========================================================================
	#######
	   #     #    #  #    #     #    #    #   ####
	   #     #    #  ##   #     #    ##   #  #    #
	   #     #    #  # #  #     #    # #  #  #
	   #     #    #  #  # #     #    #  # #  #  ###
	   #     #    #  #   ##     #    #   ##  #    #
	   #      ####   #    #     #    #    #   ####
........................................................................
	   #                             #####     #    ######
	  # #    #    #   #####   ####  #     #   # #   #     #
	 #   #   #    #     #    #    # #        #   #  #     #
	#     #  #    #     #    #    # #       #     # #     #
	#######  #    #     #    #    # #       ####### #     #
	#     #  #    #     #    #    # #     # #     # #     #
	#     #   ####      #     ####   #####  #     # ######
========================================================================


ABOUT AUTOCAD PERFORMANCE (on IRIX):

	Many factors can affect AutoCAD performance on SGI-IRIX
	workstations.  With different tests, different factors may be
	dominant.  The default settings are usually fairly good, but it
	is quite common to get a 2X (or better) speed-up just by
	changing a few system settings.  With any test, it is important
	to understand what it does, how it works and make sure it is
	repeatable.  A big step along this way is to automate the
	test.  AutoCAD can be automated through IRIX shell scripts,
	AutoCAD command scripts, AutoLISP source files and compiled ADS
	applications, or any combination of the above.  AutoCAD has
	built-in timing support and can do just about everything
	automated.  The worst "benchmark" scenario is to type in a few
	commands, look at a clock, hit Enter and then check the time
	when the task completes.  AutoCAD's internal clock uses
	"wall-time".  That is, it is the elapsed clock time from start
	to finish.  Irix, unlike some operating systems, is fully
	multi-tasking.  There may be 50 (or more) independent processes
	running:

		network activity

		remotely logged-in users

		print spoolers

		other background processes

	When running an AutoCAD benchmark test, the elapsed time may be
	increased due to other processes using CPU cycles while the
	benchmark was running.  You will get faster times by taking the
	machine off of the network, shutting down un-needed processes,
	etc.  However, in the "real world", this may not be practical.
	Also, you may want the benchmark results to reflect a "real
	world" environment.


Obtaining maximum AutoCAD performance involves several tasks:

	1. Choosing and configuring the right machine for your tasks.

	2. Configuring the operating system to work optimally.

	3. Configuring the application to work optimally.

	4. Configuring the display driver to work optimally.

	5. Integrating AutoCAD efficiently into the entire design cycle
	   with collaborative tools.


Optimum machine configuration:

	Physical memory:

		Autodesk recommended configuration:

			32 MB on all Unix platforms.

			 8-16 MB on DOS and Windows platforms.

		If you notice faster performance on a 32MB Pentium vs.
		a 32MB Indy, realize you are using several times the
		recommended memory on the PC and the bare minimum on
		the Indy.

		You will want additional memory if you intend to work
		on large drawings.  Consult the "AutoCAD Installation
		and Performance Guide" for guidelines.  I recommend
		64MB to start with, preferably in one bank of 4-16MB
		SIMMs to allow room for future expansion.  Unlike most
		PCs, Unix workstations can make effective use of all
		the memory you can put in them.  A quite way to see if
		more memory is needed is to launch the program
		'gr_osview -a' and watch the various resource indicator
		displays while running AutoCAD.  If you see a lot of
		CPU time spent waiting for swap (yellow bar) or the
		free memory indicator getting small (green bar) this is
		a sign that memory contention is happening.
		
		See discussion below on memory vs. paging vs. swapping.

	CPU and memory architecture:

		All MIPS CPUs include a primary cache.  Beginning with
		the R4000SC, secondary cache support is usually
		standard.  The MIPS CPU architecture really benefits
		from a large cache.  AutoCAD is an application that
		benefits from a large cache.  You will see a least 2X
		performance gain from a secondary cache.  AutoCAD
		R13_c4 has been optimized for increase floating point
		performance the R4000 (or greater) CPU family.

		In testing, I have found:
		
			R4000PC suffers from lack of secondary cache.
			R4000SC runs twice as fast as R4000PC.
			R4600PC runs quite well without secondary cache.
			R4600SC runs quite well without secondary cache.
			R4400SC is nearly twice as fast as the R4000SC.
			R5000SC is faster than the R4400SC.

