11 replies to this topic

[Hlib2]

Recently, just for fun, I decided to compare the characteristics of three calculators and chose: 1) fx-9750g plus - the most economical in its class; 2) fx-9750gii (SH3, V.02.00) - they say that this one is very fast because of the well-optimized OS; 3) fx-9860gii (SH4, V.02.09) - equipped with newer firmware and newer processor. All three devices include monochrome LCD 128×64 pxl, very similar and compatible versions of the programming language.
Since the fx-9850gb plus (BE) became my first full-fledged graphical calculator from CASIO in 1998, I have been watching with interest the development of this (or like this) line for about 20 years.
For a more objective assessment of the speed of calculations, I took two different programs. The first of them is the well-known 8-queens benchmark. The code with the List variable was selected as less effective for the fx-9750gii/9860gii.

0➝A~Z:8➝R
R➝Dim List 1
Lbl 0:X=R⇒Goto 4
Isz X:R➝List 1[X]
Lbl 1:Isz S:X➝Y
Lbl 2:Dsz Y Deg
Y=0⇒Goto 0
List 1[X]-List 1[Y]➝T
T=0⇒Goto 3
X-Y≠Abs T⇒Goto 2
Lbl 3:List 1[X]-1➝List 1[X]
List 1[X]⇒Goto 1
Dsz X:Goto 3
Lbl 4:S

The second one solves a fun problem: how many integers (from 1 to 9999) are evenly divisible by the number of letters in their name? First one is 4 (divisible by f-o-u-r (4 letters): 4÷4=1). Last one happens to be 9999 (divisible by n-i-n-e-t-h-o-u-s-a-n-d-n-i-n-e-h-u-n-d-r-e-d-n-i-n-e-t-y-n-i-n-e (33 letters): 9999÷33=303).

[[0,3,3,5,4,4,3,5,5,4]
[3,6,6,8,8,7,7,9,8,8]
[0,0,6,6,5,5,5,7,6,6]]→Mat K
-1→N:0→K
For 1→E To 10:Mat K[1,E]+8(E≠1)→S
For 1→D To 10:Mat K[1,D]+7(D≠1)+S→T
For 1→C To 10:Mat K[3,C]+T→U
(C=2)+1→B
For 1→A To 10:Mat K[B,A]+U→V
N+1→N:V≠0⇒(Frac (N/V)=0)+K→K
Next:Next:Next:Next:"K=":K

spoiler: K=359.
This code contains the operations For...Next and Mat_ , which fx-9750gii/9860gii perform very quickly.
The results are shown in a table from my paper notebook. The measurement and rounding error does not exceed 2%.

When comparing the results obtained from 9750gii and 9860gii, an unexpected conclusion arises: the speed of the calculator is directly proportional to its power consumption. And if I didn`t know anything about their hardware stuffing, I would say based on these tests: the fx-9750gii is a 56 MHz overclocked fx-9860gii. No special technological advantages of the 9860gii calculator compared to the 9750gii have been observed. A similar proportionality between speed and power is seen when comparing results between fx-9750g plus and 9750gii/9860gii.
Probably, the CPU clock speed can not serve as a true characteristic of the calculator, since the manufacturer also regulates the speed in other ways.
And brief conclusions that I made for myself: 1) such a calculator as 9750g plus is still relevant today; 2) I will not upgrade the 9750gii. Even with the old OS 02.00, it is an excellent calculator; 3) there are no complaints about 9860gii.

Edited by Hlib2, 20 June 2020 - 12:41 PM.

[sentaro21]

Hi HliB2,

 

Thanks a lot for the very interesting calculator comparison!

I'm very interested in these comparisons.

When I get a new calculator, the first thing I do is benchmark it.

 

I think CASIO does a very good job of ranking their calculators, 

but I didn't like the fact that they deliberately slowed it down when I switched from SH3 to SH4A.

Half power consumption and half speed is useless.

Half the power consumption and double the speed is the right evolution.

[Hlib2]

@ sentaro 21
I also check all my calculators in benchmarks, just like you. I test all the graphical calculators I have at 25 technical parameters in order to use them most effectively in specific tasks.
I don`t understand, for example, why TI-85 (6MHz, 1993) with a lower maximum power consumption is faster than FX-9860gii-2 (29MHz , 2012) in operations with lists. TI-85 is only 1.3 times slower than gii-2 in speed when performing vector operations.
TI-83+SE (15MHz, 2001) is the same as 9860gii-2 (29MHz, 2012) and 9750gii (29MHz, 2009) in the speed of graphical constructions, but it is ahead of them with lists, although it loses to them in operations with matrices by 1.4 ... 2.9 times.
Thank you for C.BASIC, which demonstrated the speed at which modern CASIO calculators with an interpreted programming language could theoretically work.

