[HP49G] Very poor graphing performances

[Bruno]

Hi All,

While I'm comparing graphing capabilities beetween my HP48S/GX/49G,
I was wondering seeing that in some cases the 48S was faster than the 48GX, and the 48GX was faster than the 49G !

In fact, the 48GX and the 49G INFORM's dialogs add extra computation time VS a simple << ERASE DRAX DRAW >> .

But, the 3D graphing tools show me an even bigger computation time difference, see below :


Benchmarked calculators :

HP48S ROM rev J + SUITE3D 4.0
HP48GX ROM rev M (the slowest I own, only 3.6Mhz !)
HP49G ROM 2.10-7
HP49G ROM 1.24
HP49G+ ROM 2.15


Benchmark test (from the HP48GX Graphics book by Ray Depew) :

PPAR / VPAR:

EQ: 'SIN(X)*SIN(Y)'
XVOL: -PI PI
YVOL: -PI PI
ZVOL: -1 1
XXRNG: -PI PI
YYRNG: -PI PI

WIREFRAME:
X-STEP: 10
Y-STEP: 8
PS-CONTOUR:
X-STEP: 20
Y-STEP: 16

Program:

<< << ERASE DRAW >> TIMED >>
(TIMED: from HP48 Insight by Bill Wicks)


Benchmark results:

WIREFRAME:

HP48S: 44.33s
HP48GX: 32.45s
HP49G: 47.98s


PS-CONTOUR:

HP48S: 2mn 52s
HP48GX: 5mn 03s
HP49G: 18mn 09s
HP49G+: 4mn 54s


Is there any flags that can explain why the HP49G was so slow ?

[Han]

(08-25-2015 08:40 AM)Bruno Wrote:  Hi All,

While I'm comparing graphing capabilities beetween my HP48S/GX/49G,
I was wondering seeing that in some cases the 48S was faster than the 48GX, and the 48GX was faster than the 49G !

In fact, the 48GX and the 49G INFORM's dialogs add extra computation time VS a simple << ERASE DRAX DRAW >> .

But, the 3D graphing tools show me an even bigger computation time difference, see below :


Benchmarked calculators :

HP48S ROM rev J + SUITE3D 4.0
HP48GX ROM rev M (the slowest I own, only 3.6Mhz !)
HP49G ROM 2.10-7
HP49G ROM 1.24
HP49G+ ROM 2.15


Benchmark test (from the HP48GX Graphics book by Ray Depew) :

PPAR / VPAR:

EQ: 'SIN(X)*SIN(Y)'
XVOL: -PI PI
YVOL: -PI PI
ZVOL: -1 1
XXRNG: -PI PI
YYRNG: -PI PI

WIREFRAME:
X-STEP: 10
Y-STEP: 8
PS-CONTOUR:
X-STEP: 20
Y-STEP: 16

Program:

<< << ERASE DRAW >> TIMED >>
(TIMED: from HP48 Insight by Bill Wicks)


Benchmark results:

WIREFRAME:

HP48S: 44.33s
HP48GX: 32.45s
HP49G: 47.98s


PS-CONTOUR:

HP48S: 2mn 52s
HP48GX: 5mn 03s
HP49G: 18mn 09s
HP49G+: 4mn 54s


Is there any flags that can explain why the HP49G was so slow ?

On the HP49, did you specify the plot parameters as "real" numbers and not "integers" ? The algorithms for the plots are based on "real" input. Any non-real input is automatically converted, and adds overhead.

[Bruno]

All VPAR and PPAR parameters are real numbers (with the dot)

[Han]

Haven't had any time or interest to poke into the HP49G ROM, but I wonder if the graphing code resides in Flash memory, and whether there is a speed difference between memory chips on the HP48 and HP49.

[Bruno]

Thanks for your answer Han,
I think they are at least 2 reasons since the extra time (vs the GX) are too different for the WIREFRAME (47% slower) and PS-CONTOUR (260% slower!!) graphs.
One common reason could be the hardware and/or the operating system implementation difference beetween the HP48GX and the HP49G,
and another reason specific to the PS-CONTOUR graph could be the different implementation of the 'differentiate' function (rewritten for the CAS ?)
used by the PS-CONTOUR algorithm.

I'm amazed that nobody was shocked before.

[Han]

(09-01-2015 03:09 PM)Bruno Wrote:  Thanks for your answer Han,
I think they are at least 2 reasons since the extra time (vs the GX) are too different for the WIREFRAME (47% slower) and PS-CONTOUR (260% slower!!) graphs.
One common reason could be the hardware and/or the operating system implementation difference beetween the HP48GX and the HP49G,
and another reason specific to the PS-CONTOUR graph could be the different implementation of the 'differentiate' function (rewritten for the CAS ?)
used by the PS-CONTOUR algorithm.

I'm amazed that nobody was shocked before.

If I am not mistaken, both the 48G and 49G use the original Saturn chips for the CPU. They have different bankswitching methods. In particular, code residing in covered ROM behave differently. Covered ROM on the 48G series gets copied into RAM before the code is executed (from RAM), whereas on the 49G many flash pointers are run directly from ROM. Only a few ROM routines are copied to RAM and executed from there (the MASD compiler behaves this way).