Running at 2.5Mhz is fantastic, but I recon it can go much higher than that. I was surprised when I got my breadboard CPU running at 2Mhz and none of it was on PCBs and the clock was a 555 timer, not even a crystal. You sounded so happy I thought you were going to cry. Well done!
I know my cap is about 4mhz with parts running "in spec" so I'm pleased with where it's at. I've actually identified 2 changes I could make to the build that take the theoretical max to 8mhz, I mighty do a video or two on that when I'm done with everything else. Thanks David!
@@weirdboyjim ah I see, I would have thought those would be really fast. Good to know though because I was just about to order some for my z80 computer rom, they'll probably be fine for testing but I'm hoping to get up to 20MHz so I might need something faster
Definitely, if you want to achieve frequencies like that you would be far better of having nothing but ram but building a circuit that can copy the contents of a rom chip into ram at startup. If you use something like my reset circuit, use the S/R latch to grab the z80's busrq line and have a counter run through memory copying it.
@@weirdboyjim Thanks for the reply, I had similar thoughts, however I thought I'd use the Z80 itself (probably downclocked to some lower speed) to copy the ROM into RAM, then disable the ROM and connect the 20MHz crystal. I thought I could have reads come from ROM and writes go to RAM on then lower 32k so that all the Z80 would do is read from and write back to the same location to copy it to RAM. However I'd not really thought of doing it all in hardware: that is probably a better choice, although I'll have to do some more research, and try to design each circuit (and software in the case of the first design) and see which is simpler/works better.
This is so GREAT to watch! Congratulations James! Seriously, this is one of the few channels I watch that I look forward to seeing a new video in. Thank you!
Don't worry, there are plans for big projects beyond this one. I can see at least 4 months of videos to finish this build, which probably means another year in real terms.
Wow - this video's not only dropping one bombshell (>2 MIPS) - which by itself is far more than impressive, ... but three! Audio and Video those are - yes, I heard you right, Sir ;-). As you can imagine - I'm totally psyched right now! This means much more to come :) Nevertheless I'm looking forward to the next "smaller" steps, too. With the PCB versions of the Clock- and ALU modules this board will look even more impressive, than it is already. Wish I could press thumbs up more than once.
I really appreciate the support Andreas. The current plan is that the next cpu video will add a single transistor to the build, but it should have an interesting use.
Congratulations Jim, that was awesome. I was sitting here shaking my head thinking there is no way you will get 2Mhz on breadboards. How wrong I was! It just goes to show there are advantages to building your own CPU rather than using an off the shelf one. I am looking forward to when you get around to creating the video circuitry.
The video is going to involve much higher frequencies, fortunately it's a bit more fault tolerant. A glitch may be just visual rather than crashing the program.
I remember the great debate (or was it an all out war) between the BBC, Speccy, and 64.... another kiddie and I were "exempt" because he had a QL and I had a Jupiter Ace.... we just didn't count. ;) Great to see your build getting up to microprocessor speed even with the ALU still on breadboard.... that's pretty damn good!
Congratulations! In terms of peak MIPS, my Magic-1 is just barely holding its own against your machine. Magic-1’s clock runs 4.09 Mhz and register-register instructions take 2 clocks, so its peak MIPS is 2.045 - just barely edging out your 2.025. I’ve no doubt that you’ll blow past it in short order, though. When Magic-1 first came alive, I couldn’t push past 3 Mhz. After analyzing critical paths, the bottleneck was in the way I was forming a physical address via the page table. I redesigned the memory access mechanism to split address translation across 2 clock cycles and that enabled me to get to 4 Mhz. I then went searching for the next critical path to attack, but found multiple (mostly around setting condition codes and branching). So, 4.09 Mhz is Magic-1’s upper limit. Of course, peak MIPS isn’t that useful. Can’t do too much just moving data around registers. It’s memory access that matters. I don’t recall exactly, but a typical memory reference instruction for Magic-1 probably takes 5 or 6 clocks. So, given a typical instruction mix it’s well under a 1 MIPS machine. In the old days, we used the Dhrystone benchmark to measure MIPS. A Vax 11/780 scored 1757 Dhrystones and was considered a 1 MIPS machine. Magic-1 is capable of running Dhrystone, and scores 529. So, on that scale, it’s a 0.30 MIPS computer. Dhrystone is largely an integer and string benchmark. There was also Whetstone for double-precision floating point. Via a software floating point emulation library, Magic-1 can actually run Whetstone. Amusingly, it scores 0.000578 MIPS. In other works, 578 instructions per second - “instructions” not “millions of instructions”. Once you firm up your instruction set architecture send me the details. I’ve done enough retargetings of the C compiler that Magic-1 uses (lcc) that I may be able to quickly cobble together something that could work for you. Then, you might be able to run Dhrystone (and other C programs) yourself.
