🚀 go-pugleaf

Kralizec Craig wrote:
> "Daniel J. Pavon-Moreno" <djpavon@neoholistic.org> writes:
> 
> 
>>Hi, people, I'm new to this group.
> 
> 
>> I've got this very old/obsolete (but so cool) SS20 that I use mostly
>>as an X terminal. It came with the Leo framebuffer which is unbearably
>>slow (how a supposedly 'gfx accelerator' can be so slow is beyond me),
>>so recently I switched to the onboard SX fb (8 MB VSIMM). Having read
>>some old Sun brochures, and being the cg14 a local bus framebuffer, I
>>expected the SX's performance to be at least on par with an old S3 PC
>>gfx accel card from around 1995, but actually it's much slower:
> 
> 
> Remember that the SX/cg14 has *no* 3D hardware acceleration. It's strictly a
> very enhanced 2D 24-bit framebuffer with a fancy pallette DAC and MMU
> interface.
> 

Well, it can be configured to have a z-buffer of sorts using main 
memory. But what I question is its 2D performance.

> Remember also that in 1993/94 when the Sparcstation 20 was Sun's premium
> machine, the 8 MB VSIMM gave the SS20 graphics performance which had almost
> never been seen before in the Unix workstation market. SGI was about the
> only place putting lots of R + D into graphics hardware/software development
> at that time.
> 

An entry level sgi from that period runs circles around an SX SS20. 
Which is ok,
as they had the fastest gfx on desktop systems. As a comparison, Doom 
only runs smooth on my system if I run the DGA version; sgi Doom, which 
is a standard X11 version (uses XShm) runs smoother on an entry level Indy.

> You will not find any PC framebuffer from the 1993/4/5 period with anything
> like the same level of capability, particularly the same level of
> integration, that the SS20's on-board framebuffer provided. Sure the raw
> performance is lacking because of the host platform it's designed for.
>

Not quite right. My old S3 Virge is MUCH faster than the cg14/sx in 
about every aspect. A friend's Number Nine (don't remember the model but 
it was an S3 chip) is also a lot faster. Both cards are from 1995. Both 
use dram, which should be slower than vram. Both are installed on 90-100 
MHz Pentium systems (about the same 'speed' as 75-90 MHz SuperSPARCS). 
And both were far, far cheaper than the cg14 VSIMM was whe it was new.

> Make absolutely sure you configure the Xservers file correctly too - to get
> the best performance you *must* run Xsun in Truecolor mode. This means you
> need to specify:
> 
> -dev /dev/fbs/cgfourteen0 defdepth 24 defclass TrueColor
> 
> or similar in the line of the Xservers file which starts Xsun on the local
> display.
> 
> If you don't, Xsun will come up in PseudoColor mode and the colormap
> processing will hinder performance.
> 
> 

But I still have the 24 bit visual available to any program that uses 
it; on the other hand, running the server at defdepth 24 breaks a lot of 
software.

>>My hardware setup is: SM71 cpu, 256 MB RAM, 2 GB hd, 8MB 60ns VSIMM,
>>Solaris 2.5.1, running at 1152x900.
> 
> 
> Using a single SM71 CPU doesn't help either. You *need* as much CPU
> brute-force as possible when using any 24-bit video platform, and since the
> SX/cg14 is tied directly into the local buses on the SS20 system board,
> better CPU performance translates directly into better video performance.
> 

I see. Well, it makes sense. Maybe if I find a cheap cpu upgrade module 
I'll try it.

> Regards,
> 
> Craig.

Thanks for your help, Craig.

> --
>  SUN RIPENED KERNELS - Surplus Sun Microsystems Equipment, Parts + Accessories
>     Waterfall, NSW, Australia - Operated by Craig Dewick - Founded in 1996
> Main site: www.sunrk.com.au - Ebay Shop: www.ebayshops.com.au/sunripenedkernels
>   Ph: +612-9520-2547 - Fax: +612-9520-2557 - Mobile: 04-2163-0547 (int. +614)


Daniel