EEVblog

Which produces the better PCB layout, a human or a computer?
By popular request, a walkthough to see what the Altium Situs Autorouter can do on the Nixie tube PCB layout. Does it beat Dave's human layout?
How useful are autorouters?
Forum: www.eevblog.com...
EEVblog Main Web Site: www.eevblog.com
The 2nd EEVblog Channel: / eevblog2
Support the EEVblog through Patreon!
/ eevblog
EEVblog Amazon Store (Dave gets a cut):
astore.amazon.c...
T-Shirts: teespring.com/s...
💗 Likecoin - Coins for Likes: likecoin.pro/@...

[@guillep2k]

I use autorouter in highly populated boards to check how good is my component placement. If the autorouter completes or gets near 98%, with only a few contentions remaining, then I kown the placement is good enough to proceed to manual routing (I scrap everything done by the autorouter, of course). If the autorouter has a bad time trying to connect the board, it means my component placement is horrible and I should reconsider something.

[@adamruck]

That makes a lot of sense.

[@m4d3ng]

Make autorouting a CAPTCHA ... Profit

[@theLuigiFan0007Productions]

Yes. That's a really good idea, provided you have some sort of AI compare information from multiple users. I'd say use autorouting captchas for really important things, where you have to make absolutely sure it's not a bot. It'd certainly be a 1-2 minutes kind of thing, requiring a decent amount of mental effort to be put into it.

[@theLuigiFan0007Productions]

Juan Herrero I never thought of the trolling possibilities. XD

[@tpowell453]

That was weak, man. Very weak.

[@ArthursHD]

Captcha would need to be put on very specific sites, another solution is to copy the Duolingo model and teach people how to design PCBs Automatic or Semi-Automatic PCB builder could be next level AI. Imagine software that could draw a design from schematics, suggest better IC's based on specs and price, optimize existing designs to minimize size, increase efficiency, etc.

[@Gameboygenius]

Request: Try routing the same board using other autorouters such as Freerouting (an open source one) and TopoR (an "organic" autorouter with a good reputation).

[@superdau]

That's no use (as Dave said at the end). Unless everything is exactly the same (and by exactly I mean numerically identical) any difference in the result could be attributed to the different starting conditions. That's just how algorithms work.

[@Gameboygenius]

+superdau I disagree. While it wouldn't lead to a rigorous scientific truth, it's still a useful starting point. If one autorouter can't complete the board with default settings while the other can and even does it neatly, then that's at least an indication for which autorouter you might want to try if you need to. Yes, the "inferior" one might be able to perform better if you tweak it, but the other thing to keep in mind is that time is money. A video on the topic could be more light-hearted, like this one, and just do a scratch and sniff of different ones.

[@andhag]

I second this. I've had some great results with FreeRouting, to the point that I've only changed stuff for visual appearance before making my board.

[@gxcreator]

+1 for TopoR

[@CZghost]

EAGLE has quite decent autorouter (yep, even free education version). It tries to not violate the design rules and it is also aware of acid traps and always constrains to 45 angles. The only thing I found rather silly given how proffessional editor EAGLE is, that's that fact there's only a global design rule for every route. You cannot set design rule for individual routes (let's say power routes and routes where you need thicker copper route to allow more amps flow through. Just to clarify, I use the free education version of EAGLE, so I don't know how many features I am missing from paid versions. But freeware alternatives have that function to assign different design rules to specific nets, so I guess that wouldn't be much an issue to have that feature in EAGLE, too.

[@roboterson]

"Quick follow up..." Ok Dave.

[@EEVblog]

It could have been an hour...

[@technodaz]

It could have been 2 and we still would have watched it.

[@thcoura]

I can grab the sarcasm flying in the air

[@DoRC]

Roboterson 30 mins is quick for this channel:)

[@SupremeRuleroftheWorld]

we would be suprised if it would be less then 30 minutes.

[@Conenion]

The problem with autorouting is that it is a multiple(!) NP problem. If you would simply want to route a _single_ net this is already the Steiner tree (with obstacles) problem which is NP-hard. I wrote a simple autorouter during my studies. It is hard, really really hard. And mine was a very simple one. Could just do rectilinear and DIL. It had rip up and retry, though.

[@illyadass4862]

Of course, and the key to success is to forget about true solution and search for good enough.