Thanks Bill! My 2.025 is actually a real world code test. Register/Register operations on my cpu are all 1 cycle, so I could claim a peak of 2.5mips at this clock rate. I've analyzed my critical paths, 4mhz is reasonable but to go beyond that I either need faster memory or a I need to add some pipeline steps to spread things out a bit. I'm not claiming superiority though, your ISA is much more advanced than mine. 16 Bit operations on my cpu are only inc/dec and I only have one addressing mode. In the 8bit era yours is closer to a mini computer. I could probably make a fairly good showing of a hand coded equivalent to Dhrystone but targeting lcc might be painful. My cpu has some idiosyncrasies that lcc would struggle with. I have thought about extending the assembler so it can process "flat" code into something that works (i.e. re-order instructions or nop the branch delay slots etc..) That might make it easier to target a compiler.
Thanks Jerril. I'm trying to plan the ALU's pcb's at the moment, it's actually quite tough to plan the modules when you are trying to make it fit a fixed rectangle.
@@weirdboyjim Sadly that is a skill I have never mastered for my own projects. I never usually have a script. It I don't mind watching it back then it is good to ship in my opinion and I really enjoyed your video so I bet you'd be happy to watch it back.
About the 555 astable - you can use the output pin through resistor to trigger + threshold with the capacitor to ground to get almost ideal 50% pulse wave with less parts ;-)
@@weirdboyjim www.electronics-tutorials.ws/waveforms/555-circuits-part-1.html First picture in this site I think ;-) EDIT: and to elaborate a little bit, you might have noticed that when you change the frequency potentiometer on your current circuit your clock pulsewidth changes - with proposed circuit it stays 50% no matter where the frequency potentiometer is set. Otherwise it works identical - less resistance = higher frequency.
Congratulations on achieving 2+Mips. My first computer was a ZX81, (1K version), then a CPC464 and then an Amstrad PCW something or other. My first job was in 1990 and it involved using MicroVAXes. In 1996 bought my first PC, between the time I'd seen it in the store and bearing to part with my money, the CPU had gone from 133MHz to 166MHz; thought I had the fastest PC I would every need 😂
@@weirdboyjim Ha Ha, you missed my comment on your snake game, about how I suck at games. But it's incredible your CPU is outperforming commercial chips. Although I wonder if there was no point in making the CPUs of the 1980s "fast" because they would have just been waiting for memory all the time. It would be interesting to see how much time your CPU was waiting for the UART to send data. Maybe re-write to store the values in memory rather than write to display, then flag completion of calculations and then begin memory dump.
Great video again James, but the VCC symbol not pointing up and the Ground not pointing down makes my brain go Noooooo. Sorry it's just the way I have been programed over the years. :)
I'm never going to be able to keep everyone happy. Pretty much bit of eda I've ever done was recorded and put up on this channel (or will be) so pretty much the only training i've had is people complaining about things. I'll try and keep that one in mind.
@@weirdboyjim Was not complaining just a bit of OCD!! I was lucky to have worked where the circuits were draw up by the drawing office and they did the PCB layouts as well. We just did the circuit on sheets of A4 and they put it down on paper as it was in them days. So I had to learn to do it how they liked it or got things thrown at me. No elf and safety in them days and they were good shots. So hardest part was checking their drawing to my bit of scrap paper. How things have changed.
@@weirdboyjim yes, just finishing the build of the program counter and revised clock PCB. Multiplier PCB finished and also nearly finished Instruction decode, so it's starting to fly quickly now.
That's awesome! Meanwhile, my breadboard transistor clock is so sensitive to noise that it can't do a few hertz without flickering if I touch anything... :) Side note: the same circuit on a PCB is surprisingly stable.
