[rescue] WTB: SGI

Björn Ramqvist rescue at sunhelp.org
Tue Jun 19 09:09:57 CDT 2001

"Joshua D. Boyd" wrote:
> I just read that paper.  I had a question.  A two TE machine has nearly a
> gigaflop of transformation power.  So, does this mean that a G4 can
> transform more geometry in software than an Impact can in hardware?

Well, in theory that is, yes. This also depends on what that
G4-processor is doing in between transforming those polys.
However, we're comparing apples and pears again. Let me point out why:

First of all, I believe that article is looking at the "first
generation" Impact boards used in the early generation Octanes. The SI,
SSI and MXI was replaced somewhere around '97-'98 to a new set of TE and
GE ASICs, which turned out to be SE, SSE and MXE graphics. Benchmarks
went up 10-50% for geometrical and texture-oriented calculations cause
of that ASIC-boost, leaving the ol'e I2 MaxImpact in the dust, atleast
in 95% of the cases.
Octane was introduced somewhere '96-'97 and the Impact graphics set were
dated back '93, so the graphics hardware really did need a refreshment.
The VPro family of graphics ('99-'00) is a breeze fresh air for the
aging Impact hardware.

Second, It's not uncommon for CPUs to be quicker than their graphics
hardware, where one specific example of this behaviour happened when
they released the Indy R5000:

Item           MFLOPS
 XZ Hardware     128
 R5000 150MHz    300
 R5000 180MHz    360


So generally, it wouldn't be such a bad idea to put G4's on a
graphicsboard and use them as geometric engines. Although a probably
expensive solution, it would in theory be a simple idea to offload the
main CPU.
(like SGI's graphics wasn't expensive?)


More information about the rescue mailing list