[@oraz.]

I was wondering that, just dimly, if it's actually TSP type problem.

[@Conenion]

@orazdow TSP is a different, but also NP problem. You /can/ route a net in a TSP style, but note that this would not give you the minimum length for the traces. The minimum rectilinear Steiner tree (MRST) introduces what are called Steiner (or Hanan ) points to allow for traces to go the shortest path to a Hanan point. en.wikipedia.org/wiki/Rectilinear_Steiner_tree (Obstacles "simply" increase the number of Hanan points.) Note that, although the MRST is NP, there a good heuristic approaches of which it can be shown that they are not far from the optimum. A much harder problem is to figure out the order of connections to route.

[@yurikirsanov8763]

Yes, Dave, what do you think about Topor? Topological router that uses free-routing style? You can use that with Altium and there's a free 'Lite' version of it, maybe give it a go? Thanks for your great videos anyway! :)

[@garrettscott8886]

Thanks for the video Dave! I am a relatively new PCB designer working on a project for the US Atlas Collaboration and have been using Altium for our front end electronics design. We are designing a 16 layer board that is 2.5 in by 9 in and it has 10 custom bga asic's on it. Lots of high speed differential signals lots of cross talking power lines! Its always nice to see that some of the issues I have are not unique to me and my lack of experience!

[@bbreeuwer4577]

I have been a professional designer for many years and I think you have been extremely generous to an autorouter in your video. The only way it really works is for local routing with bga components etc. Otherwise it's just the silliest invention ever. Most of the time you will spend more time, energy and frustration setting everything up, cleaning up traces and making sure there are no bizarre errors. Still it doesn't take safety and reliability issues into account or proper routing against noise which really can ruin your design (even with a nixie clock with a step-up converter for example). It's even worse with mixed signal (analog, digital and power) sections (like professional audio equipment). In the end it's all about how much trust you will have at the end of your design. Auto-routers are not good enough to be completely trusted (and i doubt they will ever be)

[@EEVblog]

Yes, much tidy up is usually required, impossible to trust totally. However if your ultimate goal is a faster layout, then you can make it work in some limited circumstances. I have used selected autorouting successfully, but yes, most the time I've found it a waste of time and effort.

[@martinzhang5533]

For small low freq boards, when copper pouring for ground is done manually, auto router can give acceptable results.

[@atomicpollywaffle]

BINGO!

[@bbreeuwer4577]

Small low freq boards are also done (very) quickly by hand, with much higher certainty. Especially if you already do all the ground layers, it's a matter of minutes to do the rest by hand. The risk of spending maybe half an hour until one hour extra is less of a problem than sending a non-working board to the manufacturer (with all the hassle of troubleshooting it later on)

[@Niosus]

As someone with a background in computer science, the algorithm used here seems pretty dumb. It looks like it's taking a greedy approach with backtracking when it gets stuck (which is why it gets to 50% really quickly, but slows down dramatically near the end). That's just a really poor strategy for these types of problems where every part of the solution brings along more constraints. What you really need for a problem like this is a system that can detect patterns and work out partial solutions that work well for those specific patterns. Something that uses a (well-designed) genetic algorithm would not get stuck in a corner so easily. Alternatively I'm pretty sure you can also apply neural networks to this problem effectively, since they essentially tackle problems in a very similar way that humans do: You break down the problem in smaller components, which you break down even more until the solution becomes trivial. However those techniques are still very much on the cutting edge and incredibly difficult to get working properly and reliably. You also need boatloads of training data to train them, but with PCB layouts it seems like there should be plenty of learning material in easily processable formats. I expect that the capabilities of auto routers will rapidly increase to match and exceed human capabilities in the near future. Depending on the level of R&D devoted to it, it may take another 10 years... But it'll happen much more suddenly than you'd expect. At this point it's really just a matter of which company is willing to commit to such a project to make it happen.

[@DoRC]

Thanks for the followup. I'm one of the comments that asked this question and I'm definitely convinced in this instance at least.

[@piratk]

I think you said it in the last video, that it would have been possible to swap some pins around to get better paths, however, that is a feature that would have been very useful for the autorouter to know about. Something like "This group of pins needs to go to this group of pins"-feature, when software definition can take care of the rest. Usually when I think of features like this for programs, I start to look for it in the software, and most of the time, I actually find it, already implemented.. :) Also, the ground-fill feature you wanted, must of course already exist, it should just be a matter of instructing the autorouter.