Thanks! Indeed, the breadboard quality can make a big difference but older components can build up a bit of tarnish on their legs that doesn't help. Some times it's difficult to tell the difference between a circuit problem and a breadboard problem.
Indeed. I'm intending to run my video circuit at 25.175mhz, you actually have to start jumping through some hoops to get things to run that fast. The big benefit with video though is that it's more forgiving, the odd glitch just means a little corruption and not a crashed program.
Congratulations!! Lovely seeing this milestone. I never had any doubt you'd hit 1 Mips. I'm betting you'll do fine up to about 4Mhz, after that it would depend on how fast those EEPROMs you selected are? ( 70ns or 150ns ). Cheers!
My EEPROMS are 150ns, that and some extra chained logic means about 4mhz is my theoretical max with parts running in spec. I have designed some changes that could get that over 8mhz but it involves some compromises to the instruction latency. 2.5mhz is double what I need, so I'm happy with that.
Absolutely amazing! How much faster will it go when it's all on PCB? Can't wait to find out. It would be fun to take out all but the last value and "Done" UART output for a proper comparison at high speed. :) FWIW I still don't like your block symbols, especially for gates... is that the only option with that tool, or does it have "unit" based symbols like KiCAD?
Thanks Lawrence! A while ago I worked out that about 4mhz was as fast as it would legitimately go before I'm over clocking some of the components, but most components these days are faster than their official ratings.
Ah so 250nS parts somewhere. Maybe those parts could be replaced with faster ones. Also regarding MIPS: for a totally fair comparison your routine would be written in Z80 (etc) assembly and compared that way. There's points in your ISA where multiple instructors are needed which I'm pretty sure most other 8 bits could do in one. Anyway it's an incredible achievement and I'm absolutely itching to see your video and sound circuits :)
There is no single part like that, but you have to look at chains of logic and ask yourself what the longest path thru it is. The slowest bits though are the EEPROMS which add up very quickly.
@@weirdboyjim ah of course. In terms of the EEPROMs you could swap them for some one time programmable 45nS parts. You better finalise the micro code before using them though! They are surprisingly cheap:
To be honest, 1 Mips would have been fast enough for my plans. You say "There's points in your ISA where multiple instructors are needed which I'm pretty sure most other 8 bits could do in one" but that's always been a source of arguments. there have been various attempts over the years to make standards with series of known operations. These days people usually talk in terms of flops, I may have to write a floating point library just to get close to a standard benchmark against z80 and 6502.
Thanks Leonardo! If I build a processor on FPGA it may have some common design elements but I'd take advantage of the flexibility. There is some pain in this design because the GPR's and address registers are different sizes, a fair amount of complexity goes to dealing with that. On an fpga could multiply things up to 32bits which is the first size that's pretty comfortable as both an address and gpr, makes the design simpler. No more xfer bus or xfer register.
The led’s on the left show the value on the address bus. Much of the time they will show the instruction fetch address but when an instruction that reads or writes memory hits pipeline stage 2 the address will jump to that location. I’d need to watch the video on something bigger than my phone screen to be sure.
Unfortunately not. The load lines are demultiplexed so I can only load one register in each grouping in a cycle. Saves control lines and the only real limitation is that you can't broadcast like this.
Congratulations! You can probably go onto 4 MHz clock already now. Once you have it all on PCBs you go to 16 MHz. Haha. You should remove that waiting for the UART and just write the results into RAM instead, and just check the results in the end. That way you could get faster execution. Have you thought about making a more compact card-edge connected passive backplane system which fits into a box, similar to the mini-computers of old? Or you want to pin this setup to the wall, right?
This build is about seeing it, hence all the led's and the deliberate padding between modules. I'm hoping it will make an awesome wall mounted piece when it's done.
The UART currently runs 115200, but it transmits in a16bit frame at moment. When I add the transmit buffering I'll tune that down and get a 10bit frame for max speed. That will increase my effective transmit rate by about 50%
You and me both! I have worked out that roughly 4mhz should be the upper limit based on the known critical path. That said the rated speed of the parts causing that appears to be very conservative and their are faster options available. I'll have more of a play when it's all been converted to pcb. It would benefit from the final vga display as well, right now I don't have anyway to get data back out quick enough to do 2.5mhz justice, it just ends up waiting for longer.