Edited by Hlib2, 16 February 2021 - 08:36 PM.

[MJim]

I miss my old fx-9750GII SH3 version.  It was pretty beaten up when I got it, but I still feel that the display was better than the near new fx-9750GII SH4 version I got to replace it.  It also had some crazy good quality Flash memory rated at 1,000,000 cycles and 20 years data retention.

 

I just recently got a Classpad 300+ which uses the SH3 processor, and it does feel pretty slow compared to the fx-9750GII/fx-CG50 I have.  Interestingly I discovered during a teardown that the Classpad 300+ & Classpad 330 which are both SH3 and use the same firmware happen to have different flash memory chips.  My Classpad 300+ has a 2MB (whoops 2Mbit so 256KB) + 16MB flash chip, while the 330 has a 1MB + 32MB flash chip...so I guess the 330 never makes use of that extra flash storage.  The 1MB flash chip on the 330 is 1,000,000 cycle @ 20 years, so the 330 seems to have better hardware than the 300+ even though they are both SH3's running the same firmware.

 

That comparison you did with the TI-85 reminds me of the fx-3600pv from 1989/90 where the integration is nearly the same speed as my fx-991MS Plus 2nd edition from 2018 (despite both using the same Simpson's method of integration).  I wonder if there is some sort of emulation engine running, though I'm not sure whether the SH4 is clock for clock faster than the SH3 (I would expect so, but perhaps it focused on power savings?).

 

That fx-9750G+ comparison is interesting as well; seems like it was a much more efficient calculator.  I only just sold a fx-9750G+ that I repaired, but I wonder if the new display on the fx-9750GA changes the power consumption much.

Edited by MJim, 19 February 2021 - 02:10 AM.

[Hlib2]

@ MJim

I miss my old fx-9750GII SH3 version.  It was pretty beaten up when I got it, but I still feel that the display was better than the near new fx-9750GII SH4 version I got to replace it.  It also had some crazy good quality Flash memory rated at 1,000,000 cycles and 20 years data retention.

The display in the FX-9750gii SH3 is very good, but not the best. The most high-contrast and quality monochrome displays in calculators (also without parasitic stripes at a high level of contrast) I only saw were in the afx-2.0 and ti-83+ in the very first batches (1999-2000), when they were produced in Malaysia and Taiwan, respectively.
The FX-9750GII SH3 has excellent rubber in the keyboard and sufficiently good paint in the lettering on the keys and above them. In terms of the production quality, I put it on a par with such masterpieces as TI-83+ (pre-2001), AFX-2.0 (pre-2000), HP-48GX (some releases), FX-9750GII SH3, and FX-9860GII SH4 (very first release).
As for saving RAM to flash memory every time you turn off the calculator, with CASIO gii, you don`t have to worry about it, as my experience suggests.

That comparison you did with the TI-85 reminds me of the fx-3600pv from 1989/90 where the integration is nearly the same speed as my fx-991MS Plus 2nd edition from 2018 (despite both using the same Simpson`s method of integration).  I wonder if there is some sort of emulation engine running, though I`m not sure whether the SH4 is clock for clock faster than the SH3 (I would expect so, but perhaps it focused on power savings?).

I will also note that the ti-85 still surpasses the modern CASIO calculators of the gii/giii series in terms of programming language flexibility (although, it loses out to them in the number of functions with String variables). MyList(MyVar)➝ListABC(myMatrix(1,2)+iPart (Procedure1)): in my opinion, CASIO calculators still do not know how to do this, unlike ti-85 (1993)

That fx-9750G+ comparison is interesting as well; seems like it was a much more efficient calculator.  I only just sold a fx-9750G+ that I repaired, but I wonder if the new display on the fx-9750GA changes the power consumption much.

The fx-9750g+ is one of the most effective calculators in characterizing the ratio of the number of calculations per energy unit consumption. If the cut-off threshold in it was no more than 4 volts (as in the gii series), then it would be the longest-playing graphical calculator in the world. The display here does not make any changes for the power consumption. But for example: the fx-9850gb+ calculator, which is identical to 9750g+ in technology, has 1.4 times more battery consumption even in programs that do not use graphics (LCD, obviously, can be excluded here).

