A hack to run Mathematica 6 on Solaris x86 with an Intel CPU - Solaris

This is a discussion on A hack to run Mathematica 6 on Solaris x86 with an Intel CPU - Solaris ; Mathematica is supported on Sun's Solaris x86 operating system, but only on AMD 64-bit CPUs, not any Intel CPUs. http://www.wolfram.com/products/mathematica/platforms/ says: Sun x86 Solaris 10 64-bit Requires AMD 64-bit machine. Attempting to run the text-based kernel on an Intel Duo ...

+ Reply to Thread
Results 1 to 12 of 12

Thread: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

  1. A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    Mathematica is supported on Sun's Solaris x86 operating system, but only
    on AMD 64-bit CPUs, not any Intel CPUs.

    http://www.wolfram.com/products/mathematica/platforms/

    says:

    Sun x86
    Solaris 10 64-bit
    Requires AMD 64-bit machine.


    Attempting to run the text-based kernel on an Intel Duo Core 2 CPU
    causes it to crash immediately.

    ld.so.1: MathKernel: fatal:
    /usr/local/Wolfram/Mathematica/6.0/SystemFiles/Libraries/Solaris-x86-64/libsunperf.so.1:
    hardware capability unsupported: 0x100 [ AMD_3DNow ]

    Apparently modern Intel chips support *most* of the AMD instructions,
    but appears (for my Intel CPU at least) not the AMD_3Dnow instructions,
    which one of the libraries supplied with Mathematica needs.

    However, following the suggestion of Marc (marc.glisse ||at||
    gmail.com) on the comp.unix.solaris newsgroup, and posts from several
    others in quick succession (which would never have been possible on
    comp.soft-sys.math.mathematica with the delays the moderation causes on
    there), I had a solution. I replaced the library libsunperf.so.1
    supplied with Mathematica 6.0 with a later one supplied by Sun on their
    Solaris Express Developer Edition. (The actual file is part of the Sun's
    Studio 12 development system, so one only needs that, which is free. As
    is Solaris of course too).

    That allowed Mathematica to go a bit further, but then it failed with
    the message

    ld.so.1: MathKernel: fatal: libfsu.so.1: version `SUNWprivate_1.1' not
    found (required by file
    /usr/local/Wolfram/Mathematica/6.0/SystemFiles/Libraries/Solaris-x86-64/libsunperf.so.1)
    ld.so.1: MathKernel: fatal: libfsu.so.1: open failed: No such file or
    directory
    Killed

    The file libfsu.so.1 is also part of the Sun Studio 12 development kit,
    but Mathematica does not look for it. But one can copy that into the
    directory where Mathematica looks for libraries (or, as I did, make a
    symbolic link).

    Finally when all that is done, the kernel (math) works, but the front
    end (mathematica) crashed almost immediately.

    But that issue for Solaris is in the support FAQ for Mathematica,

    http://support.wolfram.com/mathemati...ntsolaris.html

    which says the -noTrueTypeFont option can fix problems with Mathematica
    crashing on Solaris. (Who is at fault here, Sun or Wolfram?)

    That did not help immediately, but I then realised the previous crashes
    probably resulted in corrupted cache data, so I started Mathematica with
    the -cleanStart option too.

    Then, it works!!! Mathematica is now running on a computer running
    Solaris x86 with an Intel CPU, not the supported AMD CPU. Obviously
    this hack will not be supported by Wolfram, but it appears to work.
    (Perhaps they would add support for Intel chips, as it seems possible
    with little effort).

    I would add it runs the Mathematica 6 benchmark quite slowly, scoring
    1.37 or so on a machine which scores 2.00 on Vista. (That said, the x86
    version was 6.0.0, and the Vista version 6.0.1). But Mathematica does
    work on Solaris x86 with an Intel CPU and Solaris is in my opinion a
    much better operating system than Vista.

    Use at your own risk, but the steps are below. Assuming Mathematica is
    in the default installation, you will need to be root.