@@weirdboyjim When the VGA display gets going, it would be really nice to see this thing running full on at 4MHz+ As usual I'll be following this project very closely!
I usually make the schematics / pcb designs public at easyeda once they are tested and running in the build. (Check them out here - easyeda.com/weirdboyjim ), if I forget to make something public just post a friendly reminder on the video where I show it.
I hadn't planned one for my build as the UART gives me that functionality already, actually the ps2 interface is extremely close to the uart, you just get to drive the clock for both ends from the host instead of synchronising it. Might be able to do a supplement video in the uart series if there is call for it.
nstruction and memory address and content displays and switches would be great instead of relying on "EEPROM's" knobs for run/stop, clock speed,single step,deposit,deposit next,examine,examine next next would be great
I have switches and led's for most of that already, but I don't have an easy way of inspecting memory while the processor is stopped. I'll have to give it some thought. I'd need some way of overriding the address section lines, but they are driven from a chip with tri-state outputs that are just always on, so it is possible.
Running at 2.5Mhz is fantastic, but I recon it can go much higher than that. I was surprised when I got my breadboard CPU running at 2Mhz and none of it was on PCBs and the clock was a 555 timer, not even a crystal. You sounded so happy I thought you were going to cry. Well done!
I know my cap is about 4mhz with parts running "in spec" so I'm pleased with where it's at. I've actually identified 2 changes I could make to the build that take the theoretical max to 8mhz, I mighty do a video or two on that when I'm done with everything else. Thanks David!
2.5MHz! Wow I didn't expect that, but I guess modern 7400s logic is probably faster than I expect.
74 series logic is pretty fast, but the slowest parts in the build are those EEPROM's
@@weirdboyjim ah I see, I would have thought those would be really fast. Good to know though because I was just about to order some for my z80 computer rom, they'll probably be fine for testing but I'm hoping to get up to 20MHz so I might need something faster
Definitely, if you want to achieve frequencies like that you would be far better of having nothing but ram but building a circuit that can copy the contents of a rom chip into ram at startup. If you use something like my reset circuit, use the S/R latch to grab the z80's busrq line and have a counter run through memory copying it.
@@weirdboyjim Thanks for the reply, I had similar thoughts, however I thought I'd use the Z80 itself (probably downclocked to some lower speed) to copy the ROM into RAM, then disable the ROM and connect the 20MHz crystal. I thought I could have reads come from ROM and writes go to RAM on then lower 32k so that all the Z80 would do is read from and write back to the same location to copy it to RAM.
However I'd not really thought of doing it all in hardware: that is probably a better choice, although I'll have to do some more research, and try to design each circuit (and software in the case of the first design) and see which is simpler/works better.
I cheered for you at 1 MIPS. Congrats on such an amazing project.
Thanks Adam! I really appreciate the support!
Awesome! Congrats on this milestone. It was great following this series up to this point.
I appreciate it Alex! There will be plenty more before I'm done.
@@weirdboyjim Can't wait for graphics!
This is so GREAT to watch! Congratulations James! Seriously, this is one of the few channels I watch that I look forward to seeing a new video in. Thank you!
That's high praise thanks Bob, I hope you continue to enjoy.
"Hopefully brings this entire build a bit closer to finishing." I hope not! This should never end!
Don't worry, there are plans for big projects beyond this one. I can see at least 4 months of videos to finish this build, which probably means another year in real terms.
Wow - this video's not only dropping one bombshell (>2 MIPS) - which by itself is far more than impressive, ... but three! Audio and Video those are - yes, I heard you right, Sir ;-).
As you can imagine - I'm totally psyched right now! This means much more to come :)
Nevertheless I'm looking forward to the next "smaller" steps, too. With the PCB versions of the Clock- and ALU modules this board will look even more impressive, than it is already.
Wish I could press thumbs up more than once.
I really appreciate the support Andreas. The current plan is that the next cpu video will add a single transistor to the build, but it should have an interesting use.
Congratulations Jim, that was awesome. I was sitting here shaking my head thinking there is no way you will get 2Mhz on breadboards. How wrong I was!