		If you are using a CPU without the benefit of a
		secondary cache (i.e. the R4000PC or R4600PC) you might
		look into upgrading your CPU to one with a secondary
		cache.  AutoCAD is a large program and greatly benefits
		from the 512KB-1 MB secondary caches on the *SC CPU
		models; i.e R4600SC, R4400SC, and the R5000SC (the
		latter has an optimized GL graphics library).

		CPU	 Clock MHz (1)	Primary KB (2)	Secondary KB (3)
		-------	 -------------	--------------	----------------
		R3000 (4)      20-33		64		n/a	

		R4000PC		100		16		n/a
		R4000SC		100		16	       1024

		R4600PC	    100-150		32		n/a
		R4600SC	    100-150		32		512

		R4400PC     100-150		32		n/a
		R4400SC     150-250		32	     1024/2048 (5)

		R5000PC         150		64		n/a
		R5000SC     150-180		64		512
	
		Table: CPU vs. Clock speed and Cache sizes

		Notes:	(1) Typical, may not be available on all platforms.
			(2) Total of instruction plus data cache.
			(3) Typical, may vary by platform.
			(4) This CPU not supported with R13_c4.
			(5) 1M on Indy, 2M on Indigo^2.
			
	Graphics hardware:

		AutoCAD will run on all SGI graphics hardware.  The
		"best" hardware to use depends on your needs, like
		screen resolution, color resolution, amount of graphics
		acceleration, etc.

		Screen Resolution:	1024x768
					1280x1024
					
		Color Resolution:	8-bit (with virtual 24 bit output)
					24-bit true color output
					
		Graphics acceleration:	Host-based graphics processor
					1-8 dedicated graphics processors.

		The default X/Motif driver will usually exhibit similar
		performance on all platforms.  The GL driver
		performance will scale with increased graphics
		sub-system performance.

	Hard disk configuration:

		AutoCAD can be a disk-intensive application, depending
		on the task.  Not only is the drawing and its ancillary
		files (menu, fonts, XREFs, etc.) being read from and
		written to hard disk, but also, there are temporary
		files including the page file and rendering face files,
		in addition to the IRIX swap files.  Run the program
		'gr_osview -a' during an AutoCAD run and see a lot of
		CPU wait activity in the IO or PIO areas, this is a
		good indication of a disk-bound process.  If you have
		more than one disk drive and/or more than one SCSI disk
		controller (i.e. Indigo^2) you might want to
		considering setting up AutoCAD and IRIX so that the
		various disk accesses occur on separate disk drives.
		You can control the swap file locations with the IRIX
		Swap Manager tool.  You can control AutoCAD's page file
		locations via the environment variables ACADPAGEDIR,
		AVEPAGEDIR and AVEFACEDIR.  You can control the
		placement of AutoCAD's other temporary files via the
		 _CONFIG
		  ->Configure operating parameters
		   ->Placement of temporary files menu screen.
		Ideally, you should place all these various disk
		activities on separate disk drives for maximum
		performance.  If you do not have enough drives to
		permit this, try various combinations to see what works
		best for you.

		File system performance:

		You could also try the new high performance XFS file
		system available on IRIX 5.3 and 6.2.  With it you can
		use larger file systems (1TB or more), file sizes,
		block sizes and disk striping across multiple disk
		drives to increase disk I/O performance.  Aside from
		the basic bigger/faster file system features, behavior
		in areas like directory lookup have been improved.  The
		older EFS file system uses a linear directory format
		that slows dramatically as the number of entries in the
		file system increases.  XFS on the other hand uses a
		tree based directory structure that maintains its
		performance better with larger file systems.

			Entries		EFS	XFS
			-------		----	----
			   100		5970	8972
			  1000		1596	7089
			 10000		 169	6716

		Table: Directory entries vs. lookup operations/sec


IRIX CONFIGURATION:

	IRIX swap file size and use:

		More swap space will allow larger drawings to be edited
		(see also: ACADMAXMEM).  More swap space will not
		affect AutoCAD's speed.  Rule of thumb: Set the swap
		size to 2X physical RAM size.  There are several ways
		to change swap file size, you should use the IRIX Swap
		Manager (accessible from the System Manager) to change
		settings.  Generally, swapping to a dedicated partition
		(raw swap) is somewhat faster than swapping to a local
		or NFS file.  However, changing a dedicated swap
		partition involves re-partitioning and re-formatting
		a new or existing hard disk.

