Dear Bob, You obviously know you tech stuff. The number crunching I (we) do is huge floating point calculations used in drug design (quantum dynamics/quantum mechanics). At the moment the fastest results come from the Opterons with 1Gb ram. I'm not sure of their spec. But, we have found calculations to be faster on 64bit machines rather than 32.
For imaging we use SGI Octane or Octane 2 however I have found recent Nvidia graphics cards to be as good, here at least I stick to crt SGI monitors; they are cheap and perform. Also, Irix (SGI OS) is awful. For weeks now I've been trying to install ssh. It WON'T. I was debating last night whether to install Gentoo on the Octanes. My fear is, If I do: a. Sybyl, the MOST useful drug design software won't run and. b. I will have to reinstall Irix. I did it once. It's a farce. It took all day (no breaks). You even partition when it's installed. I think SG have had their day. We won't buy any more. Furthermore, the Opterons are faster and cost 1/10 the price. > > It depends. In general, looking at MTBF numbers, Itanium systems are > less reliable than most 2P to 4P x86 servers out there. But more reliable > than most RISC based servers, though IBM's Blue Gene blade servers may > have an equivlant record or even be better. (Note: Realibility when I > use it means 24x7 full time compute load, not shutting it down at night > nor the system standing idle.) > > As to performance, it kind of depends. The bone stock Intel chipset > Itaniums have fairly constrained FSB to memory bandwidth. Itanums where > companies have used their own chipsets - such as us (SGI) and higher end > HP, among others, can deliever some impressive bandwidth to memory and > I/O. Thus improving overall computing throughput and high sustained > compute power. > > And depending upon the problem set, an Itanium based system may be faster > than amd64 based systems. If your problem set is more integer based or > transaction based, the amd64 will perform better. If it's more FP based, > memory constrained, threaded, and I/O bound, then an Itainium based system > might be better. > > If you need more than 1 TB physical memory and up to 512 cpus in a single > system image, then amd64 solutions won't be there until HyperTransport III, > sometime next year, perhaps. > > Downside - you have to recompile your apps to get the best out of them. Any > x86 32bit code will run in an emulation layer, which can be very fast or > very slow. Setiathome, 32 bit client screamed on our Itanium's (ran it > on a 512P system). Oh, and you'll need Intel's compiler. > > On the AMD side, figure about 80% FP performance and 102% Integer performance > compared to an Itanium cpu. Memory bandwidth tends to match when AMDs are > used in 4P or greater configs. I/O bandwidth is a bit constained in some > instances, but seems pretty tolerable and really isn't much of an issue > until more than 4 PCI-X slots are needed. But a lot of that is based > on which chipset is being used. And to get the est out of your apps, you'll > need to compile them with the PathScale compiler. > > If you're interested in an in-office/deskside style system. the AMD solution > is probably a better match, especially if your application needs only 1 or > 2 Gfx pipes, and fits in less than 16 GB memory. > > If you need department scientifc serving, then benchmarking your apps on each > platform is the better way to find which fits better. > > If you need serious visualizaton ability, with 3 or more - up to 16, Gfx > pipes, then there isn't much choice, no AMD vendor has produced a system > capable of doing that, though with some special Nvidia cards, it's possible > to get 4 Gfx pipe on 2 PCIe x16 slots. That's not to say it works well, as > it's > pretty much beta hardware and drivers at the moment. > > Bob > -- > - > QA Curmudgeon. > Wacky and bizarre testing(TM) performed while-U-wait. > - > See http://www.gnu.org/philosophy/no-word-attachments.html > - -- Dr Gavin Seddon School of Pharmacy and Pharmaceutical Sciences University of Manchester Oxford Road, Manchester M13 9PL, U.K. -- gentoo-amd64@gentoo.org mailing list