[MJim]

The display in the FX-9750gii SH3 is very good, but not the best. The most high-contrast and quality monochrome displays in calculators (also without parasitic stripes at a high level of contrast) I only saw were in the afx-2.0 and ti-83+ in the very first batches (1999-2000), when they were produced in Malaysia and Taiwan

 

Those stripes seem to be the bane of not just dot-matrix calculators, but dot matrix displays in general.  The 9750GA has a better display then the fx-9750GII SH3 (9750GA>9750GII SH3>9750GII SH4), but has problems with the contrast stripes as well.

 

As for saving RAM to flash memory every time you turn off the calculator, with CASIO gii, you don`t have to worry about it, as my experience suggests.

 

Unfortunately it's a quirk of mine ever since I started using flash devices in the early 2000's; reliability is always important and it isn't just the number of cycles, but the unpowered data retention lifespan.  Most modern high density flash devices (SATA SSD's for example) rate their cycle life for 1 year unpowered data retention, so theoretically you could end up with corrupted data on a SSD if you leave it disconnected long enough as it is approaching it's rated lifespan.  I worry that newer calculators may start using this much higher density cheaper flash that will have their OS's becoming corrupted after a decent amount of time in storage.  Perhaps I'm a little fanatical about this (I've even set my auto-off to the longest amount of time of 60 minutes), but good quality flash with long endurance and data retention is something I look for regardless

 

The fx-9750g+ is one of the most effective calculators in characterizing the ratio of the number of calculations per energy unit consumption. If the cut-off threshold in it was no more than 4 volts (as in the gii series), then it would be the longest-playing graphical calculator in the world. The display here does not make any changes for the power consumption. But for example: the fx-9850gb+ calculator, which is identical to 9750g+ in technology, has 1.4 times more battery consumption even in programs that do not use graphics (LCD, obviously, can be excluded here).

 

I'll have to do some power measurements sometime on the fx-9750GA to see how it compares, but I do remember having to do a double take when I spotted the power consumption figure on the back (0.06W).  Do you know why the power consumption is more on the fx-9850GB+ (excluding the LCD consumption) over the 9750G+?

 

UPDATE:

Unfortunately I'm not particularly good at programming and so I couldn't get that nqueens program to work on the fx-9750GA.  In any case the average current draw when building a table or running the HP summation benchmark (https://www.hpmuseum...hread-9750.html) for about 5 minutes is about 4mA @ 5.2v so around 21mW.  Interestingly the idle power consumption is a bit lower on the fx-9750GII SH4 (3.7mW vs 6.8mW), but when put to work the fx-9750GA is rather low power.

 

The ClassPad 300+ is a bit of a monster, ~ 280mW peak (53mA @ ~5.2V) while symbolically integrating.  Idle (in menu screen) power consumption was much better @ 15.7mW (2.75mA @ ~5.7V).

 

I'm pretty sure that the ClassPad, like the fx-9750GII & fx-9860GII uses SRAM instead of DRAM, which drastically lowers the power consumption while off as compared to the fx-CG50 (after an hour being off, I believe it stops refreshing the DRAM and goes into a deeper sleep, while on the fx-9750GII/9860GII it doesn't really need to).

Edited by MJim, 20 February 2021 - 11:46 AM.

[Hlib2]

MJim

... Most modern high density flash devices (SATA SSD`s for example) rate their cycle life for 1 year unpowered data retention, so theoretically you could end up with corrupted data on a SSD if you leave it disconnected long enough as it is approaching it`s rated lifespan. I worry that newer calculators may start using this much higher density cheaper flash that will have their OS`s becoming corrupted after a decent amount of time in storage ...

I think your concerns due to CASIO are well-founded here. Currently, manufacturers of cheap gadgets reduce the cost of the electronic components used as much as possible and limit the resource of their products so that we constantly spend money. In the old manuals for CASIO graphic calculators, in particular, it was indicated: the service life of the main batteries in the OFF mode is 2...3 years, the service life of the CR2032 backup cell in the absence of the main batteries is 6 months.
Currently, despite the advances in technology, in the OFF mode, gii (64kB RAM) consumes three times more microamps than, for example, afx (146kB user RAM). Also, in modern manuals, the battery life in the OFF mode is reduced to one year. It seems that memory is consuming more power than it did 20 years ago. But I can`t say for sure yet.
A similar pattern was observed with calculators from TI. The 83+SE (1.5 MB flash, 15MHz) consumes three times as much power when turned off than the 83+ (163kB flash, 6MHz). This consuption of 83+SE in the OFF (sleep) mode is comparable to the characteristics of costly modern 11...14nm smartphones, which have 2...4 GB of RAM. I can only draw one conclusion: modern graphic calculators have become a common consumable, in which no one is interested in the introduction of modern technologies.
My ti-83+ and PDA CASIO PV_s660 from 1999 were lying around without batteries for about 15 years, but all the files in the flash memory and the OS are preserved in them. I can`t say how reliable modern graphic toys from CASIO are in this aspect.

