Survey posted 2000-11-28 03:13 by Andy Selle.

Contribute ideas to surveys by sending a mail to survey@ticalc.org.

I deleted all of my games and now all of my programs run very fast.

huh? amount of mem free is not proportional to how fast programs run... in my experience anyway...

The less programs you have, the quicker your program menu will display (at least on the 83+). Also, the time taken to search through the VAT for someting would probably decrease an unnoticable millisecond or two.

understood;
I was only thinking about after a program is up and running (sombody catch it...)

On the 86 BASIC programs run slower when the rest of the memory is filled with ASM programs

Duh....

You filled up 96K with Assembly programs? Im guessing they're games, based on the topic...

Ummm.... I have 10 games on my 83+ and it runs faster than my friends 86. If you have figured out like i have, you can store games on the ROM instead of the RAM. The ROM has about 6 times the memory as the RAM. With every game on the ROM my calc runs really fast.

Well the 86 and 83(well the 82,83,83+,85,and86) all use a 6 mhz Z80 processor, so the speed shouldnt be much diff, sept on 89/92(+) which have 10 mhz, or if you overclocked.

Actually...
the 83+ uses a 8 mhz z80 processor, but it is downclocked to 6 mhz.

I have a 83 and 89 and when i have my mem free, the games don't run slower, but loading them, that takes more time.

Well, aren't all the TI calcs are underclocked? I think they are.

Yup, all underclocked massively to conserve battery power, a wise move by TI. unless, that is you are trying to do an extreme task, which wouldn't even be necessary

Why doesn't TI put on a switch where you can elect to have the overclocking turned on and off? After all, one might want to have their 3D graphs calculate faster for a minute or two, since running it overclocked for a short time wouldn't affect the batteries very much.

That would be awesome especially for ion and asm cause then you could have really kick ass grapichs i really don't think people are worried about their batteries too much. Besides if TI really wanted you to save on batteries they would've made it a AA slot instead of AAA.

Since the calcs are so underclocked, the term "overclocked" should just be "clocked" ;)

Actually, HW2 TI-89's are clocked at the full 12MHz instead of the underclocked 10MHz.

although the 83 and the 86 use the same processor, 86 BASIC runs noticably slower than 83 BASIC (not including "compile time" for the 86). why, i don't know....

The TI-86 has larger memory and needs to do RAM page flipping, reducing speed. I'm pretty sure that the TI-86 runs slower when the RAM is full but I don't have a numerical benchmark. In particular, when I add items to the end of the list, it goes really slowly (maybe because is has to move more data in memory, because vars can't be fragmented?)

actually, i think (but do not know this for a fact, so no flames please) that the slowdown is due to the fact that the "ram" is in what we would normally call the processor cache (whether it is l1 or l2 i do not know), alongside the os. the 86 is the only z80 calculator made by ti that uses off-processor ram (albeit only about 64 kilobytes). this is most likely the reason for the significant performance differences.

these performance differences are not limited to program execution, whether it is basic or assembly. yesterday, i ran a test of times for graphing the following equations on my ti-83, ti-85, ti-86, and ti-89:

y1={1,2,3,4,5,6,7,8,9,10}*sin(x)
y2={1,2,3,4,5,6,7,8,9,10}*cos(x)
y3={1,2,3,4,5,6,7,8,9,10}*-sin(x)
y4={1,2,3,4,5,6,7,8,9,10}*-cos(x)
y5=0

these were to be graphed in connected mode simultaneously on a graph that is the zsquare (zsqr) version of the zstandard (zstd) screen and timed using the timer built into my timex expedition watch.
the times are:

89: 3:20.78
83: 3:54.47
85: 6:07.19
86: 8:35.80

that is the order that i had guessed beforehand, but the time differences are still shocking. my explanation for the relatively rapid performance of the 83 is because it has such a small screen when compared to the 85 and 86, as well as the fact that the 83 approximates to ten decimal places, while the 85 and 86 approximate to twelve decimal places, and the 89 approximates to a maximum of fourteen decimal places. my 89 runs a 12 megahertz motorola 68000 processor as it is a hardware 2 calculator. the hardware 1 calculators run 10.5 megahertz motorola 68000 processors, so those would have an even narrower margin of victory over the 83.

i would be interested in seeing times of the same test of other ti calculators, particularly the 82, 83+, and hardware 1 89.

