Is the Prime the fastest calculator ever made?

[zeno333]

...if not, what would be faster..(I refer to raw number crunching...)

[AlexFekken]

Yes of course!

(he said, hoping that no one would prove him wrong)

[cyrille de brébisson]

Hello,

It definitely has one of the fastest CPUs around.

Is it the fastest is a different question. The fastest doing what?

In raw number crunching, discarding user program management (such as when you do a large amtrix multiplication or inversion), it probably is the fastest.

In other applications, it might not be as fast because it's breath of possible options might cause some overhead that take away at the raw speed.

Cyrille

[pier4r]

Search for calculator benchmark. On the wiki4hp there are some and also on the articles of the mohpc. I'm from mobile so it is clumsy to link .

From memory I'd say that the HP 50g using hpgcc or arm assembly is the fastest. But the point is usability in a calculator, if a calculator is super fast but does only the four operations it does not help much.

[compsystems]

A comparison

https://tiplanet.org/forum/viewtopic.php...39#p224539

[salvomic]

(12-20-2017 06:10 AM)cyrille de brébisson Wrote:  It definitely has one of the fastest CPUs around.

Is it the fastest is a different question. The fastest doing what?
...
Cyrille

yes!
and even if it wouldn't the fastest, definitely it is really one of the more complete calculator at present, and that, for me, is already enough

If only it had also Longfloat library and user-friendly soft keys... But, who know? in a next release of the FW... Santa or not Santa...

Salvo

[Claudio L.]

(12-20-2017 05:06 AM)zeno333 Wrote:  ...if not, what would be faster..(I refer to raw number crunching...)

The next calculator made will be the fastest ever made. Every new product will likely be faster than previous generations.

(12-20-2017 06:10 AM)cyrille de brébisson Wrote:  In raw number crunching, discarding user program management (such as when you do a large amtrix multiplication or inversion), it probably is the fastest.
Cyrille

I doubt it. The Cortex M4 in the Numworks has an FPU, so if raw number crunching is the measurement, I don't think the Prime can even get close.

(12-20-2017 06:51 AM)pier4r Wrote:  But the point is usability in a calculator, if a calculator is super fast but does only the four operations it does not help much.

I agree, the question "is the fastest ever made?" should be answered with a "who cares?".

(12-20-2017 02:37 PM)compsystems Wrote:  A comparison

https://tiplanet.org/forum/viewtopic.php...39#p224539

Python is probably a slower interpreter than PPL, the Cortex M4 is probably more than twice as fast (at same clock) than the ARM9 in the Prime, without even counting the FPU, so that difference from 100 MHz to 400 MHz doesn't mean much these days. If the code takes advantage of the FPU, it's game over for the Prime.

[smp]

(12-20-2017 03:34 PM)Claudio L. Wrote:  ... the Cortex M4 is probably more than twice as fast (at same clock) than the ARM9 in the Prime, without even counting the FPU, so that difference from 100 MHz to 400 MHz doesn't mean much these days.

Claudio, you mentioned twice that the Numworks calculator has a Cortex M4 processor in it. When I look at the Numworks site, their specs say it has an 100 MHz ARM v7 processor.

Is ARM v7 the same as Cortex M4, or am I confused?

If the Numworks is faster, I am looking for the fastest handheld platform for my 8080 simulator.

Thanks,
smp

[pier4r]

(12-20-2017 02:37 PM)compsystems Wrote:  A comparison

https://tiplanet.org/forum/viewtopic.php...39#p224539

Hmm, that is a nice idea. Instead of the 8 queen problem one can have a sort of iterative/recursive function that is actually a bit more intensive (than only integers) and then compare. So a fractal is a nice idea.

(12-20-2017 03:34 PM)Claudio L. Wrote:  I agree, the question "is the fastest ever made?" should be answered with a "who cares?".

Aside form the OP and some others, not many. But anyway any thread has a possibility to give nice inputs. For example your info from the Numworks CPU.

[Claudio L.]

(12-20-2017 04:52 PM)smp Wrote:  Claudio, you mentioned twice that the Numworks calculator has a Cortex M4 processor in it. When I look at the Numworks site, their specs say it has an 100 MHz ARM v7 processor.

Is ARM v7 the same as Cortex M4, or am I confused?

If the Numworks is faster, I am looking for the fastest handheld platform for my 8080 simulator.

Thanks,
smp

ARMv7 is the architecture version, while Cortex M4 is the CPU name. Many different CPUs can implement the same architecture ARMv7, but they differ in other things they include in the dice. Look at the black diagram on the right here to get an idea of what's included in a Cortex M4 processor, the ARMv7 core is one of many things.

EDIT: I don't know if the Numworks will be faster doing emulation. If you do numeric calculations using the FPU it is, but an emulator I don't know if it will be faster.