		To see what you currently have, use "swap -s":
		> swap -s
		  total: 19.40m allocated + 48.80m add'l reserved = 68.20m 
		  bytes used, 135.64m bytes available

		On this system, I have the default 40MB swap partition
		on the root disk plus 64MB added in a local file system
		in addition to the 64MB of RAM or a total of 168MB or
		virtual memory.

		Hint: Setting up swap space, either dedicated or file
		system vs. using ACADMAXMEM vs. installing additional
		RAM can be complicated, at best.  A simple way to
		approach the problem us to rank the techinquies
		according to performance vs. ease of implementation.

		ACADMAXMEM: (to force AutoCAD to use its internal paging
		mechanism) is a simple as setting an environment
		variable, it writes to a temporary disk file only when
		needed and the space is recovered afterwards.  However,
		the performance may be slower on a fragmented hard
		disk.  If you only work on large drawings occasionally
		this is a good approach.  You should consider running
		the file system defragmenter (fsr) to ensure a disk with
		lots of contiguous free space.

		LOCAL SWAP FILE is easy to set up with the Swap
		Manager.  The swap file size can be changed in a matter
		of minutes to meet changing virtual memory
		requirements.  This is a good choice if you sometimes
		need to work on larger projects for a period of time.
		The swap performance via a local (or NFS) file system
		is usually faster that AutoCAD paging (because the swap
		file is set up in a contiguous block).  However, since
		the swapping goes through the file system, it is not
		quite as fast as raw swapping to a dedicated
		partition.

		SWAP PARTITION gives the fastest swapping performance.
		By bypassing the IRIX file system and writing directly
		to the hard disk, swapping can be very efficient.  If
		you have the luxury of a spare hard disk, an effective
		technique can be to dedicate that disk to swapping.  Be
		sure it is a fast disk, though.  Otherwise, you'll have
		to re-partition an existing disk drive to change swap
		file size.  You'll need to backup and restore any data
		on that disk.  In any event, you'll want to consult
		your system administrator unless you are comfortable
		with the steps involved.  However, if you are setting
		up a new system for AutoCAD, you might choose this
		approach.  Use the swap space rule of thumb (above) to
		determine a swap size based on the amount of RAM in the
		system (currently or planned).  For example in one
		machine I set up, I initially started with 64MB of RAM,
		but anticipated that I may upgrade to 128MB in the
		future, so I set the swap partition to 256MB
		initially.

		MORE RAM will give you better performance that swapping
		to disk, but the cost is higher.  You should always add
		RAM in the biggest increment you can afford and no
		smaller than the biggest bank of RAM already in the
		machine.  If you need to replace a bank with higher
		density RAM, replace the one with the lowest density.

	Shared Memory Xserver transport:

		One excellent way to get increased AutoCAD performance
		is to take advantage of a little known feature of the X
		Window system, namely the Shared Memory Transport
		(shm).  This feature is only available on local
		connections, that is AutoCAD running and displaying on
		the same machine.  The feature is also a limited
		resource, but when used properly, gives excellent
		performance.  Basically, shm replaces the default
		Xserver network-based connection (i.e. `uname -n`:0)
		with a shared memory buffer (shm:0).  Since the memory
		buffer is shared between the client and server
		applications, no copying of data is required between
		them.  This is a big win in interactive operations
		(fast turnaround time) and in applications with large
		amounts of data, such as redrawing million vector
		AutoCAD drawings.

		Since shm is a limited resource, you don't want to use
		it for everything.  Rather, use it where it makes most
		sense, for example only for the AutoCAD Graphics
		window.  Fortunately, AutoCAD has an easy way to do
		just this:

			setenv ACADDISPLAY shm:0
			
		You can do this setting in ~/.cshrc or in your personal
		AutoCAD start-up script.  I have observed about a 10%
		performance increase in some benchmark tests.

	Backing store (on/off):

		Backing store is a feature of the X Window System
		whereby a backup copy of each window is maintained,
		which is used to repaint the window when needed.

		Setting it off is best for GL driver in the Bonus
		Pack.

		Setting it on is best for Motif driver, but it can be
		run without it.  On will consume more memory,
		especially over time (see also: Xserver termination,
		below).