    1) Change to the Directory Mathematica uses for libraries:
    # cd /usr/local/Wolfram/Mathematica/6.0/SystemFiles/Libraries/Solaris-x86-64

    2) Rename the original library to something else
    # mv libsunperf.so.1 libsunperf.so.1.orig

    3) Create a link from the later version of the library in the Sun Studio
    12 development system

    # ln -s /opt/SUNWspro/prod/lib/amd64/libsunperf.so.2 libsunperf.so.1

    4) Create a link from the file /opt/SUNWspro/prod/lib/amd64/libfsu.so.1
    to the directory used for the libraries.

    # ln -s /opt/SUNWspro/prod/lib/amd64/libfsu.so.1 libfsu.so.1

    5) Launch Mathematica with the options -cleanStart and -noTrueTypeFont

    $ /usr/local/bin/mathematica -cleanStart -noTrueTypeFont

    6) Next time, it should work without the -cleanStart option, but the
    -noTrueTypeFont one is still needed.

    $ /usr/local/bin/mathematica -noTrueTypeFont

    It now works!!

    It would be so nice if comp.soft-sys.math.mathematica newsgroup was not
    moderated, so allowed a fast exchange of views between people. It allows
    problems that need contributions from many to be solved quickly,
    something that seems impossible on comp.soft-sys.math.mathematica, but
    was on the unmoderated Sun newsgroup.

    Obviously the above will not be supported by Wolfram Research, so use at
    your own risk, but it does appear to work for me at least.

    Solaris x86 has been known as a very good operating system for
    Mathematica, with various records having been set on that platform

    http://www.sun.com/third-party/srsc/...Wolfram_ss.pdf

    My initial benchmark results suggest it does not work as well on the
    Intel CPUs as it does AMD, as the benchmark results are quite sluggish
    given the hardware. That said, this is a laptop, so it may be a result
    of the CPU being slowed by power management systems. On Vista,
    Mathematica does even slower (scoring about 1.00) if Windows is
    configured for maximum battery life, rather than Maximum performance. My
    guess is an AMD is a better CPU for Mathematica, but it does work on the
    Intel one, with the above hack.

    Overall, for best performance it seems hard to beat Solaris x86 on an
    AMD CPU, but this is quite usable.

  2. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    Dave wrote:
    > Mathematica is supported on Sun's Solaris x86 operating system, but only
    > on AMD 64-bit CPUs, not any Intel CPUs.
    >
    > http://www.wolfram.com/products/mathematica/platforms/
    >
    > says:
    >
    > Sun x86
    > Solaris 10 64-bit
    > Requires AMD 64-bit machine.


    I should add, I was using Solaris 10, not an earlier version.


    -bash-3.00$ cat /etc/release
    Solaris Express Developer Edition 9/07 snv_70b X86
    Copyright 2007 Sun Microsystems, Inc. All Rights Reserved.
    Use is subject to license terms.
    Assembled 30 August 2007
    -bash-3.00$ isainfo -v
    64-bit amd64 applications
    cx16 mon sse3 sse2 sse fxsr mmx cmov amd_sysc cx8 tsc fpu
    32-bit i386 applications
    ahf cx16 mon sse3 sse2 sse fxsr mmx cmov sep cx8 tsc fpu



    It's less likely this hack would work on older releases of Solaris, but
    given that Solaris is free (Sun will even ship you a DVD and them pay
    the postage)

    http://www.sun.com/software/solaris/...xpress/get.jsp

    it is hard to justify using an old version of Solaris. Not tested with
    earlier releases of Mathematica - it might well work.

  3. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    On Fri, 23 Nov 2007 10:11:04 +0000, Dave wrote:


    > Attempting to run the text-based kernel on an Intel Duo Core 2 CPU causes
    > it to crash immediately.
    >
    > ld.so.1: MathKernel: fatal:
    > /usr/local/Wolfram/Mathematica/6.0/SystemFiles/Libraries/Solaris-x86-64/libsunperf.so.1:
    > hardware capability unsupported: 0x100 [ AMD_3DNow ]
    >
    > Apparently modern Intel chips support *most* of the AMD instructions, but
    > appears (for my Intel CPU at least) not the AMD_3Dnow instructions, which
    > one of the libraries supplied with Mathematica needs.
    >

    This is pure nonsense. The fault is in Mathematica's code, not in x86_64
    implementation of Intel. There were some minor irrelevant differences
    with AMD64 in first version, but that was with P4, not Core 2 Duo.

