Stack Exchange
log in

Kartman

 Open Source Electronics

Developing an open-source GPU model tailored for ML training e...
Kartman
May 17, 2024 14:26
Maybe look at the Pynq board. There’s so many options from the very simple to the bleeding edge. Arduino Uno? Surely you jest - it’s not even in the ballpark!
Kartman
May 17, 2024 14:22
What do you mean by ‘open to the public’? Xilinx and Intel have a variety of boards and there’s many others. Some are even open source. There’s Pynchon
Kartman
May 16, 2024 12:12
If you want to get a taste of high speed systems, get yourself a fpga dev board and go crazy with your vhdl/Verilog and see how much performance you can squeeze out of it. You’ll be nowhere near NVIDIA, but you’ll have a much better understanding of what is involved. You never know, but if you work hard at it, NVIDIA might want to hire you!
Kartman
May 15, 2024 14:49
Your knowlege of memory systems is lacking. The fastest memory is sram the highest density memory is dram. Cache memory is sram. Your 80G of store is dram. The registers are most likely sram. You could conceptually design a system to equal or better NVIDIA. How are you going to actually make a chip that implements your design? You’ll be needing millions if not billions of dollars to pay for the fabrication of a bleeding edge chip. NVIDIA had to invest billions to get their level of performance.
Kartman
May 14, 2024 13:17
In the case of a cpu like the RISC-V you mentioned. The cpu is relatively simple. Now you have to hook it up to some memory in the outside world and try to keep your cpu fed with instructions. That’s where the problems begin as the cpu is much faster than the memory. This is where the concept of caching kicks in. To implement a cache you need a slab of high speed sram and a lump of logic to manage it. The cache has now consumed way more transistors than the cpu it serves.
For your gpu/AI accelerator, you’ll need lots of high speed sram for intermediate storage and cache. I’d suggest you sea…
 

 Discussion on question by Silverspur:…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Feb 5, 2024 03:12
For learning pcb design, I find it useful to look at commercial circuit boards. Things like washing machines etc have similar circuitry and requirements to your circuit. You can see how the designers have addressed the many requirements. Learn from the best! It seems these days one picks up information from YouTube vids - the problem is that many of these channels just rehash the same bad habits.
Kartman
Feb 5, 2024 03:12
@Silverspur. As such, 45degree tracks are ok, but if we look at the neutral track on the right hand side of the pcb, you’d usually make the tracks between the connector pin and main track at 90 degrees. Nothing to say you ‘must’ do this, but is more conventional. The ground fill is normally used when you have digital logic to form a ground plane. This isn’t the case with your circuit. Again, nothing to say you shouldn’t do it, but in your case it is not necessarily advantageous, especially with clearance issues. Optos are really only useful if you have separate, isolated supplies.
Kartman
Feb 5, 2024 03:12
K1 Com2 is way too close to the ground fill. As well, use slots (cutouts) to extend the creepage distance. You can use normal diodes like 1N4004 across the relay coils. The optos are doing nothing really. As both sides share a common gnd. No protection is gained. You really don’t need a ground fill and your angled tracks aren’t something you’d normally see on a commercial pcb. You can make the tracks fatter if required and run a parallel track on the other side if you want more current capacity.
 

 Discussion on question by Franklin Kn…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Aug 21, 2023 18:42
How about some big resistors/resistance wire? With little electronics experience you don’t want to mess with designing buck converters.
 

 Discussion on answer by Justme: How t…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Nov 10, 2022 14:43
CE means ‘China Export’ not conformite European.
 

 Discussion on question by Pritesh gup…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Nov 7, 2022 12:56
When you lose power the stepper motor stops. When you restart, you need to determine the current position as mechanical systems have inertia, then recalculate the system trajectory and then get the system up to speed again. You need to store your system state, not the cpu state.
 

 Discussion on question by Daniel Van …

Imported from a comment discussion on electronics.stackexchang...
Kartman
Aug 18, 2022 19:02
Use your ‘scope to view the current waveform. If the choke is saturating, you’ll see sharp changes in the current as the choke will cease being an inductor. Measuring the current can be a challenge - make another winding on your inductor of a couple of turns. Put a 100 Ohm resistor across it and feed into your ‘scope. Poor man’s current transformer.
Kartman
Aug 18, 2022 19:02
If you’re feeding rectified AC through that inductor it may actually be saturating. If this is the case, putting more turns won’t fix the problem.
 

 Discussion on question by Thomas Wilk…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Aug 17, 2022 14:13
Sounds like your clock source might be the problem.
 

