Summation based benchmark for calculators

[Helix]

Sharp PC-1401 Basic
n=10 -> 9s result: 13.71183501
n=100 -> 91s result: 139.2971873

Sharp PC-G850VS Basic
n=100 -> 11s result: 139.2971873
n=1000 -> 102s result: 1395.346559

HP 50g HPGCC 2.0 - 75 Mhz
n=10000 -> 2s result: 13955.85790429154
n=100000 -> 23s result: 139560.8013952589

My comments:
I think the Sharp PC-1401 was the first model of this line combining a scientific calculator and a basic programmable computer, with a QWERTY keyboard. The PC-G850Vs is the very last one in this line. So it's interesting to see the evolution of speed between them.
The Sharp PC-G850VS is also programmable in C, but I've not tried this possibility. We lack a good English documentation.

HPGCC3 should be even faster than HPGCC 2.0, because it can switch the processor frequency up to 200 MHz. However I don't use it, because its installation and use are way too complex for me.

[pier4r]

added! Thanks for the report.

At first I thought "hmm, the sharp did not really improve from the first system of the line to the last". I did not see that the iterations where 10 times larger.

The 50g with HPGCC2 is a little beast.

[polbit]

Another HP 71b score, based on the program given on this thread before (and CPU freq of 655Hz!): 171s. Please correct this egregious error immediately!

[brickviking]

Here's the last results from me for a while, this should add to my previous results. First, the 9750gII upgraded to 9860gII OS running at normal speed:

• {sum:eqn,x,1,10000} gives me 13955.8579 in 3m19s (199 seconds). This seems to be faster than the previously mentioned 9860gII which was about 233 seconds.
• {sum:eqn,x,1, 100000} gives me 139560.9761 in 33m29.5s (2009.5 seconds).
  

These two results feature the 9750gII with the PLL jacked up to 471MHz and the CPU frequency sitting at 235 MHz (F5 in FTune):

• {sum:eqn,x,1,10000} gives me 13955.8579 in 32 seconds.
• {sum:eqn,x,1,100000} gives me 139560.9761 in 5m23s, or 323 seconds.
  

I can certainly conclude that these two results compare much more favourably with other calculators.

BrickEdit: modified to add CPU frequency for the souped-up PLL.
(Post 165)

[pier4r]

(02-03-2018 01:40 AM)polbit Wrote:  Another HP 71b score, based on the program given on this thread before (and CPU freq of 655Hz!): 171s. Please correct this egregious error immediately!

This for 1000 iterations?

edit: brickwiking. Are you sure that FTtune brings the cpu to 471mhz? I know from previous searches that the 98xx and 97xx of the casio have the cpu (SH3 IIRC) around 30 mhz, when they are pushed this varlue should reach a mximum of ~120 hmht. 470 should be really high.

I'add those results asap, thanks!

[toml_12953]

(01-29-2018 04:44 PM)Helix Wrote:  Sharp PC-1401 Basic
n=10 -> 9s result: 13.71183501
n=100 -> 91s result: 139.2971873

Sharp PC-G850VS Basic
n=100 -> 11s result: 139.2971873
n=1000 -> 102s result: 1395.346559

HP 50g HPGCC 2.0 - 75 Mhz
n=10000 -> 2s result: 13955.85790429154
n=100000 -> 23s result: 139560.8013952589

When running benchmarks, you should normalize the results to a standard. Either make n the same for all calculators or if the results are too small to measure, increase the number of trials and then divide the results to be the equivalent. In the case above, you should divide the G850VS results by 10 and the 50g results by 1000 so they can both be directly compared to the 1401.

[pier4r]

I guess the benchmark results are clear enough that normalization is not needed. Those results go to compete with other calculators in the same category (defined by the max x)

[pier4r]

results added. Please double check.

Impressive FTune although I do not know how long the system will last under overclock.
http://pm.matrix.jp/ftune2e.html

Surely it is pretty impressive.