		Certain hardware/software combinations may not support
		backing store.

		To see how its set now, use "/bin/ps -ef | grep Xsgi":
		> /bin/ps -ef | grep Xsgi
		> ... /usr/bin/X11/Xsgi -gamma 1.700000 -c -bs -pseudomap 4sight 
							   ^^^
		"-bs" means backing store is off, otherwise its on.

		To change the setting, edit the file:
			/usr/lib/X11/xdm/Xservers

		Note: In our testing with the X/Motif driver, simply
		having backing store enabled results in 5-10%
		performance loss in AutoCAD.  If any of the AutoCAD
		windows are further obscured a further 10% loss in
		performance is observed, due to the Xserver having to
		maintain a separate image of the obscurred window.

	Xserver termination on logout:

		Setting terminateServer to True recovers Xserver memory
		every logout.

		It is a very good idea to use this if you have backing
		store enabled, and won't hurt anything if you don't

		The time for a logout/login cycle will be a second or
		so longer.

		To change the setting, edit the file:
			/usr/lib/X11/xdm/xdm-config

	AutoCAD process priority:

		Normally an Irix process (like AutoCAD) is run at a
		default priority setting.  That is, it competes equally
		for CPU cycles and other system resources with other
		processes on the system.  The Irix "nice" command can be
		used to alter this default process priority.  As a
		regular user, you may only lower a process priority
		(making the process behave "nicer").  As super-user, you
		may increase the process's priority.  Lowering the
		process priority may be useful when running AutoCAD in
		the background, especially with the "AutoScript" command
		feature, or when running multiple AutoCAD sessions on a
		given machine (in conjunction with the ACADMAXMEM
		setting).

	Kernel Tuning Parameters

		Please consult the IRIX Administration Guide: System
		Configuration and Operation section (Appendix A) for a
		complete description of IRIX kernel tuning parameters.
		Below are extracted, for your convenience, several
		parameters that can affect AutoCAD operation under
		IRIX.


	bdflushr - specifies how often the dirty file system buffers
		   are flushed

		Description of bdflushr

		The bdflushr parameter specifies how often, in seconds,
		the bdflush daemon is executed; bdflush performs
		periodic flushes of dirty file system buffers.

		Value of bdflushr
			Default: 5
			Range: 1-31536000

		When to Change bdflushr

		This value is adequate for most systems. Increasing
		this parameter increases the chance that more data
		could be lost if the system crashes. Decreasing this
		parameter increases system overhead. Operations, such
		as loading an assembly or drawing, cause significant
		I/O waits.  Sometimes increasing this value to 60 or
		120 seconds can help reduce this bottleneck.


	nbuf - number of buffer headers in the file system buffer cache

		Description of nbuf

		The nbuf parameter specifies the number of buffer
		headers in the file system buffer cache. The actual
		memory associated with each buffer header is
		dynamically allocated as needed and can be of varying
		size, currently 1 to 128 blocks (512 to 64KB).  The
		system uses the file system buffer cache to optimize
		file system I/O requests. The buffer memory caches
		blocks from the disk, and the blocks that are used
		frequently stay in the cache. This helps avoid excess
		disk activity.  Buffers are used only as transaction
		headers. When the input or output operation has
		finished, the buffer is detached from the memory it
		mapped and the buffer header becomes available for
		other uses.  Because of this, a small number of buffer
		headers is sufficient for most systems.  If nbuf is set
		to 0, the system automatically configures nbuf for
		average systems.  There is little overhead in making it
		larger for non-average systems.

		Value of nbuf

			Default: 0 (Automatic if set to 0)
			Formula: 100 + (total number of pages of
			memory/40)
			Range: up to 6000

		When to Change nbuf

		The automatic configuration is adequate for average
		systems. If you see dropping `cache hit' rates in
		sar(1M) and osview(1M) output, increase this parameter.
		Also, if you have directories with a great number of
		files (over 1000), you may wish to raise this
		parameter. Less than 85% on the %wcache and/or less
		than 60% on the %rcache indicates that the system is
		I/O bound.


	dwcluster - (EFS) number of delayed-write pages to cluster in
			  each push.

		Description of dwcluster

		This parameter sets the maximum number of delayed-write
		pages to cluster in each push.