It just goes to show there are advantages to building your own CPU rather than using an off the shelf one. I am looking forward to when you get around to creating the video circuitry.
The video is going to involve much higher frequencies, fortunately it's a bit more fault tolerant. A glitch may be just visual rather than crashing the program.
The first "hands-on" computer I used was a DEC PDP-15. It had a knob on the front panel which varied the clock rate from 1 Hz up to full speed.
Nice! My mid speed clock I can smoothly adjust, I quite enjoy watching it sometimes when it's slow enough to see the actual operations.
Waw that is indeed awesome. Faster than my c64 for sure. Brilliant. Gives me some ideas for something I've been thinking about for a future project
Thanks Dennis! I look forward to seeing what you come up with.
Congratulations. This is quite an achievement. I am amazed that given these of breadboards and the speeds that it is stable.
Thanks Jack! It is indeed running faster than I had any hope to expect. I've been running the lock faster still recently.
woohoo.
when it's all set into PCB, it may go to 5MHz !
congrats !!
With the parts operating within spec 4mhz would be my cap. Not sure I want to "overclock" anything since this is more about the design. Thanks!
@@weirdboyjim you can at least try, just for kicks
Amazing ! Well done James. I didn’t know about the power monitor chip. Thanks for the tip. I’m going to include that in my build. Cheers.
Glad that was helpful Ced. It's saved me some time on a few occasions.
Congrats on hitting 2.5 Mips with half the computer still using breadboards.
Technical that's 2.5 mhz and roughly 2mips but I appreciate the sentiment. ;-)
@@weirdboyjim doh oh yeh, didn't think of that at the time.
I remember the great debate (or was it an all out war) between the BBC, Speccy, and 64.... another kiddie and I were "exempt" because he had a QL and I had a Jupiter Ace.... we just didn't count. ;)
Great to see your build getting up to microprocessor speed even with the ALU still on breadboard.... that's pretty damn good!
I'd call it a debate, it didn't come a true war until the time of Amiga verses Atari-ST, I side stepped that by switching to pc.
2.5Mhz is a great result for a system with still some breadboards. CPU pipelines are working great...cheers!
I actually get it into 4 in a later video
Congratulations! In terms of peak MIPS, my Magic-1 is just barely holding its own against your machine. Magic-1’s clock runs 4.09 Mhz and register-register instructions take 2 clocks, so its peak MIPS is 2.045 - just barely edging out your 2.025. I’ve no doubt that you’ll blow past it in short order, though.
When Magic-1 first came alive, I couldn’t push past 3 Mhz. After analyzing critical paths, the bottleneck was in the way I was forming a physical address via the page table. I redesigned the memory access mechanism to split address translation across 2 clock cycles and that enabled me to get to 4 Mhz. I then went searching for the next critical path to attack, but found multiple (mostly around setting condition codes and branching). So, 4.09 Mhz is Magic-1’s upper limit.
Of course, peak MIPS isn’t that useful. Can’t do too much just moving data around registers. It’s memory access that matters. I don’t recall exactly, but a typical memory reference instruction for Magic-1 probably takes 5 or 6 clocks. So, given a typical instruction mix it’s well under a 1 MIPS machine.
In the old days, we used the Dhrystone benchmark to measure MIPS. A Vax 11/780 scored 1757 Dhrystones and was considered a 1 MIPS machine. Magic-1 is capable of running Dhrystone, and scores 529. So, on that scale, it’s a 0.30 MIPS computer.
Dhrystone is largely an integer and string benchmark. There was also Whetstone for double-precision floating point. Via a software floating point emulation library, Magic-1 can actually run Whetstone. Amusingly, it scores 0.000578 MIPS. In other works, 578 instructions per second - “instructions” not “millions of instructions”.
Once you firm up your instruction set architecture send me the details. I’ve done enough retargetings of the C compiler that Magic-1 uses (lcc) that I may be able to quickly cobble together something that could work for you. Then, you might be able to run Dhrystone (and other C programs) yourself.
Thanks Bill! My 2.025 is actually a real world code test. Register/Register operations on my cpu are all 1 cycle, so I could claim a peak of 2.5mips at this clock rate. I've analyzed my critical paths, 4mhz is reasonable but to go beyond that I either need faster memory or a I need to add some pipeline steps to spread things out a bit.