[polbit]

(02-05-2018 12:52 PM)pier4r Wrote:  

(02-03-2018 01:40 AM)polbit Wrote:  Another HP 71b score, based on the program given on this thread before (and CPU freq of 655Hz!): 171s. Please correct this egregious error immediately!

This for 1000 iterations?

Yes it is.

[brickviking]

(02-06-2018 08:35 PM)pier4r Wrote:  results added. Please double check.

Impressive FTune although I do not know how long the system will last under overclock.
http://pm.matrix.jp/ftune2e.html

Surely it is pretty impressive.

There's the obvious battery life reduction under overclock, but for most of these tests I used the calculator while it was plugged into the USB, so battery life wasn't adversely affected. I also noticed that overclocking this calculator in particular (can't do it with the 9750g+, no support for external plugins) led to a better response time and screen redraw time. It's all a balancing act: can I tolerate the speed I'm running at for its resultant battery life? This really applies to I/O-heavy or CPU-heavy programs, and not so much with light usage such as "waiting for slow user to respond".

Still, it's nice to be able to pull out these sorts of figures against HP calculators on arbitrary benchmarks.

(Post 170)

[BartDB]

Casio fx-2700P

n=10
t~16s
result=13.711835

n=100
t~158s
result=139.29719
.

[toml_12953]

(02-06-2018 12:46 PM)pier4r Wrote:  I guess the benchmark results are clear enough that normalization is not needed. Those results go to compete with other calculators in the same category (defined by the max x)

Quite a few people have misinterpreted the data at first look. That says the results should have been normalized.

[pier4r]

(02-25-2018 11:47 PM)toml_12953 Wrote:  Quite a few people have misinterpreted the data at first look.

Well a first look can be misleading, it is not really a strong justification. If people would be puzzled even after an explanation, then maybe something should change.

We are talking about a simple benchmark with categories here, nothing difficult.

[Michael de Estrada]

I realize that this is not exactly a calculator per se, but I tried Free42 on my iPhone SE with iOS 11.

n = 100,000 Time = 9.19 seconds Sum = 139,560.976141

I used the same code that I used to benchmark the DM42.

[pier4r]

No problem, comparisons are never bad.

100k takes 9 seconds, on an iphone?
Which one? I mean which model. 6,7,8,10 ? (SE is the form factor, right?)

I mean that would be interesting. I know from other threads that the free42 is quite optimized, so having the prime "only" 100% slower is quite impressive.

[Michael de Estrada]

(03-31-2018 12:11 AM)pier4r Wrote:  100k takes 9 seconds, on an iphone?
Which one? I mean which model. 6,7,8,10 ? (SE is the form factor, right?)

The SE is a separate model that has the 4” screen size of the old iPhone 5, but the A9 processor of the 6s. It is currently the lowest priced model in the Apple iPhone lineup, or about a third the cost of the iPhone X.

Also, I changed the angular mode to radians, and the time dropped to 7.92 seconds.

An alternative code by grsbanks that uses the stack instead of saving to named variables takes 7.87 seconds.

[pier4r]

Updated (some of the last results were missing)

http://www.hpmuseum.org/forum/thread-975...l#pid86751

http://www.wiki4hp.com/doku.php?id=bench...g_exp_root

This remembers me that I have to update the "most common calculators in the forum" results since months. Good that the data is there for the moment, because I cannot allocate time to it so far.

[Michael de Estrada]

(03-31-2018 10:01 AM)pier4r Wrote:  Updated (some of the last results were missing)

http://www.hpmuseum.org/forum/thread-975...l#pid86751

http://www.wiki4hp.com/doku.php?id=bench...g_exp_root

Correction to your postings above, it should say “iPhone SE and 6s,” not “iPhone 6 SE.” The SE and 6s are different models that use the same A9 cpu. The 6 is an older model that uses a slower A8 cpu.

[pier4r]

Added the casio fx C.Basic entries. I am not sure how versatile is the environment, but it is pretty quick.

[pier4r]

Can someone test the prime G2 with the benchmark in the first page?

(and whatever other calculator is welcomed. The more the data, the better)