3d-bench
- To: MISC
- Subject: 3d-bench
- From: jfox@xxxxxxxxxx (Jeff Fox)
- Date: Thu, 3 Aug 1995 13:38:30 -0700
Dear MISC readers,
Dave wrote:
>Keep in mind that this benchmark also is comparing P21Forth (few CODE words)
>with LMI UR/Forth (highly optimized).
>
>P21 is generating video, and running from DRAM. With no video and running
>from SRAM, P21 should go 3 times faster. But, I haven't measured that
>configuration.
>
>CPU Seconds
>P21/SRAM 10 *estimated*
>486DX40 19
>P21/DRAM 30
>386DX20 35
>
>Given the efficiency differences between P21Forth and LMI UR/Forth, it looks
>to me like P21 *with video* is about 486DX40 class.
Having writen P21Forth I can say it generates code about an order of
magnitude slower than what an optimizing compiler could generate. I
have no exerience with LMI UR/FORTH but I know it is a much more highly
optimized implementation than P21Forth. So should include a factor
for this. I would say it is a bigger factor on this example than you
indicate.
F21 will have a much faster clock and a faster memory interface than
P21. It will also be easier to generate more highly optimized code
on F21.
In a multi-processor most nodes are just doing computation, not generating
video or other high memory bandwidth activities. On your app if you
combine the faster clock and more optimization and *no video* and
faster sram in F21 you might see:
F21node 1 (estimated)
I do thank you ( Dave ) for having posted some of the details of this MISC
benchmark.
Jeff Fox