Water Cooling a TI-84

[AlexTheGreatish]

 

It is natural to fiddle with your calculator while bored in math class. Maybe you installed some games, wrote some programs, or made art... but what if we took it to the extreme.. with water cooling.

 

Buy : TI-84

On Amazon (PAID LINK): https://geni.us/4SBLqDD

On Newegg (PAID LINK): https://geni.us/GZ4JcDM

Purchases made through some store links may provide some compensation to Linus Media Group.

 

 

 

[Haro]

This brings a smile to my face 

Edited May 27, 2020 by TofuHaroto

[minibois]

Sure, you watercooled it. Very impressive...

Just wake me up when it can divide by 0.

[nick name]

15 minutes ago, AlexTheGreatish said:

 

It is natural to fiddle with your calculator while bored in math class.

 

Phrasing.  

[Rolling2405]

The TI-84 is a low-end calculator though. This would be better with the TI-nspire CX II Cas!

 

@AlexTheGreatish @CPotter @LinusTech @GabenJr

 

Please do an OC guide on the TI-Nspire CX II Cas!

[Zodiark1593]

Ok, cool I suppose. A 26 MHz Z80. Whoopie.  How does it fare against my handheld with a whopping 16.78 MHz 32-bit ARM7TDMI (That pipelining and fast hardware multiplier, Z80 doesn’t even do native multiplication), with 32K of work RAM, 256K of external RAM, and 96K of VRAM.
 

(Bonus react for anyone to recognize the system)

[kelvinhall05]

1 minute ago, Zodiark1593 said:

Ok, cool I suppose. A 26 MHz Z80. Whoopie.  How does it fare against my handheld with a whopping 16.78 MHz 32-bit ARM7TDMI (That pipelining and fast multiplier), with 32K of work RAM, 256K of external RAM, and 96K of VRAM.
 

(Bonus react for anyone to recognize the system)

Nice GBA

[Nystemy]

Ah. good old TI 84 have one myself, though, use my TI82 far more....

Though, the Z80 processor in these aren't really designed to run fast, so even getting beyond 20 MHz is actually rather impressive to be fair.
The main thing holding them back is due to their architectural implementation. Unlike modern X86 CPUs that work with fairly large pipelined architectures, the Z80 does its whole fetch, execute and store operations in the same cycle. _{(Massively oversimplifying things here.)} This means that its honestly doing a lot more per cycle than other architectures like ARM and X86, PowerPC etc do.
 

One downside with doing it all in one cycle is that it gets more sensitive to clock skew due to propagation delay in the logic. This really isn't a major issue down at 16 MHz, but as the clock speeds increase, the impact of skew gets proportionally larger. (Since the skew is a rather fixed amount of time.)

 

But one large upside with doing it in one cycle is that one doesn't need to spend as much resources on signal buffering and such. And can thereby get a bit lower power draw.

Not to mention that the whole chip is designed to run power efficiently, not fast. So the individual cmos stages aren't just slamming from one state to the other, but rather taking a more relaxed approach to ensure that they don't have both the P-channel and N-channel transistors on at the same time. (Doing this obviously adds circuit complexity, and signal delay, but one isn't wasting power for a few percentile of the switching time of the transistor stage.)

Though, the Z80 chip used in this calculator is in the world of micro controllers very power hungry, and there is other processors on the market that consumes a fraction of the power. (There is a few that consumes a handful of µW, instead of the tens of mW the Z80 uses.)

Also, a note about ripping pads.
Don't heat the board too much, it makes the copper delaminate itself from the underlying fiberglass.
One method of avoiding this is to not use the warmest air possible, but rather be a bit more moderate with temperature.
Though, with enough experience one can almost use a blow torch, since one rather looks at the solder melting, than anything else.

Though, an interesting video non the less.

[Zodiark1593]

4 minutes ago, kelvinhall05 said:

Nice GBA

The very same unit that I got in Christmas of 2004. Battery was only swapped out last year. 
 

A shame no calculator app really officially existed for the GBA. Controls aside, the underlying hardware would’ve been quite good at it. 

[sub68]

1 hour ago, TofuHaroto said:

This brings a smile to my face 

yes except my ti84 is 6inches away and probaly think oh no not me

[sub68]

1 hour ago, AlexTheGreatish said:

 

 

 

Linus is holding both a ti 84 plus and ti83

@AlexTheGreatish also wasent there a photo where Alex was holding his ti84 plus with a heart drawn on the screen

[Zodiark1593]

57 minutes ago, Nystemy said:

Ah. good old TI 84 have one myself, though, use my TI82 far more....

