From mn1@helix.nih.gov Fri Aug 1 19:14:55 1997 Received: from helix.nih.gov for mn1@helix.nih.gov by www.ccl.net (8.8.3/950822.1) id TAA13418; Fri, 1 Aug 1997 19:10:58 -0400 (EDT) Received: (from mn1@localhost) by helix.nih.gov (8.8.5/8.8.5) id TAA12951; Fri, 1 Aug 1997 19:10:58 -0400 (EDT) Date: Fri, 1 Aug 1997 19:10:58 -0400 (EDT) From: "M. Nicklaus" Message-Id: <199708012310.TAA12951@helix.nih.gov> To: CHEMISTRY@www.ccl.net Subject: Re: Alpha + Linux ? Cc: mn1@helix.nih.gov, choic@gusun.acc.georgetown.edu On Fri, 1 Aug 1997, Cheol Choi wrote: > According to the SPEC benchmark, Dec Alpha seems to be the > fastest processor currently. > And I know some companies are selling Alpha workstations with Linux OS at > very reasonable price. > Is there anyone who has good experience with Alpha + Linux combination? We've had some experience with Alpha + Linux recently, and are going to have more very soon. Has it been good? Yes and no. The Alpha chip is indeed a very fast processor. The new motherboards with the DEC Pyxis chip set promise to be even faster, 600 MHz systems are selling (albeit still quite expensive), and processors with clock rates of 800 MHz and higher are in the pipeline. What's the drawback then? Mostly the software. The Linux for the Alpha (axp-linux) is much less mature and stable than Linux for the x86 type processors. A big problem are the compilers. f2c is plain buggy, and gcc and the linker also have their problems. If you want to know more (probably more than you want to know...) about these technical axp-linux issues, you can follow the discussion on the mailing list axp-list@redhat.com (to subscribe: send e-mail to axp-list-request@redhat.com with 'subscribe' as the subject), or check the newsgroup comp.os.linux.alpha. The software we tested on axp-linux was: CHARMM runs GAMESS runs AMSOL runs Gaussian 94 doesn't compile (see also previous postings on the CCL) When I last checked some of the Web sites on scientific applications for Linux, such as http://SAL.kachinatech.com/Z/2 or http://chpc06.ch.unito.it/chem_linux.html, I counted more than 70 Chemistry/Biology related programs. We sure haven't tested them all, but I'd guess if you get the source code for them, you could give it a shot at compiling them on an Alpha Linux system. If you have access to a Digital Unix system, you can actually compile your program on the DEC system and then run it on the Alpha Linux machine -- the executables are binary compatible (if you use static linking for the libraries). This doesn't always work (didn't work for G94), but we've used this method repeatedly for other programs. The benchmarks we ran showed very nice results indeed. The speedup we saw for various application relative to a Pentium PRO 200 MHz system running under Linux was between 1.5 and 2.2. From the clock rates, you would expect 2.5 (we used a 500 MHz Alpha chip). The advanced chip architecture itself might allow even higher numbers. The reason we havent't seen those numbers (yet) are most likely compiler issues. I can only recommend that you test various compiler options (optimization etc.) if not different compiler to see which gives you the fastest code. The bottom line, from my point of view, is: It's a great platform, but still in its early stages. So, if you like to patch operating system source code, rebuild the kernel, tweak the compilers, and do other system 'hacking' regularly, go for it (...it's not really that bad, but you get my drift). We *are* using such a system to do real comp.chem. work. However, if you want a system that is totally out-of-the-box, runs every known application without a problem, has plenty of stable compilers, etc., then you'd probably want to wait a little bit longer. Marc ------------------------------------------------------------------------ Marc C. Nicklaus National Institutes of Health E-mail: mn1@helix.nih.gov Bldg 37, Rm 5B29 Phone: (301) 402-3111 BETHESDA, MD 20892-4255 USA Fax: (301) 496-5839 http://www.nci.nih.gov/intra/lmch/MCNBIO.HTM Laboratory of Medicinal Chemistry, National Cancer Institute, & Lab. of Structural Biology, Div. of Computer Research and Technology ------------------------------------------------------------------------