Slashdot Mirror
Big-Iron to Open Up for AMD
vincecate writes "Traditionally the key chips that have allowed companies to
scale multiprocessors to large numbers have been proprietary.
Some examples are the
Cray SeaStar,
SGI NUMAlink,
HP sx1000,
and the
IBM X3/Hurricane.
This proprietary paradigm is about to change to a more open one.
Two companies have developed key chips for
building large Opteron multiprocessors,
and they will be
commercial off-the-shelf parts.
PathScale has
released
InfiniPath
which can be used with an
Infiniband
switch to make
a high-bandwidth low-latency interconnect for a
supercomputer cluster.
The other company is
Newisys,
which
will soon release
the
Horus chip.
This chip will make it possible to build 32 socket
(64-core) shared memory Opteron systems."
Ehh, maybe. Normally "Big Iron" is associated with IBM but according to Wikipedia, the submitter is correct in using the term.
Just wanted to point out that the link to Newisys is just a blurb stating that AMD is releasing the Horus chip, and doesn't really have anything to do with Newisys, other than the fact that a couple of the people behind the AMD Horus release used to work there.
Oh, and the Horus link is a PDF whitepaper... please warn when a link points to a PDF.
"Trolls they were, but filled with the evil will of their master: a fell race..." -- J.R.R. Tolkien on Olog-hai
Actually, I don't think it would help much. Most games now don't benefit from 2 way SMP, so the benefit from 64 way is debateable to say the least. Still for servers, this thing might help. I suspect that most server applications/os's will have servere scaleability problems once you go this far SMP though.
BTW, Has anyone heard of the MLX1. Makes you wonder what would happen if you put a bunch of these on a chip with some clever caching and the mother of all memory controllers. x86 Niagra anyone.
echo -e 'global _start\n _start:\n mov eax, 2\n int 80h\n jmp _start' > a.asm; nasm a.asm -f elf; ld a.o -o a;
It's the same as the regular Opteron (otherwise stuff would break).
There is a Cray XT3 that runs at 15 Teraflops at Sandia and made out of 2ghz opterons and is currently the 10th fastest computer in the world. There is a similar machine over at Oak Ridge National Labs that runs at 14 Teraflops and is the 11th fastest computer in the world.
In fact, those lowly AMD kids seem to also have their chips on the fastest machine at the Pittsburgh supercomputing center (ranked 33rd fastest computer in the world) and the US Army Research Laboratory (ranked 39th fastest) . The latter was actually being built by IBM for ARL, you know those guys who coined the term "big iron".
The special thing about InfiniPath is that the adapter is not a PCI-(e|X) card but rather connects directly to the HyperTransport interface on the cpu (requiring a special MB with a "HTX" connector), giving slighlty lower latency than a normal IB adapter.
Umm, I know there's this odd phenomenon where many people tend to label any processor that's made by either Intel or AMD "non-proprietary" and any processor made by another company "proprietary", but even still this article is a little silly. SPARC processors have been in use since the late '80s, most people consider SPARC-based machines "Big Iron", and the SPARC processor architecture is fully open -- anyone who wants to can make SPARC processors. SPARCproductDIRectory lists a bunch of companies who currently do. In fact, there are probably just as many (if not more) SPARC manufacturers as there are X86 manufacturers.
But, it's not really designed for this.
And what makes you qualified to state this? Opterons were designed with EXACTLY this in mind, right off the drawing board - i'd dig up some old articles about it but I'm at work. Research Non-Uniform Memory Access (NUMA), supercomputing, and the Opteron's architecture as it related to those two. AMD knew what they were doing when they designed the Opteron - Intel has been completely out-engineered.
Quoth twiddlingbits yet again:
I don't think that you understood the word 'proprietary' in this context. Sun's technology, as well as (IBM's, SGI's, Cray's, etc.) is something that they each independently developed solely for their own company's use. They do not share or sell this technology to third parties to enable those third parties to build their own large multi-way servers using that technology. That is why their existing solutions are called 'proprietary'. It has nothing to do with if the bus that they use to connect the components is based upon some published open standard, whether that be PCIe, HyperTransport, or one that you think defines 'non-proprietary'.The two companies listed in TFA have developed solutions specifically to allow third party companies, whether they be Dell or ASUS or someone else, to build and sell multi-CPU solutions without developing the CPU interconnect technology themselves. That is why these new solutions are not 'proprietary'. They are not kept only for the private use of the company that developed them.
I hope that makes this point clear.
And as you tread the halls of sanity, You feel so glad to be, Unable to go beyond. I have a message, From another time..