		Value of dwcluster

			Default: 64 

		When to Change 

		It should not be necessary to change this parameter.
		The automatically configured value is sufficient.


	autoup - (EFS) specifies the age, in seconds, that a buffer
		       marked for delayed write must be before the
		       bdflush daemon writes it to disk.

		Description of autoup

		The autoup parameter specifies the age, in seconds,
		that a buffer marked for delayed write must be before
		the bdflush daemon writes it to disk. This parameter is
		specified in /var/sysgen/mtune. For more information,
		see the entry for the bdflushr kernel parameter.

		Value  of autoup

			Default: 10 
			Range: 1-30

		When to Change 

		This value is adequate for most systems. 


	semmni - Maximum number of semaphore identifiers in the kernel. 

		Description of semmni

		The semmni parameter specifies the maximum number of
		semaphore identifiers in the kernel. This is the number
		of unique semaphore sets that can be active at any
		given time.  Semaphores are created in sets; there may
		be more than one semaphore per set.

		This parameter is specified in /var/sysgen/mtune/sem.

		Value of semmni

			Default: 10

		When to Change semmni

		Increase this parameter if processes require more
		semaphore sets. Increasing this parameter to a large
		value requires more memory to keep track of semaphore
		sets. If you modify this parameter, you may need to
		modify other related parameters.

Note:   1. AutoCAD and ADS:

	AutoCAD uses semaphores for interprocess communication with ADS
	applications including ASE.  To get an idea how many semaphores
	you need load one AutoCAD+ADS session and run the command
	'ipcs' and note the results.  Now load a second session and
	re-run 'ipcs'.  The change will tell you how many semaphores
	are used per session.  You may want to do this after rebooting
	to ensure you have no 'lost' semaphores.

	2. AutoCAD and ASE:
	
	You may sometimes experience 'lost' semaphores and/or shared
	memory message queues with ASE due to improperly disconnected
	SQL sessions.  If you are experiencing these kinds of problems,
	you can use the 'ipcrm' command to 'clean-up' the 'lost'
	resources but you should also correct the situation causing the
	problem.  This most frequently occurs when initially trying to
	establish connections to Oracle and Informix database servers
	and having improper user names and passwords.  Once the proper
	connection method has been established, this problem should not
	re-occur.

Setting Tunable Parameters

	The directory, /var/sysgen/mtune, contains the files defining
	tunable parameters that are typically tuned with the systune
	command. An example of changing a parameter is shown below.

Updates will be made to running system and /unix.install

	# systune -i
	systune-> nbuf
	nbuf = 509 (0x1fd)
	systune-> nbuf 1600
	nbuf = 509 (0x1fd)
	Do you really want to change nbuf to 1600 (0x640)? (y/n) y

In order for the change in parameter nbuf to become effective, reboot
the system

	systune-> quit
	# autoconfig -v
	# reboot

An alternative method of setting parameters is to edit the file
/var/sysgen/stune (as root) and add lines for each tunable parameter
you wish to change.  With this method, you can preserve the original
system settings (in /var/sysgen/mtune/*) and keep your changes confined
to the local stune file.  After editing the file, run:

	# autoconfig -v
	# reboot

as described above.


AutoCAD Settings:

	AutoCAD's memory pager (controlled with ACADMAXMEM):

		AutoCAD, by default will use all available
		(physical+swap) memory space, competing with other
		programs for resources.

			Using the default setting may cause the
			workstation to run slower, as parts of AutoCAD,
			the driver, or the O/S get swapped to disk.

		Setting ACADMAXMEM in the 10-16MB range on systems with
		32MB or less is a good staring point.  A value about 1/2
		the physical RAM is a good starting point.

		If you have additional memory, increase this setting
		appropriately.

		If running multiple copies of AutoCAD, reducing
		ACADMAXMEM is crucial.

		AutoCAD's memory pager is quite intelligent and
		efficient.

	VIEWRES setting:

		This AutoCAD setting controls the number of vectors
		used to display circles, text, etc.

		Using too large a value can inflate memory use,
		especially in 32-bit display list mode.

		Try to use a value in the 100-1000 range for most
		purposes.  Values way less than 100 are even better
		(performance-wise).  Set it as low as possible without
		affecting your drawing accuracy.
		
	SPLINESEGS setting:
	
		This AutoCAD setting controls the number of line
		segments that make up a spline curve.
		
		Using too large a value can inflate memory use,
		especially in 32-bit display list mode.