 Discussion on answer by Tony Stewart …

Imported from a comment discussion on electronics.stackexchang...
Kartman
May 3, 2022 05:19
There must be a zillion examples of a uart in verilog - its pretty much the standard peripheral to design in fpga courses.
 

 Discussion on question by dev_000: I2…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Mar 17, 2022 05:46
With the low cost of WiFi microcontrollers, they are a cheap solution for moving lots of data at a swift pace. They're not without other issues and it seems like a brute force solution but nevertheless a viable one.
 

 Discussion on question by Craig Calfe…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Feb 2, 2022 18:20
@linuxfansaysReinstateMonica - i gave enough Google hints. Actually I was thinking of a resolver, but said synchro. A resolver is 3 phase.
Kartman
Feb 2, 2022 18:20
You can’t really beat the old synchro. Engineer it so there is no mechanical adjustment. If you need to eliminate tolerance, then do it in code. Being a rotary transformer, it is made from the same stuff the motor is, so no issues with temperature etc. Downside is the interface needs a little more hardware and software. And redundancy is built in!
 

 Discussion on question by mrneedshelp…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Oct 1, 2021 13:55
lm1117 should be more than adequate current wise. What is the input voltage when the esp01 is connected? Have you fitted the required capacitors for the lm1117?
 

 Discussion on question by dtn: Altern…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Jul 9, 2021 12:30
Sorry, your graph only serves to confuse. The axis have no units. What are we observing? femtovolts/parsecs?
Kartman
Jul 9, 2021 12:30
Time in units of what? Solve one problem at a time - you’re trying to solve more than two.
Kartman
Jul 9, 2021 12:30
I think you might be combining the concept of pwm with your control loop. The frequency of the pwm is much higher than your control loop so consider them separate. For the purposes of simulation you may be able to consider pwm as a linear black box.
 

 Discussion on question by wootie11: P…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Jul 8, 2021 18:09
What is the reason for 3 clock generators? And them not being synchronised. Surely the teensy could do the job. Personally, i would’ve leaned towards the esp32 as it has the radio as well. I see no readon for a 4 layer pcb, 2 layer should be more than adequate as you’ll get a good groundplane under the led drivers.
 

 Discussion on question by Oli: TRIAC …

Imported from a comment discussion on electronics.stackexchang...
Kartman
Jun 18, 2021 17:55
@Oli, your code has a number of defects. You do not need to disable interrupts in an isr on an AVR as the already does this. Do not use |= on TIFR, just use = to clear a flag. It is much simpler to use a diac, resistor and capacitor for a triac soft start circuit. You may be exceeding the triac’s I2T rating, an inductor would help here. You probably need a snubber as well to clamp any high voltage spikes. The triac will also need to be heatsinked. A dual thyristor may be required at this current level rather than the triac. This should have a much better I2T rating.
 

 Discussion on question by Shocked: Ar…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Jun 12, 2021 13:35
Sure there are. There's valve opamps if you wish to research, then there's transistors then we came to integrated circuit opamps.
 

 Discussion on question by Veysi ADIN:…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Jun 11, 2021 13:46
You have two relays but no flyback diode? Driven by a logic device? Do you really want your ground fill around the relay contacts and the connectors? I’d be keeping the ground fill well away. The 2N3904 is upside down. But is it really a 2N3904? The parts seem to be placed randomly - this is not necessarily bad, but having some consistency makes the board look more professional.
 

 Discussion on question by TheCodeNovi…

Imported from a comment discussion on electronics.stackexchang...
Kartman
May 13, 2021 13:53
Google MOV 20mm 30V
Kartman
May 13, 2021 13:53
What is the frequency of the pwm? Generally I’d put a varistor (MOV) across the motor terminals as close to the motor as possible. Something like a 20mm 30V device.
 

 Discussion on question by Marcos: Get…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Apr 24, 2021 01:40
@marcos. You only put what information YOU think is relevant. I gave you the answer early on - maybe I wasn’t explicit enough. add a 10k resistor from the output to 24V (ie pullup resistor) then add your low pass filter. Simulate. Vary component values to suit.
Kartman
Apr 24, 2021 01:40
@Marcos - you can Google those two terms. The first term is of more use to you.
Kartman
Apr 24, 2021 01:40
@Marcos, this extra information should be added to your question. The general idea of Stack Exchange is based on the notion of a concise question/concise answer. This isn't Facebook.
Kartman
Apr 24, 2021 01:40
Add a pullup resistor. You've not told us what PWM frequency you expect to be using and other things like what sort of output impedance would you like etc. Armed with this information, we can give you better guidance. Note that 'low pass filter' doesn't really tell us much. Show us exactly what you did.
 

