STR912 + PSRAM performance benchmark
mrousseau
running from internal Flash of the STR912 in comparaison with external
PSRAM.
At this time, STR912 is evalboard are only available without external
memory.
Marc
An Schwob in the USA
Marc,
not done with this one but with other ARM7 that have external bus,
even non-multiplexed 32-bit wide data. The performance of an STR912
will drop to (best case in thumb mode) less than 50% of what it is
capable of from internal flash. Going external there are always waits
involved, even if the PSRAM is 0ns. On top of that you need to latch
the address, you can only go 16-bit wide... It will be crawling. If
you need to execute program from external and need speed as well, go
for a device that supports non-mux and 32-bit wide data.
An Schwob
Spen
The str9 is a arm966 based core so will perform better than the arm7.
For a start it harvard architecture, so fewer wait states om code/data etc.
One thing i would say go for the bga version - the emi has a burst
interface that the other packages do not.
No timing info to offer as i used it to interface a tft driver - but the
screen update was good.
One other thing to remember there are new 1 and 2 MByte devices out now
(internal flash) - so you may not need external flash.
Cheers
Spen
Spen
The str9 is a arm966 based core so will perform better than the arm7.
An Schwob in the USA
Spen,
guess you never built a system with an STR9 using external memory and
compared it to a 32-bit wide non-mux, otherwise you would make such
statements!?
While the 966 core performs better than an ARM7, the memory system on
the STR9 even for on-chip memory kills most of the advantages. During
my current and my previous job, both companies benchmarked the STR9 to
several ARM7s. The fastest ARM7 was the NXP2468 running from internal
Flash at 72 MHz and depending on the program, was much faster than the
early STR9 and approx. 10% slower than the latest version of the STR9.
In other words, the STR9 was running 133% the clock rate and provided
110% of the speed. In absolute terms the STR9 was faster, in relative
terms it was slower. Don't know what has been screwed up with the
memory interface of the STR9 because it reads very nice in the manual,
just does not perform well.
Now back to the original question. Having a multiplexed external 16-
bit bus in an ARM966 versus a non-multiplexed 32-bit bus on an ARM7,
the ARM7 will provide approx. 2x the performance in ARM mode
btw. the STM32 performs pretty much with identical speed from internal
flash as the STR9 running 33% faster and it is not just because the
Cortex M3 core, it is much more for the better memory bus. The STR9
uses 2 chips in the package while the STM32 is just one die.
An Schwob
Ulf Samuelsson
If you want performance out of an ARM core with external memory
you are probably much better of with an ARM926 than an ARM966.
AT91SAM9260 is running at 200 MHz from the cache, and I have just
shipped some 400 MHz chips to some lucky.
Eagerly waiting for my first SAM9XE512 samples which will run at 200 MHz.
Should leave most other flash controllers in the dust ;-)
--
Best Regards,
Ulf Samuelsson
This is intended to be my personal opinion which may,
or may not be shared by my employer Atmel Nordic AB
Spen
> Spen,
>
> guess you never built a system with an STR9 using external memory and
> compared it to a 32-bit wide non-mux, otherwise you would make such
> statements!?
Guess i may have done a few, but using revH silicon and the burst emi
interface of the bga part.
>
> Now back to the original question. Having a multiplexed external 16-
> bit bus in an ARM966 versus a non-multiplexed 32-bit bus on an ARM7,
> the ARM7 will provide approx. 2x the performance in ARM mode
>
Maybe but most people would run thumb mode on the 16bit bus.
And anyway, the op asked about PSRAM not 16/32bit buses.
> btw. the STM32 performs pretty much with identical speed from internal
> flash as the STR9 running 33% faster and it is not just because the
> Cortex M3 core, it is much more for the better memory bus. The STR9
> uses 2 chips in the package while the STM32 is just one die.
>
i am due to receive the emi version of the stm32, which i agree will be
better, think half the problem with the str9 is that it came from the
upsd division rather then the mcu division.
Cheers
Spen
werty
And now lets imagine S/W that no longer
allows any profit .
The ARM 7's w/ 144 pins will power the
future "PC" . using about a tenth the PSRAM
that M$ O.S. would use .
Say good bye to Software .
These advances will hold no place for
the former electronics giant , the U.S.A.
Its Europe and Japan that will build the
new PocketPC , using ARM .
This will crash the entire electronics
industry in the USA , silicon valley
will not exist . Everything will be built
outside of the USA .
WIFI will topple all the govt owned Cell
phone phony companys .