    See here for details:
    http://en.wikipedia.org/wiki/X86-64

    scroll to "Recent implementations". Even on P4 architecture there was
    no difference for 3dnow code.

  4. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    > > Apparently modern Intel chips support *most* of the AMD instructions, but
    > > appears (for my Intel CPU at least) not the AMD_3Dnow instructions, which
    > > one of the libraries supplied with Mathematica needs.

    >
    > This is pure nonsense. The fault is in Mathematica's code, not in x86_64
    > implementation of Intel. There were some minor irrelevant differences
    > with AMD64 in first version, but that was with P4, not Core 2 Duo.


    3DNow is not part of x86-64, it is specific to AMD processors and it
    never caught on. You may be thinking of SSE, which was Intel's
    similar extension to x86, which did catch on and was included in the
    x86-64 spec. The instruction set compatibility table at
    http://www.agner.org/optimize/instruction_tables.pdf should clear up
    any questions about what specific processors support.

  5. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    Roman Pearce wrote:
    >>> Apparently modern Intel chips support *most* of the AMD instructions, but
    >>> appears (for my Intel CPU at least) not the AMD_3Dnow instructions, which
    >>> one of the libraries supplied with Mathematica needs.

    >> This is pure nonsense. The fault is in Mathematica's code, not in x86_64
    >> implementation of Intel. There were some minor irrelevant differences
    >> with AMD64 in first version, but that was with P4, not Core 2 Duo.

    >
    > 3DNow is not part of x86-64, it is specific to AMD processors and it
    > never caught on. You may be thinking of SSE, which was Intel's
    > similar extension to x86, which did catch on and was included in the
    > x86-64 spec. The instruction set compatibility table at
    > http://www.agner.org/optimize/instruction_tables.pdf should clear up
    > any questions about what specific processors support.


    Thank you. Which is why Mathematica will not run under Solaris x86 with
    an Intel CPU unless you charge a library for one which does not use
    AMD_3DNOW instructions.

  6. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    On Fri, 23 Nov 2007 22:18:43 -0800, Roman Pearce wrote:

    >> > Apparently modern Intel chips support *most* of the AMD instructions,
    >> > but appears (for my Intel CPU at least) not the AMD_3Dnow
    >> > instructions, which one of the libraries supplied with Mathematica
    >> > needs.

    >>
    >> This is pure nonsense. The fault is in Mathematica's code, not in x86_64
    >> implementation of Intel. There were some minor irrelevant differences
    >> with AMD64 in first version, but that was with P4, not Core 2 Duo.

    >
    > 3DNow is not part of x86-64, it is specific to AMD processors and it never
    > caught on.

    Precisely. 3DNow IS NOT part of 64 bit code.

    Let me cite history of x86_64:

    MMX, 3DNow!, x87, SSE, SSE2, and SSE3 extensions all hang off of the
    original x86 ISA, overlapping in many cases. x86-64 cleans things up by
    adopting SSE and SSE2 as part of its core set of instructions and
    jettisoning MMX, 3DNow!, and the x87 FPU.

    Now, since the OP is dealing with
    Solaris 10 64-bit
    Requires AMD 64-bit machine.

    we conclude that his Mathematica is either 32 bit only, but this should
    not be compatible with the following
    /usr/local/Wolfram/Mathematica/6.0/SystemFiles/Libraries/Solaris-x86-64/libsunperf.so.1:

    which is claiming to be a 64 bit library, OR

    the code is "64 bit".

    In either case the fault is in Mathematica's code which seems to be 32 bit
    only + proprietary extensions.