Though, the Z80 processor in these aren't really designed to run fast, so even getting beyond 20 MHz is actually rather impressive to be fair.
The main thing holding them back is due to their architectural implementation. Unlike modern X86 CPUs that work with fairly large pipelined architectures, the Z80 does its whole fetch, execute and store operations in the same cycle. _{(Massively oversimplifying things here.)} This means that its honestly doing a lot more per cycle than other architectures like ARM and X86, PowerPC etc do.
 

One downside with doing it all in one cycle is that it gets more sensitive to clock skew due to propagation delay in the logic. This really isn't a major issue down at 16 MHz, but as the clock speeds increase, the impact of skew gets proportionally larger. (Since the skew is a rather fixed amount of time.)

 

But one large upside with doing it in one cycle is that one doesn't need to spend as much resources on signal buffering and such. And can thereby get a bit lower power draw.

Not to mention that the whole chip is designed to run power efficiently, not fast. So the individual cmos stages aren't just slamming from one state to the other, but rather taking a more relaxed back approach to ensure that they don't have both the P-channel and N-channel transistors on at the same time. (Doing this obviously adds circuit complexity, and signal delay, but one isn't wasting power for a few percentile of the switching time of the transistor stage.)

Though, the Z80 chip used in this calculator is in the world of micro controllers very power hungry, and there is other processors on the market that consumes a fraction of the power. (There is a few that consumes a handful of µW, instead of the tens of mW the Z80 uses.)

Also, a note about ripping pads.
Don't heat the board too much, it makes the copper delaminate itself from the underlying fiberglass.
One method of avoiding this is to not use the warmest air possible, but rather be a bit more moderate with temperature.
Though, with enough experience one can almost use a blow torch, since one rather looks at the solder melting, than anything else.

Though, an interesting video non the less.

From what little I’ve read of the Z80 programming manual, this isn’t the necessarily the case, as a number of instructions appear to take multiple cycles to complete. I’m not well versed enough in Z80 assembly to know the exact cycle counts of the different instructions. Lack of cache also adds cycles when it comes to fetching. 
 

Also consider that many operations that are tasked to the calculator are not natively supported by the Z80. I believe the accumulators may actually be 16-bit (don’t quote me here), but the ALUs are most definitely 8-bit. This can be worked around via math routines as necessary to calculate 32-bit numbers, though is very costly from an instruction count perspective (even if arithmetic instructions take a single cycle). Further, multiplication, division and square roots are also not natively supported, requiring additional math routines to perform. 

On a more modern cpu, while it may take more cycles to reach L2 cache and beyond than it takes for a Z80 to fetch it’s own instructions from RAM, bear in mind that Out-of-Order execution keeps the CPU working on other tasks while waiting for data to come in. The use of multiple pipelines allows multiple instructions to be executed simultaneously as well. 
 

In a calculator, working on only 8 bit integers, and only with basic arithmetic (no multiplication nor division), and absolutely nothing else in the background, the Z80 can probably close the gap a fair bit in cycle count for the work done. 

Edited May 27, 2020 by Zodiark1593
Fixed, as proofreading on a phone is not fun.

[boricj]

Before overjoying about Texas Instruments calculators: they've removed ASM/C support from the TI-83 Premium CE through a software upgrade. No more DOOM for you.

 

 

[kuro68k]

"Riced out"? Come on guys, no need for language like that.

[Rolling2405]

1 hour ago, boricj said:

Before overjoying about Texas Instruments calculators: they've removed ASM/C support from the TI-83 Premium CE through a software upgrade. No more DOOM for you.

Just downgrade the OS!

[kuro68k]

11 minutes ago, Thomas001 said:

Just downgrade the OS!

If only you could...

[Rolling2405]

3 minutes ago, kuro68k said:

If only you could...

You can on the Ti-nspire Cx II Cas at least I think so never tried downgrading it.

[Rolling2405]

@kuro68k  Looks like I was wrong there is a downgrade protection that prevents dongrading the OS on both the Ti-84 plus CE and my calculator the Ti-nspire CX II Cas

[Results45]

3 hours ago, AlexTheGreatish said:

 

It is natural to fiddle with your calculator while bored in math class. Maybe you installed some games, wrote some programs, or made art... but what if we took it to the extreme.. with water cooling.

 

 

Nice. It's the literal calculator equivalent to overclocking a Core i9 to 7.7Ghz or unlocking a Vega 56 to match a RTX 2070.