 Discussion on question by oğuz kaan ç…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Apr 10, 2021 05:55
Your clamp meter may not like reading pwm and will lie to you. The BTS7960 is expected to get hot if there’s 11Amps flowing. It is not a perfect switch.
 

 Discussion between Varonne and Kartman

Imported from a comment discussion on electronics.stackexchang...
Kartman
Mar 26, 2021 12:49
I'm GMT+11. Time for bed.
Kartman
Mar 26, 2021 12:48
use Fritzing or similar to draw the layout. That way we have something to work from
Kartman
Mar 26, 2021 12:46
optos are useless. you rarely see an opto in an automotive environment
Kartman
Mar 26, 2021 12:31
https://i.sstatic.net/Igdet.gif
resistor is 10K. diode is 1N4004
Kartman
Mar 26, 2021 12:29
https://electronics.stackexchange.com/questions/43498/how-can-i-use-a-12-v-input-on-a-digital-arduino-pin
R1 = 22K, R2 = 10K. 100nF capacitor across R2.
Kartman
Mar 26, 2021 12:24
you need 2 x 10k 1/4W resistor, 1 x 22k 1/4W resistor, 100nF ceramic capacitor, logic level mosfet, 1N4004 diode and a relay.
Kartman
Mar 26, 2021 10:45
what are you wanting to do? What value is the Arduino adding to all this?
Kartman
Mar 26, 2021 10:41
Slow? Do you want to run pwm in order to vary the brightness or just turn the headlights on and off? Mechanical relays are hard to kill, mosfets in the other hand need protection. Infineon have some nice high side switches that might suit. They even have four of them on an Arduino like board. That would take care of the evil you face.
Kartman
Mar 26, 2021 10:41
If all you want to do is turn the headlights on and off, use relays. For the passing switch - relay. The relay contacts across the switch. Have the Arduino control the relay via a small low side logic level mosfet. In my part of town we have a number of auto dismantlers - we pay $2 to get in. Exit with a pocketful of bolts, lamps, fuses and for you, some relays.
Kartman
Mar 26, 2021 10:41
Bad idea to interrupt the grounds. By doing this you open up the possibility of ‘sneak circuits’ ie paths that aren’t obvious. The positive side is switched for a reason. I’d suggest using automotive relays and using small mosfets to drive the coils. One thing to note - the fuses protect the wire. If you short your average piece of wire across your battery (not suggested you do this) the wire will melt and smoke. You don’t want this to happen. So use a fuse or ensure your power is via a fuse and the wire is rated for the fuse current.
 

 Discussion on question by DayDayHappy…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Mar 20, 2021 14:11
normally the three wires are active - or at least active at some time. The neutral would be wired separately from the switches. (note with Australian switches there is a free terminal that can be used to join wires). As for the wiring diagram - there is 4 possible combinations for the switches. Copy the diagram 4 times and mark with a red pencil the switch path. Follow the path in each of the 4 instances and note what happens. I can tell you both switches on and both off result in the light being off.
 

 Discussion on question by Steven Sago…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Jan 21, 2021 23:29
I’d suggest a comparator.
 

 Discussion on question by Joey: Thund…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Dec 26, 2020 16:29
@Joey - why would you think I'm angry? I'm just pointing out the errors as are the others - and there's quite a few of them.
Kartman
Dec 26, 2020 16:29
And no pullup resistor on the reset pin. Follow the recommendations in the Atmel datasheet.
Kartman
Dec 26, 2020 16:29
@Joey - of course you need a groundplane. You’re experiencing EMC issues and you’re breaking the rules! What you think is a simple circuit is way more complex than you realise. AVCC and all the ground pins of the micro NEED to be connected. Single sided boards have the tracks on the bottom.
Kartman
Dec 26, 2020 16:29
Unfortunately, the schematic doesn’t make much sense. What is a ‘gds’? The wiring to the reed switch is acting like an antenna. When lightning hits nearby a large magnetic field is created and your wiring picks it up. Use a varistor or TVS diode to limit the voltage then have a filter with a 10k resistor in series and 100nF capacitor from the port pin to gnd. Your code also need to debounce/filter the input to ignore short pulses. A valid signal is >100ms.
 

 Discussion on question by Robinson-Co…

Imported from a comment discussion on electronics.stackexchang...
Kartman
Dec 22, 2020 20:18
You haven’t solved the problem, you’ve fudged it. Understand the problem first.