  7. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    On 23 Nov, 10:11, Dave wrote:
    > Mathematica is supported on Sun's Solaris x86 operating system, but only
    > on AMD 64-bit CPUs, not any Intel CPUs.
    >
    > http://www.wolfram.com/products/mathematica/platforms/
    >
    > says:
    >
    > Sun x86
    > Solaris 10 64-bit
    > Requires AMD 64-bit machine.
    >
    > Attempting to run the text-based kernel on an Intel Duo Core 2 CPU
    > causes it to crash immediately.
    >
    > ld.so.1: MathKernel: fatal:
    > /usr/local/Wolfram/Mathematica/6.0/SystemFiles/Libraries/Solaris-x86-64/libsunperf.so.1:
    > hardware capability unsupported: 0x100 [ AMD_3DNow ]
    >
    > Apparently modern Intel chips support *most* of the AMD instructions,
    > but appears (for my Intel CPU at least) not the AMD_3Dnow instructions,
    > which one of the libraries supplied with Mathematica needs.
    >
    > However, following the suggestion of Marc (marc.glisse ||at||
    > gmail.com) on the comp.unix.solaris newsgroup, and posts from several
    > others in quick succession (which would never have been possible on
    > comp.soft-sys.math.mathematica with the delays the moderation causes on
    > there), I had a solution. I replaced the library libsunperf.so.1
    > supplied with Mathematica 6.0 with a later one supplied by Sun on their
    > Solaris Express Developer Edition. (The actual file is part of the Sun's
    > Studio 12 development system, so one only needs that, which is free. As
    > is Solaris of course too).
    >
    > That allowed Mathematica to go a bit further, but then it failed with
    > the message
    >
    > ld.so.1: MathKernel: fatal: libfsu.so.1: version `SUNWprivate_1.1' not
    > found (required by file
    > /usr/local/Wolfram/Mathematica/6.0/SystemFiles/Libraries/Solaris-x86-64/libsunperf.so.1)
    > ld.so.1: MathKernel: fatal: libfsu.so.1: open failed: No such file or
    > directory
    > Killed
    >
    > The file libfsu.so.1 is also part of the Sun Studio 12 development kit,
    > but Mathematica does not look for it. But one can copy that into the
    > directory where Mathematica looks for libraries (or, as I did, make a
    > symbolic link).
    >
    > Finally when all that is done, the kernel (math) works, but the front
    > end (mathematica) crashed almost immediately.
    >
    > But that issue for Solaris is in the support FAQ for Mathematica,
    >
    > http://support.wolfram.com/mathemati...aris/notruetyp...
    >
    > which says the -noTrueTypeFont option can fix problems with Mathematica
    > crashing on Solaris. (Who is at fault here, Sun or Wolfram?)
    >
    > That did not help immediately, but I then realised the previous crashes
    > probably resulted in corrupted cache data, so I started Mathematica with
    > the -cleanStart option too.
    >
    > Then, it works!!! Mathematica is now running on a computer running
    > Solaris x86 with an Intel CPU, not the supported AMD CPU. Obviously
    > this hack will not be supported by Wolfram, but it appears to work.
    > (Perhaps they would add support for Intel chips, as it seems possible
    > with little effort).
    >
    > I would add it runs the Mathematica 6 benchmark quite slowly, scoring
    > 1.37 or so on a machine which scores 2.00 on Vista. (That said, the x86
    > version was 6.0.0, and the Vista version 6.0.1). But Mathematica does
    > work on Solaris x86 with an Intel CPU and Solaris is in my opinion a
    > much better operating system than Vista.
    >
    > Use at your own risk, but the steps are below. Assuming Mathematica is
    > in the default installation, you will need to be root.
    >
    > 1) Change to the Directory Mathematica uses for libraries:
    > # cd /usr/local/Wolfram/Mathematica/6.0/SystemFiles/Libraries/Solaris-x86-64
    >
    > 2) Rename the original library to something else
    > # mv libsunperf.so.1 libsunperf.so.1.orig
    >
    > 3) Create a link from the later version of the library in the Sun Studio
    > 12 development system
    >
    > # ln -s /opt/SUNWspro/prod/lib/amd64/libsunperf.so.2 libsunperf.so.1
    >
    > 4) Create a link from the file /opt/SUNWspro/prod/lib/amd64/libfsu.so.1
    > to the directory used for the libraries.
    >
    > # ln -s /opt/SUNWspro/prod/lib/amd64/libfsu.so.1 libfsu.so.1
    >
    > 5) Launch Mathematica with the options -cleanStart and -noTrueTypeFont
    >
    > $ /usr/local/bin/mathematica -cleanStart -noTrueTypeFont
    >
    > 6) Next time, it should work without the -cleanStart option, but the
    > -noTrueTypeFont one is still needed.
    >
    > $ /usr/local/bin/mathematica -noTrueTypeFont
    >
    > It now works!!
    >
    > It would be so nice if comp.soft-sys.math.mathematica newsgroup was not
    > moderated, so allowed a fast exchange of views between people. It allows
    > problems that need contributions from many to be solved quickly,
    > something that seems impossible on comp.soft-sys.math.mathematica, but
    > was on the unmoderated Sun newsgroup.
    >
    > Obviously the above will not be supported by Wolfram Research, so use at
    > your own risk, but it does appear to work for me at least.
    >
    > Solaris x86 has been known as a very good operating system for
    > Mathematica, with various records having been set on that platform
    >
    > http://www.sun.com/third-party/srsc/...Wolfram_ss.pdf
    >
    > My initial benchmark results suggest it does not work as well on the
    > Intel CPUs as it does AMD, as the benchmark results are quite sluggish
    > given the hardware. That said, this is a laptop, so it may be a result
    > of the CPU being slowed by power management systems. On Vista,
    > Mathematica does even slower (scoring about 1.00) if Windows is
    > configured for maximum battery life, rather than Maximum performance. My
    > guess is an AMD is a better CPU for Mathematica, but it does work on the
    > Intel one, with the above hack.
    >
    > Overall, for best performance it seems hard to beat Solaris x86 on an
    > AMD CPU, but this is quite usable.