 

Are you guys still shooting the re-assembled 8K RED camera with water cooling?

 

Also, can you do another video with the Sub-Zero chiller where you overclock a 10900K to 5.5Ghz or higher see if a sustained all-core (or multi-core) overclock holds up for a few hours of gaming?

[Zodiark1593]

6 hours ago, Thomas001 said:

@kuro68k  Looks like I was wrong there is a downgrade protection that prevents dongrading the OS on both the Ti-84 plus CE and my calculator the Ti-nspire CX II Cas

Is it any more difficult  than jailbreaking an iPhone?

[koukimonster91]

@AlexTheGreatish In the video after you figured out you lifted the pads you said the resistors were wired in parallel, they are not (how would the CPU read 2 different values) one side of them are(the side closest to the round metal cylinder) connected to vcc. The other sides both have separate traces going back to the CPU so if you want to fix it you will have to wire one of those wires back to the cpu. I have attached a picture showing that below. If you ever need to desolder a smd resistor/cap like that in the future just apply some flux then get a glob of solder on you iron big enough to touch both pads then apply it and the resistor will pull off and stick to the glob of solder on your iron, from there you can pull it off the glob with tweezers.

[Nystemy]

10 hours ago, Zodiark1593 said:

From what little I’ve read of the Z80 programming manual, this isn’t the necessarily the case, as a number of instructions appear to take multiple cycles to complete. I’m not well versed enough in Z80 assembly to know the exact cycle counts of the different instructions. Lack of cache also adds cycles when it comes to fetching. 
 

Also consider that many operations that are tasked to the calculator are not natively supported by the Z80. I believe the accumulators may actually be 16-bit (don’t quote me here), but the ALUs are most definitely 8-bit. This can be worked around via math routines as necessary to calculate 32-bit numbers, though is very costly from an instruction count perspective (even if arithmetic instructions take a single cycle). Further, multiplication, division and square roots are also not natively supported, requiring additional math routines to perform. 

On a more modern cpu, while it may take more cycles to reach L2 cache and beyond than it takes for a Z80 to fetch it’s own instructions from RAM, bear in mind that Out-of-Order execution keeps the CPU working on other tasks while waiting for data to come in. The use of multiple pipelines allows multiple instructions to be executed simultaneously as well. 
 

In a calculator, working on only 8 bit integers, and only with basic arithmetic (no multiplication nor division), and absolutely nothing else in the background, the Z80 can probably close the gap a fair bit in cycle count for the work done. 

There were a reason I put "_{(Massively oversimplifying things here)}" in my original post.
In terms of lack of cache. This really doesn't start to matter unless one gets up to a speed where the SRAM access delay is a large portion of a clock cycle.
Most cheap SRAM has access times around 75-15 ns. We also need to account for propagation delay, to cover the 3 cm distance and back will take us about 0.4ns.
Then we also have some additional delays in memory managements and such. But at 16MHz, a clock cycle is 62.5 ns long. So if we for an example use 25ns SRAM, then we have plenty of time to access the data as if it were right inside of our core, despite sitting a couple of cm away. (Cache isn't needed for latency reasons down at these frequencies. _{And even in modern architectures, cache isn't used for latency reasons, its used to artificially increase bandwidth for shorter program loops and often used data, so that a CPU doesn't have to waste the frankly pathetically slow main memory buss on those tasks. Latency can be solved by prefetching, but that gets a bit nasty after a couple of branches. Since each branch increases our need for bandwidth since it gives us yet one more instruction stream to decode and fetch, rather wasteful if we don't need it.})
 

In terms of Accumulators, Z80 has two, one 8 bit, and an optional 16 bit one. The 16 bit registers can be split to provide more 8 bit registers. (Except the program counter and stack pointer, and two index registers that always are 16 bit respectively.) Then Z80 also has some tricks up its sleeve for BCD processing.

 

In terms of the ALU, its honestly 4 bits wide, but takes between 2 to 4 cycles to execute. This is mainly to lower the transistor count of the chip, make the chip a bit smaller, lower leakage, and generally make it more economical. (Though, some micros on the market has an internal PLL that doubles _{(or more)} the operational frequency of their ALUs, so that a small ALU looks far bigger than it actually is, while not wasting clock cycles.)

 

But Z80 is an architecture that works on 1 instruction at a time. Since that makes the control logic side of things easier to implement, since it becomes a simple state machine. The Z80 though needs between 4 to 23 clock cycles to execute 1 instruction.