		The default value is 8, but lower values are fine.  Set
		it as low as possible without affecting your drawing
		accuracy.

	MAXACTVP setting:

		This setting controls the maximum number of active
		viewports allowed in AutoCAD.  By default is can be up
		to 64, determined by the display driver's
		capabilities.  If you only use a few viewports, you can
		reduce this value to a smaller number to reduce memory
		and CPU overhead.

	DRAGMODE setting:

		This controls whether objects being dragged are
		displayed.  If the object is complex, this can make the
		dragging operation much more interactive.

		The GL display driver also has an option to try to
		cache or not cache dragged objects.  Caching is useful
		when the object is complex and not changing from frame
		to frame, such as in a _MOVE or _COPY command.  If
		however, you use a _ROTATE or _SCALE command, the object
		must be erased and redrawn differently each frame.  For
		this operation, the caching-off version is best.  These
		controls are accessable in the Configure->DriverOptions
		dialog.

	ISAVEPERCENT setting (some AutoCAD versions):

		This setting controls the incremental saving of
		drawings.  Seeting this to 50(%) can speed drawing
		saves.  With R13_c3, this value must be left a 0 to
		avoid a potential problem.

	ISOLINES setting:

		Specifies the number of isolines per surface on
		objects. Valid integer values are from 0 to 2047 and the
		default value is 4.  Setting this to 0 results in only
		no additional iso-lines being created.

	FACETRES setting:

		This setting controls the "smoothness" of rendered
		solid objects.  Values can range from 0.01 to 10.00.
		The default is 0.50.  Smaller values result in coarser
		surface meshes and therefore smaller display lists and
		for exporting to Inventor and VRML, smaller files as
		well.

		You should set this value to the smallest that produces
		adequate results.  One place you may consider using a
		higher value is in the production of multiple Level Of
		Detail (LOD) versions of your model for VRML worlds.
		For example, creating a complex, smooth version for
		close viewing, a less complex/detailed version for
		middle distance viewing and a very simple/coarse
		version for distant viewing.  You can then import and
		combine these three LODs into CosmoWorlds in the LOD
		editor.

	SCROLL Lines setting:

		AutoCAD on Unix offers the ability to maintain a
		scollable text screen and command prompt area.  Several
		X resource settings control this feature.  These are:

			AutoCAD-Graph.scrollLines:      mmm
			AutoCAD-Text.scrollLines:       nnn

		These settings control the number of lines of text to
		be maintained in the back-scroll buffer for the command
		prompt area of the graphics window and the text window
		respectively.  By default they assume rather large
		values (from 100's to 1000's of lines).  If your
		AutoCAD application or test produces any quantity of
		text output, you may find the back scroll buffer may
		impact performance.  Therefore, especially if you don't
		find this feature of value, you should consider
		reducing the size of the backscroll buffer down to a
		minimum of the size of the window.  After all this is
		the situation you get on a DOS machine, e.g. in a 25
		line text screen, the last 25 lines of text output are
		"remembered".

	3DFACES (with AVE Render - R12 only):

		Render is part of the Autodesk Visualization
		Extension.  One of its features is the ability to pass
		3D data directly from AutoCAD to the GL display
		driver.  Doing so bypasses all the complicated
		processing normally needed to display rendered images.

		The 3DFACES command enables the 3D display list
		capabilities.

			On drawings of 10,000 faces, 3D rendering can
			be more than 10 times faster than 2D
			rendering.

			This allows use of the hardware Z buffer, and
			graphics accelerator for better image quality.

			Allows the 3D display list to be replayed in
			the same viewport any number of times without
			regeneration.

		Hint: See the Inventor file export feature (below).
		
	Hidden line removal:

		In R12 this process is optimized for 32-bit CPUs and runs
		10-100X faster than R11 style hidden line removal.

		If at all possible, don't use the "R11-style" hidden
		line algorithm, as it is not optimized.

		The GL Driver also offers a hardware accelerated hidden 
		line viewing mode that is 10-100X faster, again, (see
		3DFACES).


GL Driver/Bonus Pack Settings:

	The GL driver is the primary component of the Bonus Pack
	software.  It gives you all the features of the default (Motif)
	AutoCAD display driver but also provides many enhanced
	features.  This is primarily done through providing direct
	access the GL graphics engine that is in every SGI
	workstation.  It is a full-featured AutoCAD driver and has a
	number of user-configurable options:

	Display list coordinate size (16/32 bits):

		32 bit mode is better for REGEN-intensive tests.