Ok i did the ttest for the TI 83 + and it was using the computer timer so its not that accurate but
the test ended up lasting 3:37.92 to the best of my knowledge thats preety dang good. also i have some questions about making a backlight fo rmy ti83+ and i want to know what the best way of doing that is...
email me for comments on that one

also i have 23607 in ram and 87533 in my rom so if that has nething to do with speed please reply to this message.

^Dragons^ Rule

in theory, ram should have at least a minor effect on performance, for reasons stated above. however, as your calculator has most of its ram free, it should have negligible effects on the test. the amount of flash rom free is irrelevant to this, as that is essentially the calculator equivalent of hard drive space, except that calculators have no need for a swap file. what would cause the greatest variation in performance between otherwise identical calculators is the power left in the batteries.

as far as backlighting, i have no information on how to do this, nor have i done it, but i do remember seeing something about it at richfiles.calc.org . sometime i would like to do it, but i doubt that i will have time to at least until i'm in college, approximately 20-21 months from now.

Well, looks like amount of free RAM does have effects on your calcs performance!
This test took 4 minutes and 16 seconds on my 83 with 2380 bytes of RAM.

~zkostik

i think i forgot to mention this, but when i ran the test on my calcs, the 83, 85, and 86 had nothing in their ram, and everything on my 89 was archived. it is interesting that much lower ram increased the time by about 22 seconds, or about 9.5% .

I already mailed this to you (calcfreak901) but here it is again so everyone can see it. So, with 7000 and something bytes of RAM used the time with my 83+ was ~3:40:00.

with a brand spanking new ti 83 plus, never used, brand new batteries, time for it to graph was 3:22.47

The 89 is just plain fast, the 83+ has less pixels, and the other two are old.

I don't think the 83 has "compile time." Don't the 73, 81, 82, 83, and 83+ run programs line by line? They may look up the "jumping addresses" ahead of time, but I don't think they token (as is the proper word) like all the other calculators do.
eMail me if I'm way off base. I have an 83+ and an 89

Actually, the "Compile Time" on the 85, 86, and 89/92+ (92, also, I think) is not really compile time. TI-BASIC is a scripting language, not a compiled language, on any TI calculator. What the calculator is actually doing during the "compile time" is tokenizing. This means that the calulator is taking the commands ("Disp", "Input", etc) and replacing the actual words with 1 or 2 byte "tokens" (Please tell me if you know whether they are 1 or 2 bytes). This saves space and speeds up execution of the program, but of course you must wait for the program to be tokenized the first time you run it after editing and to be de-tokenized the first time you edit it after running it.

Yeah, on the 85 and up they tokenize first, while on the other calculators they are already tokenized--example:
86: Disp "Hello World" gets tokenized into the value for disp and the token for a string, plus the actual string.
83(+): Disp "Hello World" doesn't really need tokenized because "Disp " is actually already recognized because it has to be entered through the Prgm menu or the catalog. If you type "DISP " on an 83, it will recognize it as D*I*S*P.

The tokens are one byte long, at least on the 83. I don't see any reason for them to be any longer, unless, of course, they had more than 255 tokens...

The way I've heard it is that the TI-86 tokenizes when you run the program after editing, while the TI-83(+) (and probably other calcs) tokenizes it line by line as you scroll past the line in the program editor.

Most of the tokens are one byte, but some are two. Matrix, list, statistics, etc., variables are all two bytes - one byte to say what type of variable it is and another tells the particular variable. This is how it works on 82/83/83+.

The tokenising of instructions for BASIC is not unique to TI calculators. The Commodore 64, for example, has one- or two-byte tokens for each BASIC command and also a two-byte number for each line number. A similar thing happens with assembly language. When an assembly program is assembled, it is translated into tokens or instructions that the processor can understand, but can be interpreted again by humans through the use of a disassembler. The BASIC editor on the TI calcs are essentially disassemblers in a different form.

The jumping addresses are not looked up ahead of time for my 83, since I have discovered that repeatedly jumping to a label that is towards the front of a program is much faster than jumping to a label towards the end. This is approximately a 2 kilobyte program that I tried this on. This discovery led me to move my initialization code to a separate program, so that the BASIC interpreter would not have to search through all of that code to find a label.

I once had 65535 bytes of archive devoted to games.
And yes I know that's FFFF in hex

I deleted all of my programs and now all of my games run very fast.