    Just to say it works for me too on my Intel desktop.

    The fonts don't look too nice it must be said, that seems to be an
    issue with Mathematica on Solaris, rather than specific to the
    specific use of Mathematica on Solaris x86 using Intel chips.

    Richard

  8. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    On 23 Nov, 22:54, Burzum wrote:
    > On Fri, 23 Nov 2007 10:11:04 +0000, Dave wrote:
    > > Attempting to run the text-based kernel on an Intel Duo Core 2 CPU causes
    > > it to crash immediately.

    >
    > > ld.so.1: MathKernel: fatal:
    > > /usr/local/Wolfram/Mathematica/6.0/SystemFiles/Libraries/Solaris-x86-64/libsunperf.so.1:
    > > hardware capability unsupported: 0x100 [ AMD_3DNow ]

    >
    > > Apparently modern Intel chips support *most* of the AMD instructions, but
    > > appears (for my Intel CPU at least) not the AMD_3Dnow instructions, which
    > > one of the libraries supplied with Mathematica needs.

    >
    > This is pure nonsense. The fault is in Mathematica's code, not in x86_64
    > implementation of Intel. There were some minor irrelevant differences
    > with AMD64 in first version, but that was with P4, not Core 2 Duo.
    >
    > See here for details:http://en.wikipedia.org/wiki/X86-64
    >
    > scroll to "Recent implementations". Even on P4 architecture there was
    > no difference for 3dnow code.



    Whatever you say, the program fails to run without hacking the
    libraries.

    Richard

  9. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    Dave and others,

    Although the Mathematica Platform Availability page specifically
    states that only AMD processors are supported on Solaris x86, our
    intent is to support all systems running on Sun hardware regardless of
    whether those systems are based on AMD, Intel, or SPARC processors.
    And we will work toward that end.

    In regard to the issue with the Mathematica kernel and the SunPerf
    library (aka libsunperf.so.1), this is an issue that has only recently
    come to our attention. Our internal testing on the Sun Ultra 24, which
    is an Intel Core 2-based system, did not exhibit this problem. I
    would encourage anyone who has experienced the crash due to libsunperf
    to report this to the Wolfram Research Technical Support team at
    support@wolfram.com with the details of your system configuration.
    That will help us to determine the conditions under which this crash
    occurs.

    In regard to the issue with the Mathematica front end and TrueType
    fonts, this issue is not specific to Solaris x86, but occurs on
    Solaris SPARC as well. Earlier builds of Solaris 10 did not exhibit
    this problem, and it appears to have been introduced in a Solaris
    patch. We are working closely with Sun to resolve this.

    Wolfram Research is committed to continuing and enhancing support for
    Mathematica on the Solaris on the SPARC and x86 platforms. Thank you
    for bringing these issues to our attention.

    Regards,
    Schoeller Porter
    Technical Development Specialist
    Wolfram Research, Inc.