			Uses at least twice the memory of 16-bit mode.

			Allows you to zoom in to your drawing without
			causing a REGEN.

		16 bit mode has better memory use and faster drawing.

			It is about 5% slower in REGEN's, but 10%
			faster in REDRAW's.

	Refresh Display list (on/off):

		The refresh display list management can lead to excessive
		memory use during editing and automated scripts.

		Enabling the GL driver's Refresh Display list mode
		helps refresh the graphics screen after window
		exposures; e.g. popping the graphics window.  However,
		it comes at a price, both in extra memory for the
		refresh display list as well as extra time in redrawing
		the refresh display list.  Imagine a 1M vector drawing
		in which you select all vectors for a subsequent
		operation.  AutoCAD wants to highlight each selected
		vector, these are normally non-drawing vectors, but in
		refresh mode, the driver saves them to use in case the
		screen needs refreshing.  After the selection set is
		complete, AutoCAD dehighlights all the vectors with
		more non-drawing vectors.  As a result, you end up with
		a 1M vector drawing display list and a 2M vector
		refresh display list.  The idea for turning refresh mode
		off is that once you've selected some geometry (for
		example via a dialog box), you'll probably change
		whatever has been selected upon dismissing the dialog. 
		Rather than going through the effort of 'repairing' the
		window damage under the dialog just to throw it away and
		draw the results of your change, why not just ignore the
		non-drawing stuff and proceed as if nothing happened.

			If you rarely restack windows or can tolerate
			some minor window damage (i.e. missing UCS
			icons, etc.) and want more interactive speed,
			turn refresh off.  A REDRAWALL or CLEAN command
			will always clean up the screen for you if
			needed.

			If you do a lot of window restacking and demand
			accurate graphics display turn refresh on or use
			the X/Motif driver with backing store enabled. 
			Note that refresh mode is still not perfect.

			Most benchmarks will run faster with refresh
			off.

	Display list cleaning (on/off):

		AutoCAD's display list management can lead to extra
		memory use during editing and automated scripts.

		Enabling the GL driver's RT Clean mode helps reclaim
		memory during automated tests.

			This is good while running AME and ADS
			applications.

		Turning RT Clean off is faster during manual,
		interactive operations like moving, copying and block
		insertions.

			Most benchmarks will run faster with Cleaning off.

	Display list page size (4K-256K):

		Larger pages are more efficient in terms of speed (to a
		point, 32KB is a good size).

		Smaller pages use less memory, especially with many
		viewports and are also more efficient when display list
		cleaning is enabled.

	Screen mode (Default/single/dual):

		AutoCAD runs with two windows, but it is possible to
		configure the behavior of these windows to suite your
		needs.

		The default mode is user-friendly, but causes
		un-necessary overhead.

		Single screen mode is best used when running command
		scripts.

			The text screen is not kept updated, so it is
			much faster.

			Often 2/3 of a test's time may be consumed with
			text display.

		Dual screen mode is a good compromise between the above
		two modes, the text screen is updated, but the overhead
		is less.

	Polygon Outlining (on/off):

		By default, the IrisGL interprets polygons as the set
		of pixels contained inside the polygon border.  This
		provides good performance and visual results for
		rendered scenes composed of polygon meshes that cover
		the surface.  However, AutoCAD also uses polygons to
		represent PLINES (to allow for line width).  As such,
		these polygons may not actually fill the scene and if
		the zoom ratio of the current view is just right, there
		may be no screen pixels contained within the polygon
		and the PLINE will "disappear".  To avoid the visual
		affects of this behavior, the GL driver allows an
		explicit polygon outline mode, where the polygon border
		is retraced as a line to guarantee its being visible,
		regardless of view.  This features comes at a price of
		about a 50% speed penalty in REDRAWs.  If you need this
		mode enabled, by all means turn it on, but if your
		style of drawing does not need this feature, turn it
		off (the default setting) for increased speed.  The
		SGI_OUTLINEn command controls this setting.
		
