Timing normal 32K RAM vs 32K 16bit RAM

[HOME AUTOMATION]

How did I get in the middle of this...

[1980gamer]

47 minutes ago, RXB said:

45 seconds vs 45 seconds!

Yep the test is invalid as the sample size can not show any difference therefore is invalid. 

Now run the test in a 10,000 loop and look at times and you be astounded to see they are not exactly the same anymore.

2 things....  First, the 2 times of 45 seconds could be 1.98 seconds different.  I'll bet a stop watch would be more accurate for this test!   LOL,  just being an ass.  If the clock was in 100ths or even 10ths of a second.....

Next, If I am writing or playing a game that only needs to draw the maze once per round, It doesn't really matters if the time separation is visible after 10K iterations.

 

I am 100% on board for any performance gain.

However, we have SO many other great speedups available.  The new RXB and of course the Great and Powerful O..  no, Compiler!

 

This is like watching a five hour stage of the Tour de France that comes down to a photo finish!   The overall time is considered the same.

[GDMike]

Yes. It's agreed, let's take that hundred bucks and buy a stopwatch. 

[+save2600]

This never ending argument is beyond sad. Time to lighten it up!  ?

[+InsaneMultitasker]

I hope we won't need 10,000 posts in this topic to come to a solid conclusion (or agreement).

 

I pulled my 16-bit console out of the storage bin only to realize I had swapped a few components years ago, and it was no longer a modified (or working!) console.   Instead, I tweaked my Geneve clock Call Link routine (from my BBS) to show tenths and hundredths of a second, thinking I could run a few tests for comparison.  Alas, I ran out of time but may give it a whirl tomorrow if this thread is still humming (lurching? plodding?) along. 

 

 

[Reciprocating Bill]

1 hour ago, RXB said:

if you run a 10,000 loop they will not be exactly the same

Rich, I've already shown in my video that execution speeds are not identical, which would be the reason I would decline your bet. Further, no one here has argued that using RAM expansion makes zero difference in execution speed. What everyone can see is that it makes only a small difference. For me, in 1982 or so, that difference was disappointing. 

 

These numbers are identical because your preferred method of measurement (the TIPI clock) is limited to one second increments. Differences less than that can't be accurately reported.

 

A more accurate form of comparison - say watching the processes run side by side - shows that the difference is something just under one second - an insignificant difference, for my purposes, given the 45+ second time required for each maze. As it happens, that's just over a 2% difference - consistent with "not much more than a 1-2% difference." 

 

Of course, given a one second difference, and 10,000 loops, (cue Gomer Pyle) well surprise, surprise! The sum of those differences is 10,000 seconds! What does that tell you that you didn't already know? Nothing. Does that mean that RAM expansion provides a significant performance boost after all? It does not. 

 

10,000 loops would enable us calculate that ~2% difference with more precision. But who cares? Particularly given that each run would take nearly five days, five hours. But go ahead, knock yourself out.  

 

Edited June 9, 2022 by Reciprocating Bill

[Reciprocating Bill]

Rich, as an aside, it's worth noting that the maze program runs in 37 seconds in RXB versus 45 seconds in TI Extended BASIC), a worthwhile speedup (around 21%), probably due to the faster RAND function. That's noticeable. I appreciate that, and your RXB efforts in general. 

 

I just don't see gains in the 2% range (or so) as similarly significant. 

[Tursi]

3 hours ago, RXB said:

RXB has nothing to do with it. 

If you test on Real Iron 32K vs 16bit 32K the test should be repeated so you can see the difference in time as the slower one will have less values then the faster memory.

This is not complicated rocket science here!

If you do like these guys argue just ignore that your timing method can not catch the difference as the sample size is to small to see any difference.

Which is why the failed to see any difference as the timing method they are using sucks!

 

Thus must loop that same routine in same computer with only difference being speed of memory will become apparent.

Bigger the loop the bigger the difference will become between them. This is simple logic.

 

And if there is no difference in speed it should be the same for 1 loop or for 10,000 loops, but I bet $100.00 it will not be the same ever!

Well, it's true overall that it should be slightly faster. But for the same reason that the slow GROM is not responsible for the slow speed of TI BASIC, faster RAM will not make a huge difference in XB performance. A difference, yes. A measurable difference, eventually. But a day-to-day meaningful difference, not too much. The reason is that the CPU spends far more time executing code from the cartridge, in 8-bit wait-stated memory, than time spent accessing the memory expansion.

 

The thing is, you're both* saying that, and then saying the other person is wrong, because you're hung up on whether it's meaningful. That'll depend on the person.

 

Anyway, I added the switch to Classic99 and I'm running away from this thread now.

 

(* - Well, I say both, but I think at least three people are involved... )

 

Edited June 9, 2022 by Tursi

[Willsy]

3 hours ago, Reciprocating Bill said:

Rich, as an aside, it's worth noting that the maze program runs in 37 seconds in RXB versus 45 seconds in TI Extended BASIC), a worthwhile speedup (around 21%), probably due to the faster RAND function. That's noticeable. I appreciate that, and your RXB efforts in general. 

 

I just don't see gains in the 2% range (or so) as similarly significant. 

Yes, that's a major improvement - nice job, Rich.

Edited June 9, 2022 by Willsy
typo

[apersson850]

8 hours ago, RXB said:

45 seconds vs 45 seconds!

