How fast is the HP Prime?

[jebem]

This interesting video compares a bunch of calculators resolving a "simple" definite integral, and it seems that the HP Prime leaves all of them eating dust!
This benchmark included HP, TI and Casio.

The integral is: int(e^(x^3),x,0,6)
The Prime comes with the answer instantaneously (Ans=5.96393809189E91)

When we look to the venerable and powerful HP-50G, we understand the costs of Saturn emulation... more than 60 seconds to compute the same integral!

http://www.youtube.com/watch?v=DHRsvSTGiBc

[ArielPalazzesi]

(03-09-2014 07:17 PM)jebem Wrote:  When we look to the venerable and powerful HP-50G, we understand the costs of Saturn emulation... more than 60 seconds to compute the same integral!

Awesome!!!

[eried]

Excelent! Haven't completed the video but it seems very in depth as usual, nice!.

[HP67]

(03-09-2014 07:17 PM)jebem Wrote:  When we look to the venerable and powerful HP-50G, we understand the costs of Saturn emulation... more than 60 seconds to compute the same integral!

I didn't bother looking at the video but from your description this is not an apples-and-apples comparison. There is nothing to conclude from it except relative device speeds, all things being unequal.

The 50g is Saturn emulated on 75 MHz ARM. The only way to truly understand the cost of emulation in the 50g is to fab a 75 MHz Saturn processor and run them side by side. Otherwise, you're comparing a 400 MHz ARM device running native code against who knows what. Did you not expect it would be faster even against a 75 MHz device running native code?

Did I miss something here?

[Han]

Then there's also the difference in speed between algorithms (are we sure it's the same algorithm in both calculators) as well as speed differences between the languages implemented.

[eried]

(03-10-2014 03:07 PM)HP67 Wrote:  The 50g is Saturn emulated on 75 MHz ARM. The only way to truly understand the cost of emulation in the 50g is to fab a 75 MHz Saturn processor and run them side by side. Otherwise, you're comparing a 400 MHz ARM device running native code against who knows what. Did you not expect it would be faster even against a 75 MHz device running native code?

Did I miss something here?

You are missing that jebem is testing how the device is "presented" to the user, not how it can be 'hacked' to run faster.

In the other hand, I think integration is the poorest thing implemented on the 50g, I always had troubles with EXP(X... integrations. It seems that the method used by casio and TI implements some pre-catched results (because I often noticed that the results were a bit approximated in the other devices, I am not sure if this is the case exposed in the video). In my opinion, an implementation of an algorithm like Riemann sum in the devices would represent a more truly comparison, but we will see that soon in the next video with n-queens.

[HP67]

(03-10-2014 05:21 PM)eried Wrote:  You are missing that jebem is testing how the device is "presented" to the user, not how it can be 'hacked' to run faster.

I don't know about anybody else, but I don't need a "benchmark video" to prove a 400 MHz ARM device outperforms a 75 MHz ARM device even if all other factors were identical!

How much more so when they aren't

[jebem]

(03-10-2014 03:07 PM)HP67 Wrote:  

(03-09-2014 07:17 PM)jebem Wrote:  When we look to the venerable and powerful HP-50G, we understand the costs of Saturn emulation... more than 60 seconds to compute the same integral!

I didn't bother looking at the video but from your description this is not an apples-and-apples comparison. There is nothing to conclude from it except relative device speeds, all things being unequal.

The 50g is Saturn emulated on 75 MHz ARM. The only way to truly understand the cost of emulation in the 50g is to fab a 75 MHz Saturn processor and run them side by side. Otherwise, you're comparing a 400 MHz ARM device running native code against who knows what. Did you not expect it would be faster even against a 75 MHz device running native code?

Did I miss something here?

1) Hi, HP67! I understand your point.
Of course HP Prime is faster! That is exactly my point: HP has done an excellent choice of hardware and software components for the Prime, and the benchmark just prove it!
Otherwise, the benchmark results would be poorer for the HP-Prime, even running at 400MHz.

And please let's not generalize and jump to conclusions: this benchmark is only valid for the above equation, nothing else.

2) Concerning your observation on the costs of emulation: I agree with you, of course. We cannot compare apples to apples here, as we do not have a native 75MHz Saturn processor.