... Do you know why the power consumption is more on the fx-9850GB+ (excluding the LCD consumption) over the 9750G+?

The LCD driver serves three 127×63 layers of pixels: red, green and blue. This is probably the reason for the excessive power consumption of the fx-9850gb+ compared to the fx-9750g+.

... In any case the average current draw when building a table or running the HP summation benchmark (https://www.hpmuseum...hread-9750.html) for about 5 minutes is about 4mA @ 5.2v so around 21mW. Interestingly the idle power consumption is a bit lower on the fx-9750GII SH4 (3.7mW vs 6.8mW), but when put to work the fx-9750GA is rather low power.

I checked the summation benchmark on fx-9750g+(WE) for max=1000. The result is 100s, 35.4 mW (6.8 mA at 5.2 V).

Deg:1000➝N:0:For 1➝X To N Step 1:Ans+cube_root_e_(sin_atan_X):Next
result=1395.346288

The idle mode at 5.2 V: fx-9750g+ (1.7mA, 8.8 mW), fx-9750gii SH3 (1.5 mA, 7.8 mW time-averaged value), fx-9860gii SH4 (1.5 mA, 7.8 mW time-averaged value). The last two calculators differ in power in all other modes. Also, in text scrolling mode, the 9860gii SH4 is 20% more economical than even the 9750g+.

Edited by Hlib2, 20 February 2021 - 04:52 PM.

[MJim]

Currently, despite the advances in technology, in the OFF mode, gii (64kB RAM) consumes three times more microamps than, for example, afx (146kB user RAM). Also, in modern manuals, the battery life in the OFF mode is reduced to one year. It seems that memory is consuming more power than it did 20 years ago. But I can`t say for sure yet.
A similar pattern was observed with calculators from TI. The 83+SE (1.5 MB flash, 15MHz) consumes three times as much power when turned off than the 83+ (163kB flash, 6MHz). This consuption of 83+SE in the OFF (sleep) mode is comparable to the characteristics of costly modern 11...14nm smartphones, which have 2...4 GB of RAM. I can only draw one conclusion: modern graphic calculators have become a common consumable, in which no one is interested in the introduction of modern technologies.

 

I thought the power off power consumption was pretty amazing on the fx-9750GII SH4 using SRAM; 5 minute average of 26.3uA should give me about 3 years (I use eneloop AAA's).  Classpad 300+ is 77.8uA (5 minute average), so that is a little over a year; which isn't too bad considering it's a 2MB chip (512KB available), where I think that the 9750GII may be using a 512KB chip (Sentaro's Casio basic gives access to 256KB, so I'm guessing that it is a 512KB chip).  The fx-CG50 in comparison needs 970uA in comparison to power the DRAM, before hitting a low-power mode 1 hour later and dropping down to 35.5uA. 

 

The fx-9750GA seems also to use SRAM, but probably something much smaller as it averages 6.2uA while off (nearly 15 years on AAA eneloops @ 800mA or nearly 2 years on a CR2032 (using 12.4uA current for lower voltage)).  Looks like the values I measured are roughly 4x higher than the figures quoted in the manual, though things like self-discharge, battery chemistry and environmental influences may of factored into the figures Casio used in their manuals.  Weirdly the fx-9750GA has a 512KB flash chip (1M cycle, 20 year endurance, so a good one), but it doesn't save the history, nor as far as I can see provide any way to upgrade the OS.

 

I checked the summation benchmark on fx-9750g+(WE) for max=1000. The result is 100s, 35.4 mW (6.8 mA at 5.2 V).

Deg:1000➝N:0:For 1➝X To N Step 1:Ans+cube_root_e_(sin_atan_X):Next
result=1395.346288

What is the "Deg" command?  I couldn't find it in the programming reference and is likely a good part of the reason I can't get the nqueens benchmark to work.  In any case using the standard RUN mode without any programming I get close to that time with 1:34 or 94 seconds and the same answer.

 

The idle mode at 5.2 V: fx-9750g+ (1.7mA, 8.8 mW), fx-9750gii SH3 (1.5 mA, 7.8 mW time-averaged value), fx-9860gii SH4 (1.5 mA, 7.8 mW time-averaged value). The last two calculators differ in power in all other modes. Also, in text scrolling mode, the 9860gii SH4 is 20% more economical than even the 9750g+.

