Orad Hi-Tec Systems (orad_u++at++netvision.net.il)
Wed, 12 Mar 1997 19:33:20 +0200
> Another question that I have concerns the maximum performance I can expect
> from a framebuffer copy into the texture memory.
Good question.
> At this time, I use an internal format PFTEX_RGB8 for my texture, and I
> obtain a time of 14ms for the transfer of a 1024x1024 texture with 2 RM6s.
> Is it logical ?
> Does this time depend linearly of the number of RM ?
> How doest it vary in function of internal format of texels ?
> If I take the hardware documentation, it precises that one RM6 has a 80Mb
> transfer rate between frame buffer and texture memory, so for 3 eight bits
> components texels and 2 RM6s, the practical time that I have corresponds to
> a transfer rate of 180 Mb and not 160 Mb ?
> Am I exploding the performances of the IR (I don't believe so) or am I
> making a mistake in my computations ?
You didn't take into account the fact that the framebuffer memory is
split among the RMs, while the texture memory is duplicated
among the RMs.
This would imply (against intuition?) that the performance
with a single RM is better than with two, adn two is better
than four! This is because
with >1 RM, pixels have to be copied between RMs -- each RM must
fill the whole texture, while the framebuffer data resides on
other RMs as well.
So you need the cross-RM transfer rates as well.
Am I making a mistake?
Bye,
Moshe Nissim
Orad Hi-Tec Systems
=======================================================================
List Archives, FAQ, FTP: http://www.sgi.com/Technology/Performer/
Submissions: info-performer++at++sgi.com
Admin. requests: info-performer-request++at++sgi.com
This archive was generated by hypermail 2.0b2 on Mon Aug 10 1998 - 17:54:53 PDT