2041
AmigaOS 1.0 - 3.x / NatAmi MX
« il: 04 Giugno 2011, 22:24:06 »
interessante quanto riportato da Gunnar von Boehn in merito alle prestazioni di NatAmi MX con daughterboard CPU 68060 in merito a CPU benchmark:
http://www.natami.net/knowledge.php
As most people will know the 68060 has two ALU pipes.
This means the 68060 can execute several instructions in parallel.
Sysinfo is carefully written in a way which disables this feature of the 68060.
This means a 68060 will always look much worse in Sysinfo than in real live.
The 68060 can execute 2 instruction per clock (sometimes even 3)
The 68040 can peak execute 1 instruction per clock.
The 68030 can peak execute 0.5 instructions per clock. And much lower in average.
The 68000 can peak execute 0.25 instructions per clock. Much much lower in average.
The way Sysinfo is written it is not possible to execute more than 1 instruction per clock, this means basically only halve of the 68060 is used in the test. The 68060 will not behave like a 68060 but more like a 68040 in this test.
If a 100MHz 68060 scores 4.3 times better than an AMIGA 4000/040 then this means only for this special test.
In real life with normal code or code compiled for the 68060 the 68060 will in average score A LOT better. As normal code will make use of the feature to be able to execute 2 instructions per cycle.
Also its important to mind that SYSINFO is a pure CPU test.
Sysinfo does only measure CPU clockrate.
This means a 100MHz 68060 having super fast SRAM which can transfer 266 MB/sec - and a 68060 having much slower DRAM which can transfer 30MB/sec - they will both score the same number in the SYSINFO benchmark. As sysinfo only measures CPU clockrate.
Today is not a good day for benchmarking.
Even while Annikas NATAMI MX is faster than any existing classic AMIGA. Its still running in the equivalent of "factory default settings" and it is not upgraded to fast settings.
We are confident that there is a plenty of room for running it much faster. Give us a few more days to play with this and then I'll post some benchmarks which are really impressive.
http://www.natami.net/knowledge.php
As most people will know the 68060 has two ALU pipes.
This means the 68060 can execute several instructions in parallel.
Sysinfo is carefully written in a way which disables this feature of the 68060.
This means a 68060 will always look much worse in Sysinfo than in real live.
The 68060 can execute 2 instruction per clock (sometimes even 3)
The 68040 can peak execute 1 instruction per clock.
The 68030 can peak execute 0.5 instructions per clock. And much lower in average.
The 68000 can peak execute 0.25 instructions per clock. Much much lower in average.
The way Sysinfo is written it is not possible to execute more than 1 instruction per clock, this means basically only halve of the 68060 is used in the test. The 68060 will not behave like a 68060 but more like a 68040 in this test.
If a 100MHz 68060 scores 4.3 times better than an AMIGA 4000/040 then this means only for this special test.
In real life with normal code or code compiled for the 68060 the 68060 will in average score A LOT better. As normal code will make use of the feature to be able to execute 2 instructions per cycle.
Also its important to mind that SYSINFO is a pure CPU test.
Sysinfo does only measure CPU clockrate.
This means a 100MHz 68060 having super fast SRAM which can transfer 266 MB/sec - and a 68060 having much slower DRAM which can transfer 30MB/sec - they will both score the same number in the SYSINFO benchmark. As sysinfo only measures CPU clockrate.
Today is not a good day for benchmarking.
Even while Annikas NATAMI MX is faster than any existing classic AMIGA. Its still running in the equivalent of "factory default settings" and it is not upgraded to fast settings.
We are confident that there is a plenty of room for running it much faster. Give us a few more days to play with this and then I'll post some benchmarks which are really impressive.