  10. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    schoeller.porter@gmail.com wrote:
    > Dave and others,
    >
    > Although the Mathematica Platform Availability page specifically
    > states that only AMD processors are supported on Solaris x86, our
    > intent is to support all systems running on Sun hardware regardless of
    > whether those systems are based on AMD, Intel, or SPARC processors.
    > And we will work toward that end.


    Wolfram may have intensions of supporting Mathematica on Solaris x86,
    but at the moment Mathematica has *no* chance at all of running on Intel
    systems which run Solaris x86 - unless the hack I put here is applied.

    > In regard to the issue with the Mathematica kernel and the SunPerf
    > library (aka libsunperf.so.1), this is an issue that has only recently
    > come to our attention.



    I think you are missing the point. The library supplied with Mathematica
    will *never* run on an Intel system running Solaris x86. It *needs* an
    AMD CPU.

    > Our internal testing on the Sun Ultra 24, which
    > is an Intel Core 2-based system, did not exhibit this problem.


    Are you 100% sure that Sun Ultra 24 was running Solaris, and not Linux?
    This Sun blog talks of Mathematica setting a record on a Sun Ultra 24

    http://blogs.sun.com/bmseer/entry/ma..._world_records

    But note the OS is not Solaris, but Linux.

    Same with this page

    http://www.sun.com/desktop/workstati...nchmarks.jsp#1

    It is Suse Linux, not Solaris x86.

    To the best of my knowledge, the hack I posted is mandatory if one wants
    to run Mathematica under Solaris x86 on an Intel CPU. I can't see any
    other way it can be done.

    > I
    > would encourage anyone who has experienced the crash due to libsunperf
    > to report this to the Wolfram Research Technical Support team at
    > support@wolfram.com with the details of your system configuration.
    > That will help us to determine the conditions under which this crash
    > occurs.


    I've spent quite a bit of time looking at this, but clearly I can't be
    expected to know as much as Wolfram Research. But my belief is that the
    Mathematica for Solaris x86 (both 6.0 and 6.0.1) and no hope at all of
    running on an Intel CPU as shipped.

    > In regard to the issue with the Mathematica front end and TrueType
    > fonts, this issue is not specific to Solaris x86, but occurs on
    > Solaris SPARC as well. Earlier builds of Solaris 10 did not exhibit
    > this problem, and it appears to have been introduced in a Solaris
    > patch. We are working closely with Sun to resolve this.


    Let us know when you have a solution. Perhaps you can post it on
    sci.math.symbolic. I get fed up trying to use Mathgroup, due to the
    moderation.

    > Wolfram Research is committed to continuing and enhancing support for
    > Mathematica on the Solaris on the SPARC and x86 platforms. Thank you
    > for bringing these issues to our attention.


    I get the very distinct impression that Mathematica is not tested as
    well on Solaris as it is under Windows. Every time I've seen a member of
    WRI staff with MMA it has been on a Windows environment. I get the
    feeling that the internal usage on Solaris is probably a lot less than
    on Windows.

    I'm afraid to say I have seen this problem with other software. Labview
    in particular was very buggy under Solaris. The same can not be said for
    Mathematica, but it seems to have a number of quite fundamental issues.


    > Regards,
    > Schoeller Porter
    > Technical Development Specialist
    > Wolfram Research, Inc.


    Well than you Schoeller. It's good to see some response from WRI on this
    matter - an email to support has received no useful response.


    But unless I am mistaken, there is no way you can expect people to run
    MMA on Solaris x86 with Intel chips unless you ship a different library.

  11. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    Dave,

    > > Our internal testing on the Sun Ultra 24, which
    > > is an Intel Core 2-based system, did not exhibit this problem.

    >
    > Are you 100% sure that Sun Ultra 24 was running Solaris, and not Linux?
    > This Sun blog talks of Mathematica setting a record on a Sun Ultra 24
    >
    > http://blogs.sun.com/bmseer/entry/ma..._world_records
    >
    > But note the OS is not Solaris, but Linux.
    >
    > Same with this page
    >
    > http://www.sun.com/desktop/workstati...nchmarks.jsp#1
    >
    > It is Suse Linux, not Solaris x86.