I'm not claiming superiority though, your ISA is much more advanced than mine. 16 Bit operations on my cpu are only inc/dec and I only have one addressing mode. In the 8bit era yours is closer to a mini computer.
I could probably make a fairly good showing of a hand coded equivalent to Dhrystone but targeting lcc might be painful. My cpu has some idiosyncrasies that lcc would struggle with. I have thought about extending the assembler so it can process "flat" code into something that works (i.e. re-order instructions or nop the branch delay slots etc..) That might make it easier to target a compiler.
Thanks James. Fantastic speed, especially considering the ALU is still on breadboard. Interesting video. Take care.
Thanks Jerril. I'm trying to plan the ALU's pcb's at the moment, it's actually quite tough to plan the modules when you are trying to make it fit a fixed rectangle.
@@weirdboyjim That is going to be tough, a lot of routing in there. Very exciting, lot's to look forward to. Good luck.
Well done James, I got quite excited at the end along with you.
I really appreciate that thanks David! Got any tips on not having your videos all end up 50% longer than you expected?
@@weirdboyjim Sadly that is a skill I have never mastered for my own projects. I never usually have a script. It I don't mind watching it back then it is good to ship in my opinion and I really enjoyed your video so I bet you'd be happy to watch it back.
Thanks David! That's one of the benefits of editing your own videos. Your own voice stops seeming weird.
About the 555 astable - you can use the output pin through resistor to trigger + threshold with the capacitor to ground to get almost ideal 50% pulse wave with less parts ;-)
Got a link for that? I can't quite decipher the difference between what you are suggesting and what I did.
@@weirdboyjim www.electronics-tutorials.ws/waveforms/555-circuits-part-1.html
First picture in this site I think ;-)
EDIT: and to elaborate a little bit, you might have noticed that when you change the frequency potentiometer on your current circuit your clock pulsewidth changes - with proposed circuit it stays 50% no matter where the frequency potentiometer is set. Otherwise it works identical - less resistance = higher frequency.
Nice, awesome work! Can't wait to see the PCB build!
Thanks Brandon, I have a few tweaks to do before I can set it in stone.
Congratulations on achieving 2+Mips. My first computer was a ZX81, (1K version), then a CPC464 and then an Amstrad PCW something or other. My first job was in 1990 and it involved using MicroVAXes. In 1996 bought my first PC, between the time I'd seen it in the store and bearing to part with my money, the CPU had gone from 133MHz to 166MHz; thought I had the fastest PC I would every need 😂
Thanks Alex! Wow 166mhz! That was a bit of a beast in 96' Doom must have been reasonably smooth!
@@weirdboyjim Ha Ha, you missed my comment on your snake game, about how I suck at games.
But it's incredible your CPU is outperforming commercial chips. Although I wonder if there was no point in making the CPUs of the 1980s "fast" because they would have just been waiting for memory all the time. It would be interesting to see how much time your CPU was waiting for the UART to send data. Maybe re-write to store the values in memory rather than write to display, then flag completion of calculations and then begin memory dump.
Great video again James, but the VCC symbol not pointing up and the Ground not pointing down makes my brain go Noooooo.
Sorry it's just the way I have been programed over the years. :)
I'm never going to be able to keep everyone happy. Pretty much bit of eda I've ever done was recorded and put up on this channel (or will be) so pretty much the only training i've had is people complaining about things. I'll try and keep that one in mind.
@@weirdboyjim Was not complaining just a bit of OCD!! I was lucky to have worked where the circuits were draw up by the drawing office and they did the PCB layouts as well. We just did the circuit on sheets of A4 and they put it down on paper as it was in them days. So I had to learn to do it how they liked it or got things thrown at me. No elf and safety in them days and they were good shots.
So hardest part was checking their drawing to my bit of scrap paper.
How things have changed.
When I'm soldering through hole resistors, I insist that all the tolerance bands must be aligned "correctly"... ;)
Really nice work, you've set the bar pretty high!
Thanks Paula! Managed to put any time into your build?
@@weirdboyjim yes, just finishing the build of the program counter and revised clock PCB.
Multiplier PCB finished and also nearly finished Instruction decode, so it's starting to fly quickly now.