X86 on the other hand is very pipeline dependent. For an example, on a Pentium D processor, it takes 32 clock cycles for an instruction to execute.
The Pentium D is though a clear edge case, most normal processors do it in around 4-12 clock cycles. Without out of order execution, we can still execute many instructions at once in a pipelined architecture as long as they aren't dependent on each other _{(such dependencies would create a bubble in our pipeline. Since an instruction can't progress to the next stage until an instruction before it has reached a particular stage. Doesn't always have to be the end of the pipeline though.)}, but Z80 isn't such an architecture. Mainly since implementing a pipeline increases the complexity of the control logic, and that in turn increases power consumption. (It increases performance too, but power consumption increases faster, so from a power efficiency standpoint, pipelines aren't that nice...)

How Z80 compares to a modern architecture is though a discussion that one can talk about for literal days.
And to a degree, the Z80 is a fairly close relative of x86. (Z80 is a knock off copy of Intel's 8080 architecture, and 8086 is simply the 16 bit version of the 8080, but with a janky 20 bit address space. _{(8086 and other early x86 processors used two 16 bit address registers, one though bit shifted 4 bits. Then the two are added together to form the final address. Instead of just shifting one of the registers over a whole 16 bit to provide a 32 bit address space... Intel's marketing and board wanted a 1MB address range for marketing reasons. 4GB were simply too good to be true... Then they quickly learned their mistake, and 8086 mode on modern x86 processors are still a janky hell to deal with. While all the other larger address space modes does things more logically, and simply doesn't expose all pins. (Though, DDR memory channels complicates the matter to be fair...))})

Though, I should probably not go through the whole history of computer architectures, even though the Z1 is an interesting mechanical beast....

[Nystemy]

7 hours ago, koukimonster91 said:

@AlexTheGreatish In the video after you figured out you lifted the pads you said the resistors were wired in parallel, they are not (how would the CPU read 2 different values) one side of them are(the side closest to the round metal cylinder) connected to vcc. The other sides both have separate traces going back to the CPU so if you want to fix it you will have to wire one of those wires back to the cpu. I have attached a picture showing that below. If you ever need to desolder a smd resistor/cap like that in the future just apply some flux then get a glob of solder on you iron big enough to touch both pads then apply it and the resistor will pull off and stick to the glob of solder on your iron, from there you can pull it off the glob with tweezers.

There is a big downside to putting a blob of solder on the tip of an iron.
One can drop it onto the board. (usually not too hard to clean up, but an unneeded risk)

Instead one can go and buy a wide tip to one's iron.
For an example one of these T12 tips from Hakko:

Then one can approach the 0603 resistor from the side, not risk dropping solder all over the place, and gently apply heat and lifting the component with tweezers.
Or use a hot air gun and watch, when the solder melts, one stops heating and lifts the component. (this though requires a bit more skill/experience)

Alex either had 1 of two problems.
Either he lifted off the component too early. When the solder weren't yet melted through, lifting the pads with him. (The copper doesn't adhere as well to the fiberglass when warm, so its very easy to rip off.)
Or heated the board sufficiently for the pads to properly delaminate from the underlying fiberglass, and the surface tension of the solder were enough to pull along the copper.

Though, considering the fairly clean cut at the end of the pad where it transitions to a trace, I would more suspect that the first issue is at fault. Since the later typically also pulls the traces out as well.

I used to run classes on soldering and electronics, so has seen plenty of these types of mistakes.

[devilwarier9]

That soldering job was brutal. Those look like 0603 resistors which are huge in the world of electronics. You really don't need a microscope for something that large, but if you're inexperienced and need one, that's fine, but don't solder with a video microscope, the latency and resolution will kill you. Get a real glass microscope. This is an easy job complicated by using the wrong tools. The tip on the hot air gun was way too big, and he didn't Kapton tape around the resistor to prevent damage. Even worse is that he made all these mistakes with a hot air gun when he didn't need to use one at all. Just 2 soldering irons, hold one on to each side of the resistor and it will pop off real quick. I can't believe he just yanked on it and was surprised the pads got ripped. Also, please get better soldering tips. You have a great digital Hakko iron and tips are super easy to get, but they are using the stock tip which is way too big for Surface Mount work, basically only good for Through Hole.

 

I get that Alex is clearly not an electronics engineer, he's mechanical, but this stuff can be found with a little research. Or just hire an electronics engineer.

[Egg-Roll]

13 hours ago, Zodiark1593 said:

Is it any more difficult  than jailbreaking an iPhone?

I think the proper phrase would be "is it any more difficult to root an Android?" Since that vulnerability came out... RIP Apple.