Re: [ccp4bb]: PC 3D hardware

[Warren DeLano <warren@sunesis.com>]

***  For details on how to be removed from this list visit the  ***
***    CCP4 home page http://www.dl.ac.uk/CCP/CCP4/main.html    ***

I apologize to the list for drawing out this discussion, but Jeroen
raised some points about hardware that are particularly
relavent to crystallography visualization, and which merit follow-up.
The impact of rapidly changing computer
technology on crystallography and ccp4 users would seem to be an
appropriate thread for the list, IMHO.

Jeroen Mesters wrote:

>  triangles are of no great use at all. It's the OpenGl you will need
the
>  most. For all of you that want to read up on this check out the
>  very usefull site of http://www4.tomshardware.com. You will
>  find several tests of true OpenGL versus nVidia cards.....
>  http://www4.tomshardware.com/graphic/00q4/001213/index.html
>  Cards build for games that hardly use OpenGl can have
>  40 million triangles, but an OpenGl card with only 4 million
>  triangles will always be on top, sorry!

>  It is not for nothing that SGI introduced a subgraphics layer between

>  the NVIDIA cards and the PC in order to deal with OpenGl called
>  "OpenGl on a chip". Have a look at there linux machines. With the
>  rebate you can get, the price is not so bad at all....

Jeroen,

    I disagree and have the real-world benchmarks with SGI workstations
and nVidia cards to support my position.  Sure,
nVidia cards aren't necessarily as optimized for OpenGL, but nVidia
cards still outperform SGI in crystallography
applications because the hardware is so much faster.  Plus, most
crystallography and molecular modeling applications
*also* hardly use the OpenGL API -- just antialiased-depth-cued lines
and lit-shaded-untextured-depthcued triangles.
Beyond that, the quality of the OpenGL implementation is relatively
unimportant to crystallographers.

    Furthermore, triangles *are* of particular importance in
crystallography, as any kind of surface, stick, or ribbon
rendering uses triangles, and triangle primitives have typically been
used at the hardware level to render lines (electron
density meshes) on PC cards.  That is what will ultimately determine how
fast a program like O or PyMOL can update
the screen.  In real-world tests, I've confirmed that PyMOL can rotate
an electron density map, surfaces, or a
Molscript-generated 3D ribbons much faster on an nVidia/GeForce2 (under
Windows or Linux) than a Max Impact
Octane can with "real" OpenGL and real SGI hardware (under IRIX).
Other programs show similar trends.  Your
hypothesis just doesn't fit the data.


>    In addition, the very best graphics cards donot mean nothing with
the
>    very best driver supporting it!! NVIDIA has a one for all driver
for
>    all of their new cards under linux meaning you will loose
performance
>    here. The same is true for the PC itself, you better make sure it
is
>    linux certified in order to avoid problems later.


    That nVidia uses a common driver across all its cards contributes
significantly to the stability and ease of use of their
cards under Linux. You don't have to worry about driver incompatibility,
which is a typical source of angst under Linux.
Considering that modern cards do transformations and lighting in
hardware, custom-tweaked drivers may no longer be
as important for performance.

    Buying a PC with an SGI chip in it might make sense if the total
cost of ownership is comparable to a generic
consumer machine, as you will probably get some extra support in
exchange for the premium you will pay for the
hardware (assuming that SGI's newest cards catch up to nVidia).
However, if your goal is maximum crystallography
performance for minimum expense today, and you have access to Linux
experience within your group (important!), I
don't see how you can beat an Athlon/Linux machine with an
nVidia/GeForce2 -- the "graduate student special".

   Even if Linux expertise is not available, running under Windows 2k is
becoming possible as more software packages
now support it and you can use the cygwin package to create a full
unix-like shell environment under win32 that works
surprisingly well.

    There are plenty of legitimate reasons for sticking with SGI
hardware (legacy applications, experience, other
investments, proven compilers, stereo), but graphics performance is not
one of them.

    We still lack a low-end stereo graphics for PCs! Any vendors out
there listening?

- Warren

--
mailto:warren@sunesis-pharma.com
Warren L. DeLano, Ph.D.
Informatics Scientist
Sunesis Pharmaceuticals, Inc.
3696 Haven Ave., Suite C
Redwood City, CA 94063
(650)-562-3106 fax: (650)-556-8824