Congratulations! Damn, that's impressive!
Thanks Dylan! I appreciate it.
Awesome James!!!! Congratulation!
Thank you so much 😀 I get a little more speed out of later in the series!
That's awesome! Meanwhile, my breadboard transistor clock is so sensitive to noise that it can't do a few hertz without flickering if I touch anything... :) Side note: the same circuit on a PCB is surprisingly stable.
Thanks! Indeed, the breadboard quality can make a big difference but older components can build up a bit of tarnish on their legs that doesn't help. Some times it's difficult to tell the difference between a circuit problem and a breadboard problem.
I bet it can go quite a bit faster! I got a video generator circuit to run at 8 MHz on breadboards without too many problems
Indeed. I'm intending to run my video circuit at 25.175mhz, you actually have to start jumping through some hoops to get things to run that fast. The big benefit with video though is that it's more forgiving, the odd glitch just means a little corruption and not a crashed program.
Congratulations!! Lovely seeing this milestone. I never had any doubt you'd hit 1 Mips. I'm betting you'll do fine up to about 4Mhz, after that it would depend on how fast those EEPROMs you selected are? ( 70ns or 150ns ). Cheers!
My EEPROMS are 150ns, that and some extra chained logic means about 4mhz is my theoretical max with parts running in spec. I have designed some changes that could get that over 8mhz but it involves some compromises to the instruction latency. 2.5mhz is double what I need, so I'm happy with that.
Absolutely amazing! How much faster will it go when it's all on PCB? Can't wait to find out. It would be fun to take out all but the last value and "Done" UART output for a proper comparison at high speed. :)
FWIW I still don't like your block symbols, especially for gates... is that the only option with that tool, or does it have "unit" based symbols like KiCAD?
Thanks Lawrence! A while ago I worked out that about 4mhz was as fast as it would legitimately go before I'm over clocking some of the components, but most components these days are faster than their official ratings.
Ah so 250nS parts somewhere. Maybe those parts could be replaced with faster ones. Also regarding MIPS: for a totally fair comparison your routine would be written in Z80 (etc) assembly and compared that way. There's points in your ISA where multiple instructors are needed which I'm pretty sure most other 8 bits could do in one. Anyway it's an incredible achievement and I'm absolutely itching to see your video and sound circuits :)
There is no single part like that, but you have to look at chains of logic and ask yourself what the longest path thru it is. The slowest bits though are the EEPROMS which add up very quickly.
@@weirdboyjim ah of course. In terms of the EEPROMs you could swap them for some one time programmable 45nS parts. You better finalise the micro code before using them though! They are surprisingly cheap:
To be honest, 1 Mips would have been fast enough for my plans. You say "There's points in your ISA where multiple instructors are needed which I'm pretty sure most other 8 bits could do in one" but that's always been a source of arguments. there have been various attempts over the years to make standards with series of known operations. These days people usually talk in terms of flops, I may have to write a floating point library just to get close to a standard benchmark against z80 and 6502.
I'm late to the party but, Congrats on breaking 1 Mips!
Thanks Rouverius!
Nice work!
Thanks Scott! I appreciate the support.
This is crazy James, 2.5 Mips with breadboards, awesome. Did you think after this project building the same processor using an FPGA?
Thanks Leonardo! If I build a processor on FPGA it may have some common design elements but I'd take advantage of the flexibility. There is some pain in this design because the GPR's and address registers are different sizes, a fair amount of complexity goes to dealing with that. On an fpga could multiply things up to 32bits which is the first size that's pretty comfortable as both an address and gpr, makes the design simpler. No more xfer bus or xfer register.
@@weirdboyjim I'm curious to know what it will be the clock supported by the final PCB version :D
30:40 DIAMOND MOMENT
Get's faster in the future!
What happened to the LEDs on the Memory module at 25:44?
The led’s on the left show the value on the address bus. Much of the time they will show the instruction fetch address but when an instruction that reads or writes memory hits pipeline stage 2 the address will jump to that location. I’d need to watch the video on something bigger than my phone screen to be sure.
@@weirdboyjim It looks to me like they are slowly fading out while the clock is not running...
@@miege90 ahh, figured it out. Reset is Being held so nothing is asserting to the address bus. Those lines are floating.