[@bkiffter]

"You never go full Autorouter"

[@miguelcastorena4293]

LOL

[@tohopes]

Sounds like connecting a congested net can take a lot of vias. Do you ever worry about the number of vias in a path (for low-frequency stuff)?

[@EEVblog]

Generally not a problem, no. Often when you get to the last few traces on a complex layout you'll desperate and use a ton of via on a single trace in a desperate attempt to get from one side of the board to another.

[@pornpori]

I would probably worry if the path will carry a high current, but any good designer will identify those "special" tracks and route them first.

[@kay486]

why exactly are right angle traces bad? is it because the electrons are too fast they would fly off? jokes aside, I am genuinely curious

[@jope7137]

They are not bad. Neither for signal integrity nor for modern PCB manufacturing. The only reason is aesthetics.

[@unaliveeveryonenow]

Some stupid people read somewhere that right angles might be bad, but they did not read the second part of the sentence (might be bad at microwave frequencies) and now everyone is forced to honor their stupidity.

[@TheDuckofDoom.]

he expolained it, manufacturing problems with the etching step

[@MarkTillotson]

Not a microwave engineer, clearly.

[@MarkTillotson]

[ Modern logic chips run at MW frequencies, that's the point really ]

[@bami2]

auto-spaghetti

[@0ADVISOR0]

gave me a good laugh, good day sir!

[@Gameboygenius]

When I cook spaghetti it actually usually comes out less tangled than this. And as a devout Pastafarian, I'm disturbed by your insult to my faith. In retaliation I will... do nothing but perhaps offer you noodles for lunch if we ever meet.

[@EEVblog]

Ramen.

[@ericmin6055]

EEVblog mom's spaghetti

[@physnoct]

@@Gameboygenius "And as a devout Pastafarian ..." May the sauce be with you!

[@zodak9999b]

So what's needed is an autorouter with an option for multi-chip pin swapping where you can say "All these connections are software definable, interconnect any of these inputs to any of these outputs, one to one, updating the schematic when done".

[@dtiydr]

I have read that with autorouting of two layers so is ground plane on both sides the right way to go and the autorouter get it much easier since its pretty much made to it with that as well.

[@kevtris]

I don't think those little tiny SMD resistors are a good choice for the anode resistors. They will be dissipating some power (depending on the HV rail) and I am not sure if they are rated for the voltage across them. 3 resistors of 1210 size in series would've been a bit better margin.

[@Yotanido]

What do you actually do when you need a 3-way junction on a trace? Just deal with having at least one 90° angle?

[@EwanMarshall]

Better to go through the pad at one or the other end of the trace and onto the next one rather having a junction in the trace.

[@petehiggins33]

Or add 45 degree fillets in the corners maybe?

[@TheDuckofDoom.]

360/3=120. Fillets radiused or angular are another option.

[@userPrehistoricman]

Dave used 90-degree ones on his manually routed board for the pad common to all of the tubes

[@sparkplug1018]

It seems like when your routing power traces a lot of engineers don't mind the 90s for that. Which is fair enough, its just power.

[@the2d]

I recently did my first ever pcb design and used the auto router. Hopefully it works fine. Just a little custom diy audio sampler board.

[@laernulienlaernulienlaernu8953]

I couldn't tell the difference between the auto router and your version!

[@Wren6991]

I route in descending order of importance (think this is a common strategy), and sometimes I get to a point in a design where I just have slow signals left, and I let the autorouter loose (followed by some tidy-up with the push-and-shove tool). I prefer not to though. Normally I find that the failed routes are hard to find sensible paths for, because the router has backed me into a corner. Your ground fill is also chopped into tiny pieces, although this is less critical for >2 layers.

[@Arek_R.]

Acid traps? Never heard about if before. Anyway I never make back-angles so I never had a issue with it.

[@uriba107]

In this particular case, wouldn't flooding the ground before the autorouter runs will cause it to skip most of the ground traces?

[@Gameboygenius]

Not really. The autorouter must still cut through the plane to make way for tracks, and it must also ensure that there's at least a track's width worth of connection everywhere. So in practice, the autotrouter can treat the plane as not being filled in. Filling the plane in after the boar has been routed does mask of the ugliness in the ground plane at least.

