houses, many of whom are going to be very unwilling to risk the jump to a new platform.
---
Jon E. Erikson
Jon Erikson, IT guru
The ISA is very big and complex and doesn't actually look any simpler than the x86 ISA. The designers also seem to have pretty much missed on what a good new architecture should actually do - they concentrated on trying to maximise IPC (Instructions Per Cycle - ie how many instructions you can issue per clock cycle), when current designs have pretty much already gotten as good as practically possible. So, all that complexity to increase IPC is pretty much going to go to waste...
I'm not that surprised it's late, consumes huge amounts of power (100W at 600MHz or something), and slower than expected. Also, given that they're going to cost $2000-$4000 each, and that Intel's own Pentium 4 (Willamette) will beat it in terms of pure performance, as well as price/performance, I don't think there's going to be much demand. (and multiprocessor Athlon motherboards will be out before too long, and there's plenty of good stuff from the RISC vendors if you wish to splash out...)
Intel/HP would have been much better off doing something like the Alpha 21464 (a few years away), the MAJC (few months away), or some of IBMs recent POWER chips - already on sale. I kinda pity the poor guys who actually have to implement the design since the higher level architects seem to have introduced the concept of bloatware to chip design...
Sorry, kinda been wanting to say this for a while now.
Interesting that the license says:
HP grants you a license to Use the Software solely for the purposes of (i) teaching and training of the IA-64 architecture by non-profit educational institutions and (ii) for developing software for Open Source operating systems.
It seems you can't use it to develop for a proprietary OS without special permission from HP.