[smp]

(12-20-2017 07:34 PM)Claudio L. Wrote:  ARMv7 is the architecture version, while Cortex M4 is the CPU name. Many different CPUs can implement the same architecture ARMv7, but they differ in other things they include in the dice. Look at the black diagram on the right here to get an idea of what's included in a Cortex M4 processor, the ARMv7 core is one of many things.

EDIT: I don't know if the Numworks will be faster doing emulation. If you do numeric calculations using the FPU it is, but an emulator I don't know if it will be faster.

Thanks very much for the explanation, Claudio. I was confused, and I'm glad I to get the the answer.

smp

[Tim Wessman]

That "python" comparison on the TI planet site isn't really python at all on the HP side but symbolic calculation in the CAS. It just is a layer that converts python like syntax into the underlying CAS syntax. So it still ends up doing calculations through the CAS code which is considerably slower then the normal calculator code which I suspect would be faster then the python on the numworks.

Anyone want to make some code and time it?

Again though, the "who cares" point has probably been reached...

[Arno K]

(12-20-2017 10:38 PM)Tim Wessman Wrote:  Again though, the "who cares" point has probably been reached...

as it feels to be the quickest calculator available
Arno

[cyrille de brébisson]

Hello

(12-20-2017 03:34 PM)Claudio L. Wrote:  

(12-20-2017 06:10 AM)cyrille de brébisson Wrote:  In raw number crunching, discarding user program management (such as when you do a large amtrix multiplication or inversion), it probably is the fastest.
Cyrille

I doubt it. The Cortex M4 in the Numworks has an FPU, so if raw number crunching is the measurement, I don't think the Prime can even get close.

I was thinking raw decimal float number crunching...

Numworks CPU is a newer generation, but the generation gain is far from compensing for the clock speed difference. The latest 12C does also have a cortex M4 and the speed gain was around 1.3 times compared with the ARM7 based one.

Also, the cortex FPU only supported 32 bits floats. Which is very limited for doing calculator math.

Cyrille

[smp]

(12-20-2017 10:38 PM)Tim Wessman Wrote:  That "python" comparison on the TI planet site isn't really python at all on the HP side but symbolic calculation in the CAS. It just is a layer that converts python like syntax into the underlying CAS syntax. So it still ends up doing calculations through the CAS code which is considerably slower then the normal calculator code which I suspect would be faster then the python on the numworks...

Thanks very much for that explanation, TIM! I'm not a CAS user, so I don't know how to distinguish stuff like that.

smp

[parisse]

The CAS code for the Mandelbrot is about 2* slower than the equivalent Home code, and the Home code is a little bit slower than the Numworks micro-Python code despite the fact that the Prime processor is faster. The reason are
* Home is using decimal floats, they are of course slower than CPU binary floats
* I have never optimized the CAS interpreter before, because it's better to optimize built-in commands. If you are writing a CAS program, you do that to take advantage of all the CAS commands, and the running cost of an interesting CAS program is almost certainly in the CAS commands you called, not in the interpretation itself.
For example if you stay with fractals, I have proposed on tiplanet a Newton fractal code using some CAS builtin commands, and nobody proposed an equivalent for the Numworks.
Since these kinds of benchmarks may be misused (despite the fact that a calculator is not really an exciting hardware to run fractals), I have decided to optimize the giac interpreter. It's too late for the beta HP Prime firmware, but you can try some fractals in your browser here

[Stevetuc]

(12-21-2017 06:07 AM)cyrille de brébisson Wrote:  ...
Numworks CPU is a newer generation, but the generation gain is far from compensing for the clock speed difference. The latest 12C does also have a cortex M4 and the speed gain was around 1.3 times compared with the ARM7 based one.

Also, the cortex FPU only supported 32 bits floats. Which is very limited for doing calculator math.

Cyrille

Some interesting info on 32 bit floats here:
http://www.ilikebigbits.com/blog/2017/6/...-or-double

The link to the benchmark for Cortex is broken but can be reached via Wayback:
https://web.archive.org/web/201606080657...rit_speed/

Although the Cortex achieves a very large speed increase for 32bit floats versus doubles, it seems the processor used in the Prime might achieve a factor of 2 increase if there was a mode to use this for math. Agreed its limited precision but might be useful under some circumstances.

[pier4r]

So, it always depends on what activity one performs with the tool (a calculator in this case). Though for the little summation test setup in this thread: http://www.hpmuseum.org/forum/thread-9750.html so far the prime is impressive!

The newRPL and the dm42 are impressive as well.

Surely I think that with some optimizations (es: nspire running lua, hpgcc on the 50g and so on) one could beat the prime, but realistically in all day usage one use the common functions, not the optimized ones. Therefore for common usage the prime is really fast so far.