[Freemint-list] Kronos results on Amiga + Vampire V2
Vincent Rivière
vincent.riviere at freesbee.fr
Wed Mar 1 15:50:39 MSK 2017
On 01/03/2017 à 12:35, Philippe noble wrote:
> You got more or less the same horsepower with Aranym and a 40$
> Raspberry Pi3.
You are certainly right.
The fact is that Apollo people are "real hardware" people, they don't
like emulators, so they rarely use them, or even compare Apollo
performance with them. Of course benchmarks like Amiga SYSINFO give very
good results compared to original Amiga hardware. Similar good results
can bee seen when comparing CT60 to ST or TT.
In the real world, JIT emulators such as WinUAE or ARAnyM running on
today's computers literally blast real hardware. But this doesn't matter
for real hardware lovers.
> Beside the FPU which is missing, can we expect more horsepower when
> Apollo core will be completed? Higher frequencies?
The current situation may not be so bad. I'm not sure that Kronos CPU
benchmarks are really relevant in the real world. They test only a few
instructions in loops. Real programs may behave differently. For
example, a lot of work has been put in Apollo 68080 for instruction
parallelization, branch prediction, etc. I'm not sure that these
features are properly tested with Kronos.
On the other hand, a real test case would be to run a lengthy program,
not doing much disk I/O, and see how long it runs on several machines.
There is also much room for performance gain in the FPGA. Gunnar
regularly improves performance of many instructions.
And about higher frequencies, yes, they can be used. The Appolo Team
regularly make tests with higher clocks, that's really promising. This
adds a significant performance increase.
And maybe there are also some other optimization domains we completely
missed (alignments, etc.) and which could improve performance.
IMHO, there is still much room for performance gains.
At least, this benchmark show you something concrete.
BTW: Chip RAM is very slow on Vampire, while FastRAM is very fast. It
seems that Kronos mixes the memory results, so the overall memory
performance looks bad. Look at the details, you will see the difference.
The tested EmuTOS version (optimized for Vampire) uses FastRAM for
anything possible.
--
Vincent Rivière
More information about the Freemint-list
mailing list