			SGI_OUTLINE0 is off
			SGI_OUTLINE1 is on (default)

	Entity Highlighting (on/off):
	
		According to the ADI specification, selected entities
		are to be drawn once with the entity color with a
		highlighting pattern (e.g. "- - - - ") then again in the
		background color with an alternate pattern (e.g. " - - -
		-").  The advantage of this technique is that you can
		more easily determine the selected entity, for example
		assume you draw a red line over a green (existing) line.
		A selection would probably hit the green line (as it is
		'first') and it would appear on top and highlighted. 
		However its drawback is that lines must be drawn twice
		(with a line pattern change in between each line).  An
		alternative approach (SGI_HILITE0) is to only draw the
		background pattern over the line for a highlight.  This
		cuts the drawing requirements in half and allows the
		line pattern to stay constant.  Also, in practice, if
		you have two red line on top of each other, the color
		alone is not sufficient to tell one from the other.

			SGI_HILITE0 is off (simple highlight)
			SGI_HILITE1 is on (default)

	Motif or GL user interface:

		Using the GL user interface components is faster and
		uses less memory.

			You give up the iconic pull-down menu capability.

			You still get Motif dialog box support.

		You may want to retain the Motif status line and
		toolbar features.

		Do this by setting the AutoCAD-Graph*use... items to
		FALSE in .Xdefaults.

	Window size:

		Using a smaller window can produce faster redraw speed
		and reduce memory use.

		Iconifying the graphics window can also speed automated
		operations and free screen real estate for other uses.

		The NULL display driver is quite efficient for doing
		background tasks, and is used by the AutoScript
		feature.  Check out the shell script file named
		'autoscript' in your AutoCAD executable directory.
		Finally the GL driver allows has several features to
		replace the NULL display with some added benefits.  The
		NULL display does not support the rendering interface
		and can therefore not handle operations requiring
		rendering (like 3DSOUT).  Also, the GL driver is better
		optimized for IRIX and runs faster than the NULL display
		driver.

			To enable the Autodesk NULL display:

				setenv ACADDISPLAY NULL

			To disable the Autodesk NULL display:

				unsetenv ACADDISPLAY

			To enable the GL NULL display mode:

				setenv ACADDISPLAY .
				setenv ACADTEXTDISPLAY .

			To enable the normal GL NULL display mode:
			
				unsetenv ACADDISPLAY
				unsetenv ACADTEXTDISPLAY

		To free up screen real estate with the GL driver, you
		can also re-direct the text screen to an alternate
		display screen via the ACADTEXTDISPLAY variable. 
		Setting ACADTEXTDISPLAY to a valid X display will send
		it to that display; e.g.:
		
			setenv ACADTEXTDISPLAY host:0
			
		Setting it to a TTY port will send output to a text port
		or serial terminal connected to the system.  See a
		further discussion of this feature in the GL driver
		release notes.
		
	Integration with the operating system:

		The Bonus Pack provides full integration of AutoCAD
		into the Indigo Magic environment.  Among the supported
		features are Drag&Drop, file type definitions,
		automated AutoCAD command script handling.  These
		features allow you to to work more productively.

	Printing and Plotting:
		
		Impressario gives fast network printing capabilities.

		Your host processor functions as a PostScript raster
		image processor, driving the printer at its maximum
		speed.

		The Bonus Pack includes built-in support for
		Impressario printing.

	Inventor/VRML file export:

		Inventor is a 3D visualization environment standard on
		all SGI machines.  It allows you to view, manipulate
		and share your AutoCAD drawings with other users who
		may not have access to AutoCAD or know how to use it.

		This is also the gateway into collaborative design
		allowing you to incorporate AutoCAD drawings into
		presentations, electronic mail and even with live
		interactive conferences with others.
		
		This is also a handy technique to use to preview parts
		libraries and designs in progress.  Rather than the time
		consuming HIDE, SHADE or RENDER process in AutoCAD,
		simply export the model in Inventor and have it ready to
		view at a moments notice.


BOTTOM LINE:

	The bottom line to tuning AutoCAD is to realize there are many
	factors influencing performance on a Unix workstation.  If you
	make a systematic and well-planned search through some of the
	adjustments, you'll probably find settings that produce optimal
	performance for your typical tasks.
	
	
FEEDBACK:

If you find any other factors that affect performance of AutoCAD and
would like to share that information, please send me via:

	INTERNET:	mailto:r.c.brown@ieee.org
	WWW:		http://reality.sgi.com/autocad/