 

fx-9750GA idle mode (showing menu - 5 minute average): 5.6V @ 1.20mA ~ 6.8mW.  If I'm in "RUN" mode with nothing on the display it drops to 1.05mA (5 minute average).  Perhaps the screen might be a bit more efficient, or some user/tool errors on my part.  When it was hard at work 4mA is the 5 minute average, but I have seen momentary spikes up to 7mA. 

 

Seems the fx-9860GII SH4 with a bigger screen is pretty much the same power consumption as the fx-9750GII SH4; how is the power consumption with the backlight on (can't believe Casio ditched this on their latest models)?
 

On another note, I have a guess that Casio uses BCD to store their number values.  From what I can tell the 9750G+/GA+ uses 10 bytes for decimal number storage  while the 9750GII uses 12 bytes and the ClassPad 300+ uses 16 bytes.  This means you get more efficient RAM usage on the older models as well, for example I created a table of 100 values (200 numbers) and found it would consume 2050 bytes on the 9750GA, 2516 bytes on the 9750GII SH4 and finally 3256 bytes on the ClassPad 300+.  The Classpad does support exponents of +999/-999 but the base 15 digits is the same, so it should still fit under the same 12 bytes as on the 9750GII/9860GII. 12 bytes makes sense to me for the fx-9750GII/9860 since they are 32 bit processors so work in groups of 4 bytes (10 byte number needs 12 bytes of storage).

Edited by MJim, 21 February 2021 - 12:57 AM.

[Hlib2]

What is the "Deg" command?  I couldn`t find it in the programming reference and is likely a good part of the reason I can`t get the nqueens benchmark to work.

"Deg" sets the "degrees" mode": in the program editing mode, this is Shift MENU F1 F1. In the classic "nqueens benchmark " for the fx-9750g+, this command performs the "No Operation" function (execute command with no operation). It is interesting to note that the "nqueens benchmark" algorithm for afx does not work in fx-9750g+/9850gb+ calculators.

On another note, I have a guess that Casio uses BCD to store their number values.  From what I can tell the 9750G+/GA+ uses 10 bytes for decimal number storage  while the 9750GII uses 12 bytes ...

I do not pay attention to all sorts of small things, if they do not impair the functioning of the calculator. In the old classic models, CASIO used BCD, in the gii series this is not the case. But the gii calculators, nevertheless, did not become worse from this than the previous models.
For example, monochrome images in the fx-9750g+ take up the same amount of memory as color images in the fx-9850gb+. That`s just too bad.

[MJim]

"Deg" sets the "degrees" mode": in the program editing mode, this is Shift MENU F1 F1. In the classic "nqueens benchmark " for the fx-9750g+, this command performs the "No Operation" function (execute command with no operation). It is interesting to note that the "nqueens benchmark" algorithm for afx does not work in fx-9750g+/9850gb+ calculators.

 

Thank you (for the fix and the explanation). I also needed to add a colon (I don't understand how the program works, so I couldn't figure out what was wrong):

Lbl 2:Dsz Y:Deg

Now I was able to run the benchmark and I got very close to your measured value (104 seconds vs 105 seconds).  I also ran the summation program you created and got exactly 100 seconds, so the 9750GA software seems pretty much unchanged over the 9750G+.

 

Kind of odd that they decided to switch to using a flash chip if they weren't going to make much use of it.  fx-9750GA definitely uses SRAM (32KB Cypress CY62256VNLL), since I discovered that I did take a picture and look up the parts.  Unfortunately I couldn't make out the part code on the BSI chip of my now sold fx-9750G+, but I think this will also be an SRAM chip with similar specs.  I guess the OKI chip on the fx-9750G+ is some custom ROM.

 

fx-9750GA+ and fx-9750G+ respectively:

Spoiler

Edited by MJim, 22 February 2021 - 12:10 AM.

[Hlib2]

I also ran the summation program you created and got exactly 100 seconds, so the 9750GA software seems pretty much unchanged over the 9750G+.

CASIO has always been able to sell the same outdated product under dozens of different names. But I do not blame them: they produce cheap goods with good quality. HP, on the contrary, produces a very limited range of calculators at an inflated price, but with poor quality. And this strategy is also a success. 95% of HP calculators are long gone, but we only see the remaining 5%, and we think it`s a good brand. This is the well-known rule of 5% from the field of real statistics, or the error of a survivor in a disaster. Just math :-)

[Hlib2]

@MJim
Thank You for informative and interesting images of the fx-9750g series! I also want to share an image of the elegant electrical board of the fx-9750g+ and fx-9850gb+ calculators, which are fundamentally no different. They are about 20 years old, still function flawlessly and do not require repair.
In my opinion, there are about a dozen similar models made on these boards in different cases.

Edited by Hlib2, 17 March 2021 - 05:48 AM.