New HP 15c CE speed tests

[Gene]

Some HP 15c CE speed test results:

1) N-Queens problem
https://www.hpmuseum.org/cgi-sys/cgiwrap...i?read=700

15C - 4750 seconds

DM-15 - not sure which one - 195 seconds

15C LE - 28 seconds

15C CE - 25.5 seconds

So on this test, the CE is 186 X faster than the original 15C and perhaps 5-10% faster than the 15c LE (hard to know if the 28 seconds has any rounding in it).


2) Math and trig results
https://www.hpmuseum.org/speed.htm

15C CE results

Math - RCL 2 yields 5591 divided by HP-9100 result of 679 x 100 = score of 823

Trig - RCL 2 yields 1139 divided by HP-9100 value of 40 x 100 = score of 2847

If I did this correctly these are pretty good scores...


3) The LBL A + GTO A counting speed test

Load the stack with 1, CLx.
LBL A + GTO A
counts in one minute 54,086

For reference, the 15C LE did 49,792 in one minute.


4) Valentin’s program to calculate e to 208 places from his “Long live the HP-15C!” article.

This program takes 22 seconds to compute e to 208 places with the Pause instruction at line 33 was omitted. With the Pause instruction included, the timing was 2 minutes 43 seconds). The numeric results are identical. Timing for an original HP-15c from Valentin’s article for 208 places was 62 minutes and 43 seconds.

[xerxes]

Thank you Gene. I've updated the list.

[EdS2]

Thanks Gene!

Are there good difficult SOLVE or INTEGRATE challenges which could be used as speed (or indeed accuracy) benchmarks?

[Gene]

Every example in the owners handbook and the advanced function handbook were tested and agree to every digit.

Timing remains faster than the 15c LE.

Gene

[J-F Garnier]

(05-03-2023 01:57 PM)Gene Wrote:  ...the CE is 186 X faster than the original 15C and perhaps 5-10% faster than the 15c LE (hard to know if the 28 seconds has any rounding in it).

The new manual is very conservative when it states:

Quote:typical operations are 10–20 times faster than on the HP-15C.

However, the manual then gives more details and clarifications:

Quote:To minimize power consumption, the calculator normally operates in a
low-power mode, with the processor running at a slow speed. When
running user programs (including integration), the calculator switches the
processor to full speed, which increases the demand on the batteries.
However, when the low battery indicator comes on, the calculator will
only run at the slow speed to avoid overloading the batteries

So we can assume the 10x-20x ratio is in slow speed mode.

One of the longest keyboard operations is the calculation of Cy,x combinations.
320 ENTER 160 Cy,x gives 9.5197e94 on my original 15C in about 35s.
On my 15c LE, it takes less than one second.

Another direct keyboard operation test:
8   ENTER   DIM A     ; dim a(8,8)
PI   STO MATRIX A    ; mat a=(pi)
RCL MATRIX A   1/x    ; mat a=inv(a)
Matrix is singular, but the 15c will calculate something anyway

Original 15C: about 60 s
My 15c LE: < 1s

What about the 15c CE ?

J-F

[Gene]

Ok, so I entered this program:

LBL A
50
STO 0
LBL 0
RCL Matrix A
1/x
DSE 0
GTO 0
RTN

Then I pressed 8 ENTER DIM A
PI STO MATRIX A

and did GTO A R/S.

The iterations took about 35.5 seconds total.


Then I changed the LBL 0 of that program to be
320
ENTER
160
g Cy,x

inside the loop. . .

The iterations took just under 18 seconds.

[J-F Garnier]

(05-05-2023 03:01 PM)Gene Wrote:  Ok, so I entered this program:
...

Thanks, but I meant a direct keyboard execution (not a program), to check if I correctly understood the 10x-20x speed ratio mentioned in the manual for "typical operations", that I assume means "keyboard operations".

J-F

[Gene]

Keyboard executions were one second or less.

Since that was difficult to time, I did the program in order to allow a better comparison to the 15c LE.

It would be interesting to see the times of the 15c LE on these programs.

Gene

[johanw]

(05-05-2023 08:48 AM)J-F Garnier Wrote:  One of the longest keyboard operations is the calculation of Cy,x combinations.
320 ENTER 160 Cy,x gives 9.5197e94 on my original 15C in about 35s.
On my 15c LE, it takes less than one second.