“Actually Clapping!”
Thanks again Pete! I managed to get it a chunk faster than this for the vga videos but I haven't done my final check on max speed yet.
I think you can also initialise registers B, C and D to zero by issueing a parallel load to them?
Unfortunately not. The load lines are demultiplexed so I can only load one register in each grouping in a cycle. Saves control lines and the only real limitation is that you can't broadcast like this.
@@weirdboyjim Ah I see, you’re right about that! Thanks for the rapid answer! I’m bingewatching your videos, great series :).
Congratulations! You can probably go onto 4 MHz clock already now. Once you have it all on PCBs you go to 16 MHz. Haha. You should remove that waiting for the UART and just write the results into RAM instead, and just check the results in the end. That way you could get faster execution.
Have you thought about making a more compact card-edge connected passive backplane system which fits into a box, similar to the mini-computers of old? Or you want to pin this setup to the wall, right?
This build is about seeing it, hence all the led's and the deliberate padding between modules. I'm hoping it will make an awesome wall mounted piece when it's done.
So what baud rate are you running at?
Sheesh Jonny, do you even watch my UART videos ;-)
The UART currently runs 115200, but it transmits in a16bit frame at moment. When I add the transmit buffering I'll tune that down and get a 10bit frame for max speed. That will increase my effective transmit rate by about 50%
The question is how high can it go? I want to know what the absolute limit is.
You and me both! I have worked out that roughly 4mhz should be the upper limit based on the known critical path. That said the rated speed of the parts causing that appears to be very conservative and their are faster options available. I'll have more of a play when it's all been converted to pcb. It would benefit from the final vga display as well, right now I don't have anyway to get data back out quick enough to do 2.5mhz justice, it just ends up waiting for longer.
@@weirdboyjim When the VGA display gets going, it would be really nice to see this thing running full on at 4MHz+
As usual I'll be following this project very closely!
18:40 Opportunity to butcher old "James Sharman" sticker to bottom right of new board missed ;) Branding man BRANDING! hehehe
I made that sticker before the first wire went in a breadboard!
You are genius man
I think you misspelled crazy ;-) Thanks Mati!
but how fast can you actually go slowly increase the clock until it stops working
I will do, but once a few extra bits are finished.
Loving the "Timely Annotations" :) Getting Ninja with YooToob.
I do try, editing takes more time than the building these days!
@@weirdboyjim As the guy who took "Um, erm, aaaah and oh"'s out of Radio shows for a good while. Believe me, Choir, Literally: backing singer.
Looks like you need to buy a faster crystal! 😉
I do run it a bit faster later on.
damn .... that is fast :D
Thanks Karol! Hopefully I'll be able to use it to do some more interesting things soon.
Hi James. Do you keep any online documents or schematics of it, even if they are outdated? Thanks?
I usually make the schematics / pcb designs public at easyeda once they are tested and running in the build. (Check them out here - easyeda.com/weirdboyjim ), if I forget to make something public just post a friendly reminder on the video where I show it.
@@weirdboyjim Hi, this link seems to be broken. Please can you verify it?
@@lsfornells should be fine now. RUclips was including the close bracket in the url.
@@weirdboyjim Thanks for that, I actually realised about the bracked a couple of minutes ago
Do you have any plans for a PS2 keyboard interface?
I hadn't planned one for my build as the UART gives me that functionality already, actually the ps2 interface is extremely close to the uart, you just get to drive the clock for both ends from the host instead of synchronising it. Might be able to do a supplement video in the uart series if there is call for it.
nstruction and memory address and content displays and switches would be great instead of relying on "EEPROM's"
knobs for run/stop, clock speed,single step,deposit,deposit next,examine,examine next next would be great
I have switches and led's for most of that already, but I don't have an easy way of inspecting memory while the processor is stopped. I'll have to give it some thought. I'd need some way of overriding the address section lines, but they are driven from a chip with tri-state outputs that are just always on, so it is possible.
Looks like you'll have to buy faster crystals!
I have an 8 in the same form factor, I have the 25.175mhz one I bought for the vga circuit around somewhere.
w00t!
Thanks! It gets faster in the future!
Zoom zoom!
Thanks! I get the clock rate a little higher later!