However, let me do some basic analysis here:
I have run the benchmarks myself on my HP-50G and HP-Prime, after seeing the video:

- Let us say the HP Prime takes 0.1 Sec to come out with the result (and I can guarantee it is taking less than that: the answer is instantaneous. Try it yourself to check it).
- The HP-50G takes about 65 sec.

- Calculation speed ratio of: 65/0.1 = 650 times faster for the Prime.
- Raw clock speed ratio: 400 MHz / 75 MHz = 5.3 faster clock in the Prime.
So, for a 5 times faster clock, the Prime is 650 times faster.

Analysis:
The poor HP-50G performance root cause is not the lower clock speed.
Make your choice(s) for the root causes:
- Saturn emulation.
- Inefficient Saturn cpu architecture;
- Poor choice the host ARM processor for the job;
- Inefficient HP-49 series algorithms when compared with the ones implemented in the Prime.

[CR Haeger]

This analysis is interesting but to me is only part of the "speed" of a calculator.

If the device is easy to learn, quick and error resistant to setup and execute problem(s), then it's fast in my book. I think the Prime has the potential to improve in these areas as well. If it does I think it may become indispensable for students and professionals alike.

A TI-36X I have messed with runs this in 1'25". It's slow for THIS problem but it's really fast and easy to use for non CAS, non graphing tasks.

[HP67]

(03-10-2014 07:34 PM)jebem Wrote:  2) Concerning your observation on the costs of emulation: I agree with you, of course. We cannot compare apples to apples here, as we do not have a native 75MHz Saturn processor.

Yes, but this goes directly against what you said above, which is what I quoted and objected to.

(03-10-2014 07:34 PM)jebem Wrote:  - Let us say the HP Prime takes 0.1 Sec to come out with the result (and I can guarantee it is taking less than that: the answer is instantaneous. Try it yourself to check it).
- The HP-50G takes about 65 sec.

Why aren't you timing this on the calculators? We all know the 50g has a bad clock bug but surely checking the TICKS and getting the real time is better than guessing with a stopwatch!
Until then, the rest of your numbers aren't ready for use in drawing any conclusions.

(03-10-2014 07:34 PM)jebem Wrote:  - Calculation speed ratio of: 65/0.1 = 650 times faster for the Prime.
- Raw clock speed ratio: 400 MHz / 75 MHz = 5.3 faster clock in the Prime.
So, for a 5 times faster clock, the Prime is 650 times faster.

Analysis:
The poor HP-50G performance root cause is not the lower clock speed.
Make your choice(s) for the root causes:
- Saturn emulation.
- Inefficient Saturn cpu architecture;
- Poor choice the host ARM processor for the job;
- Inefficient HP-49 series algorithms when compared with the ones implemented in the Prime.

This is the point. There are many factors here and without isolating them you cannot call this drag race a benchmark. From your SOP analysis that the Prime is 130X faster, the first thing that comes to mind is memory or memory controller deficiencies, then differences in algorithms, etc. That said, I have seen emulators degrade performance tremendously so I don't rule that out either. But we cannot conclude anything yet until we isolate all the variables. That was all I have been trying to say.

[Chris Pem10]

I've got an old Pentium 90 and a Core i5 overclocked at 4.8 GHz. Regardless; I can still only type about 20 wpm on each... because the interface never changed. My results are 99% determined by how fast I can find and press the correct button.

Where the Prime fell short in my expectations was in the interface: it has a full-color touchscreen that is under-utilized. It's 2014: any physical keyboard on a handheld device is obsolete. I wanted the ability to not just make user-defined keys... I wanted user-defined on-screen keyboards. I wanted to press one key and launch "calculus keyboard mode" or another button for "unit conversion keyboard". Would have been cool if a teacher could "lock-in" a custom keyboard of functions they wanted to allow for a particular exam. And every time I swipe the screen to the left, I get a new calculator instance for "parallel calculations"... so if I'm stuck on exam problem #2 I can just swipe a new calculator instance and come back to problem #2 later without having to re-enter everything.

If someone would re-engineer the HP48 android emulator and, instead of using the stock HP48 keyboard layout, create a custom keybaord that dynamically covered the entire screen real estate of a 10" tablet, that would be worth paying for... even if it could "only" perform the same functions as a 25 year old calc.