On my DM15 it takes about 10 seconds.

[Gene]

On the original 15c, the Cy,x of 320,160 takes about 36 seconds as I just timed it.

On the 15c CE, simply dividing the loop of the calculation which took 18 seconds by 50, you get 0.36 seconds (humorous!)

So on this, the 15c CE is approximately 100X as fast as the original 15C.

Speed varies on different functions of course.

[Eric Rechlin]

If you are putting it in a loop, then you are measuring program execution speed again, which will run at the calculator's full speed. It's hard to test performance of anything running at the slower speed, because a single operation is too quick to get reliable timing, and once you put it in a program to run in a loop then it will run at full speed.

Best way to force the slow speed is to run the batteries down until you get a low battery indicator and then run it in a program, because in low battery mode programs will still run at low speed to avoid overloading the batteries (the current draw of the CPU at 48 MHz exceeds the capability of nearly-used-up batteries to provide current).

[J-F Garnier]

(05-05-2023 03:01 PM)Gene Wrote:  Ok, so I entered this program:

LBL A
50
STO 0
LBL 0
RCL Matrix A
1/x
DSE 0
GTO 0
RTN

Then I pressed 8 ENTER DIM A
PI STO MATRIX A

and did GTO A R/S.

The iterations took about 35.5 seconds total.

Isn't it nice to use a 8x8 matrix without changing the memory configuration, and still being able to write a little program to work on it ?

J-F

[Chumango]

(05-05-2023 08:48 AM)J-F Garnier Wrote:  One of the longest keyboard operations is the calculation of Cy,x combinations.
320 ENTER 160 Cy,x gives 9.5197e94 on my original 15C in about 35s.
On my 15c LE, it takes less than one second.

The 35S takes just under 3 seconds

[Gene]

(05-05-2023 09:44 PM)rprosperi Wrote:  ... which is why the CE stays in low-power mode.

... in keyboard manual mode.

Gene

[rprosperi]

(05-06-2023 01:36 AM)Eric Rechlin Wrote:  

(05-05-2023 09:44 PM)rprosperi Wrote:  [.... wrong advice removed....]

You can actually run the calculator on one battery, but then you have only half the current capacity so you will probably start finding the processor to be short on current well before the low-battery warning.

My bad, thanks for the correction. I'll go delete my post and the portion of yours which quoted mine. Right, they are 3V not 1.5v. Doh!

[Chr Yoko]

(05-05-2023 09:05 PM)Chumango Wrote:  

(05-05-2023 08:48 AM)J-F Garnier Wrote:  One of the longest keyboard operations is the calculation of Cy,x combinations.
320 ENTER 160 Cy,x gives 9.5197e94 on my original 15C in about 35s.
On my 15c LE, it takes less than one second.

The 35S takes just under 3 seconds

And 9 seconds on the DM15L

[Chumango]

And less than 2 seconds for the 42S.

[J-F Garnier]

(05-05-2023 05:27 PM)Gene Wrote:  On the original 15c, the Cy,x of 320,160 takes about 36 seconds as I just timed it.
On the 15c CE, simply dividing the loop of the calculation which took 18 seconds by 50, you get 0.36 seconds (humorous!)

So on this, the 15c CE is approximately 100X as fast as the original 15C.
Speed varies on different functions of course.

On many emulators, the speed ratio doesn't depend much on the type of operations, and we can estimate the speed ratio with any operation, such as a counting loop.

But the Coconut CPU is a special case, because the original CPU is a bit-serial processor that fetches and executes an instruction every 56-clock cycle (1 word time).
It doesn't depend if the instruction is simple, such as setting a CPU flag, or complex such as adding two 56-bit BCD numbers.
If the emulator is designed for maximum speed efficiency and doesn't try to mimic the CPU word time, it will execute simple instructions much faster than the complex ones.

So benchmarks that involve a lot of BCD math will have a lower speed ratio than others that are, for instance, just moving data around.

J-F

[paf]

(05-10-2023 12:33 PM)Chr Yoko Wrote:  And 9 seconds on the DM15L

In 12MHz mode ? With the 48MHz mode activated, it is about 2 seconds.