    The Sun Ultra 24 we used ran Solaris 10. Our internal testing was not
    related to the performance testing by Sun. As you note, Sun's
    performance testing was done with Linux.


    > I've spent quite a bit of time looking at this, but clearly I can't be
    > expected to know as much as Wolfram Research. But my belief is that the
    > Mathematica for Solaris x86 (both 6.0 and 6.0.1) and no hope at all of
    > running on an Intel CPU as shipped.


    I am glad that you are able to get Mathematica running on your system
    by replacing libsunperf.so.1 with the Studio 12 version. I appreciate
    the effort you've put in this, and I'm sure the information you've
    provided here and on other forums will be helpful understanding the
    root cause of this problem and in finding a solution.


    > > In regard to the issue with the Mathematica front end and TrueType
    > > fonts, this issue is not specific to Solaris x86, but occurs on
    > > Solaris SPARC as well. Earlier builds of Solaris 10 did not exhibit
    > > this problem, and it appears to have been introduced in a Solaris
    > > patch. We are working closely with Sun to resolve this.

    >
    > Let us know when you have a solution. Perhaps you can post it on
    > sci.math.symbolic. I get fed up trying to use Mathgroup, due to the
    > moderation.


    Just today I learned that the TrueType problem is an issue with
    fontconfig on Solaris. We do not yet have a solution for this, but I
    thought you (and the others who've encountered this problem) would
    appreciate knowing the cause.


    Regards,

    Schoeller.


  12. Re: A hack to run Mathematica 6 on Solaris x86 with an Intel CPU

    schoeller.porter@gmail.com wrote:

    >> Are you 100% sure that Sun Ultra 24 was running Solaris, and not Linux?
    >> This Sun blog talks of Mathematica setting a record on a Sun Ultra 24
    >>
    >> http://blogs.sun.com/bmseer/entry/ma..._world_records
    >>
    >> But note the OS is not Solaris, but Linux.
    >>
    >> Same with this page
    >>
    >> http://www.sun.com/desktop/workstati...nchmarks.jsp#1
    >>
    >> It is Suse Linux, not Solaris x86.

    >
    > The Sun Ultra 24 we used ran Solaris 10. Our internal testing was not


    Do you know if your internal testing was done with a retail Mathematica
    or some development version? As far as I can tell, the library in
    Mathematica 6.0 and 6.0.1 needs the AMD_3Dnow instruction set, and there
    are no Intel CPUs with that.

    In contrast, a later version of the library do not need the AMD_3Dnow
    instructions.

    > I am glad that you are able to get Mathematica running on your system
    > by replacing libsunperf.so.1 with the Studio 12 version. I appreciate
    > the effort you've put in this, and I'm sure the information you've
    > provided here and on other forums will be helpful understanding the
    > root cause of this problem and in finding a solution.


    There were two changes. First was to replace a library, but once that
    was done, there was an error message about a missing library. I found
    that library as part of Studio 12 too. I can't recall the names, but I
    documented it, so anyone looking to do this can check themselves.

    I've had a few times where the front end could not start the kernel.
    Quitting and starting again would usually solved it. That was quite
    common on 6.0, but seems to have been fixed in 6.0.1.

    > Just today I learned that the TrueType problem is an issue with
    > fontconfig on Solaris. We do not yet have a solution for this, but I
    > thought you (and the others who've encountered this problem) would
    > appreciate knowing the cause.


    Thank you.

    I did a Google with the terms Mathematica, fontconfig and Solaris,
    which found this:

    http://www.opensolaris.org/jive/mess...ssageID=138015

    Since it's not in English, I don't know what it says, but it might shed
    some light on the problem.

    It's good to see Wolfram Research supporting Mathematica on Solaris x86.
    From what I can gather, Maplesoft have no intention of porting Maple to
    Solaris x86 and there are conflicting comments on the web about any
    intentions Mathworks may or mat not have of porting MATLAB. Considering
    both MATLAB and Maple can be found on Solaris (SPARC), this seems a bit
    odd. have SPARC ports of the products, this seems a bit odd.

    My own very limited testing (just one computer), found Mathematica under
    Vista scoring the Benchmark[] a lot high than on Solaris - 2.00 on
    Vista, 1.37 on Solaris x86.

+ Reply to Thread