Yep the test is invalid as the sample size can not show any difference therefore is invalid. 

Bingo! There you have also concluded that the difference is insignificant. Which is what we are talking about. Not there there is no difference, but that it's of no meaningful value.

[apersson850]

8 hours ago, RXB said:

Again opinion is not factual.

I'm not presenting opinions about how the machine works. I point at facts, facts that can be used to draw conclusions, which will make you understand why there's a minimal difference caused by various speeds of memory. To be able to draw the conclusions require some technical understanding, of course.

 

The only opinion I have is that if you can't see the difference during a 45 second run, then the difference that's there is not important.

[youxia]

I haven't got a clue what this is all about, but it sure was fun to watch...a good ol' nerd argument

 

 

 

[GDMike]

Nothing wrong with hashing it out. I do love this tipi has a clock, but I'll just put it aside right now, but I do have some plans on software involving the tipi clock, like a scheduler, or something.

 

If I were using any basic, id feel confident in using rxb though.

[apersson850]

12 hours ago, RXB said:

Simple science says the sample size they used is worthless for use.

Simple science says that the sample size doesn't matter when each sample is the same.

 

It's like the average of the values in these two lists:

{ 2 }

{ 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 }

 

Many more samples in the second list, but the average can be calculated just as good from the first list.

Edited June 9, 2022 by apersson850

[GDMike]

Sampling is the key. I like this point of view 

Edited June 9, 2022 by GDMike

[+OLD CS1]

Should I update the topic title, then?

[RXB]

15 hours ago, TheBF said:

Then we have a mis-communication.

Nobody said they were the same.

The evidence is that they are close enough that nobody cares.

Where did this 5% go?

If it does not matter how come your test did not show it?

That would mean the Test FAILED or your logic is flawed.

How else would the test not show the 5%?

Now the excuse "Oh does not matter as I said so..." and that is extremely lame as an excuse for a test result.

I am fact basedm,  not going to be bullied into "Because I said so..."

[senior_falcon]

Instead of telling everyone how wrong they are, now that you have the ability to test 8 bit vs 16 bit in MESS/MAME, maybe you can stop posting long enough to run some tests to actually show us. As was said in a previous post, "never have I see you post a video of proof yourself, odd huh?"

[RXB]

15 hours ago, Reciprocating Bill said:

Rich, I've already shown in my video that execution speeds are not identical, which would be the reason I would decline your bet. Further, no one here has argued that using RAM expansion makes zero difference in execution speed. What everyone can see is that it makes only a small difference. For me, in 1982 or so, that difference was disappointing. 

 

These numbers are identical because your preferred method of measurement (the TIPI clock) is limited to one second increments. Differences less than that can't be accurately reported.

 

A more accurate form of comparison - say watching the processes run side by side - shows that the difference is something just under one second - an insignificant difference, for my purposes, given the 45+ second time required for each maze. As it happens, that's just over a 2% difference - consistent with "not much more than a 1-2% difference." 

 

Of course, given a one second difference, and 10,000 loops, (cue Gomer Pyle) well surprise, surprise! The sum of those differences is 10,000 seconds! What does that tell you that you didn't already know? Nothing. Does that mean that RAM expansion provides a significant performance boost after all? It does not. 

 

10,000 loops would enable us calculate that ~2% difference with more precision. But who cares? Particularly given that each run would take nearly five days, five hours. But go ahead, knock yourself out.  

 

Why are you arguing with me all this time when you just said everything I have been telling you for all these posts?

You just repeated all the things I have been saying in these posts?

I think you just had to keep egging me on knowing you were trolling me!

[RXB]

8 hours ago, apersson850 said:

Bingo! There you have also concluded that the difference is insignificant. Which is what we are talking about. Not there there is no difference, but that it's of no meaningful value.

As asked before where did that 5% go in the test results?

The test did not show it as 45 vs 45 shows zero difference.

Thus the test FAILED to show correct results.

When anyone says take my word for it I remember a guy named Leopold Wundt that wrote a thesis on mental illness and was used for 100 years, but today is considered total crap!

[senior_falcon]

13 hours ago, Tursi said:

(* - Well, I say both, but I think at least three people are involved... )

 

Are you referring to me, or one of my multiple personalities?

[RXB]

36 minutes ago, OLD CS1 said:

Should I update the topic title, then?

Yea mostly as the name in title is basically saying come here to troll RXB!

[Reciprocating Bill]

54 minutes ago, RXB said:

You just repeated all the things I have been saying in these posts

Complete agreement! Life is good. 

[+TheBF]

2 hours ago, RXB said:

Where did this 5% go?

If it does not matter how come your test did not show it?

That would mean the Test FAILED or your logic is flawed.

How else would the test not show the 5%?

Now the excuse "Oh does not matter as I said so..." and that is extremely lame as an excuse for a test result.

I am fact basedm,  not going to be bullied into "Because I said so..."

Since our assertion was: (paraphrasing the last 10,000 posts) 

"Running programs from VDP RAM or Expansion RAM makes very little different to the execution speed"   (that' all we are saying)

 

Some simple programs showed a 5% difference.

Bill's more complicated programs show 1..2% difference or less.

This is what was predicted.

 

The assertion has now been proven true, multiple times

 

Case closed.

 

[+OLD CS1]

2 hours ago, RXB said:

Yea mostly as the name in title is basically saying come here to troll RXB!

Slide me a title and it will be done.