[@knssoftware6018]

Can you also make it layout the non-critically laid out components (i.e. everything except the tubes and socket)? Any routing (manual or auto) is only as good as the placement of the components to start with. It may be able to layout the components better.

[@docdaneeka3424]

hey dave, regarding the 'acid trap', what about 90 degree T-junctions? I would guess it could also be a problem for quite fine traces? What would you do if you had to have a junction of some sort?

[@casperorillian7393]

Forgive me if you have mentioned this in another video. but you you have any recommendations towards circuit design software? The piece you are using in the video looks very expensive so I wonder if you know of any suitable cheap/open source alternatives.

[@pomprocks]

Is there some advantage to straight line and tight corners? Seems it would be better to have curves in many places.

[@flyguille]

the position of the ICs doesn't help to the auto-router. can you set it auto place al components except the nixies?.

[@Spikejwh1]

I never autorouted in my life. Love to route boards manually. Soft music on the background and just route until it says "nothing to do!" in the lower left corner (Eagle) Very relaxing. ;-)

[@Tigrou7777]

why is the ground layer (in blue) a big flat surface covering almost the whole pcb ? Is it for electromagnetic shielding or is there another reason ?

[@necro_nemesis]

Seems relatively slow given your CPU's capability. Do you have all 8 threads of the i7 3770k working the solution? I've had reasonable success within using ground plane polygons I draw in and assign as "gnd" prior to autorouting and ensure all 16 threads of my workstation work out the routing.

[@runforitman]

I've been having fun optimising the layout of a single sided board I wonder if using the auto router could give me some new ideas to add to my own routing

[@LittleRainGames]

i just routed a board the other day the auto router was horrible with. it gave me 1000+ vias, and only 98% complete. i then did it by hand with only 20 vias with no problem at all. it was basically a board with 5 edge card connectors, sharing all the same pins, in parallel, and that was it. it was just one of those designs that are super easy in practice, but kills the auto router for some reason.

[@Wilburley]

Good to know that after the singularity happens I'll still be useful to the robot overlords.

[@minxythemerciless]

If ground is a problem, just remove it from the routing and flood-fill with ground after all the routing is done. That works fine on KiCad.

[@paulrautenbach]

He did try to do this in the video.

[@GRZNGT]

I wander how the process of routing a silicon for IC's look like. Given that some of them can have millions of components, there's got to be a lot of autorouting.

[@kaiklopsch8590]

Hi Dave! Clear win, congrats! In this case you have positioned all the parts, but would it be possible to "auto arrange" all parts on the PCB by software, in order to optimize everything? What would be the needed power (and time) to calculate all this details? Probably some parts would be fixed (locked), but others may be placed elsewhere …

[@EEVblog]

OMG don't even go there! Auto placement is worse than autorouting.

[@VEC7ORlt]

Autorouting we got pretty much figured out - with proper rules it works okay, autoplacement on the other hand is much much harder - it offer much more permutations and conditions arent as obvious, and the program needs to think ahead way more than with routing, it is also possible that it is one of those np-hard problems - where complexity grows very fast.

[@kaiklopsch8590]

Nevertheless, I would be interested to see - even if it fails …

[@niklaswallin9478]

I just feel like some of the ic's could be rotated 90 dungareees.. Is the a function to have the autorouter actually move the ic's around?

[@hyperplastic]

Since you're exploring routing possibilities what about a video on the pin swapping idea? That seemed extremely elegant to me

[@EEVblog]

I might do that.

[@brumbymg]

I wondered about auto-routing .... now I have an idea. Seemed a bit slow to me - but wondered if the screen capture had anything to do with that.

[@konsolkongendk]

Hi Dave. Could you elaborate on the "acid trap" traces and how you would have done them differently. I have only done a few small hobby boards, and I think I missed your point of why the angle of the traces were bad :)

[@gettingpast4391]

Well this was enlightening. My friends who have much more experience tell me they never use autoroute. I always wondered why.. surely it can't be THAT bad. OK well I learned that it can.

[@Darryl603]

Could you give the tubes a bit more breathing room or does the board have to be that specific size? It seems to me that a little more room between the tubes would make a big difference wether it be manual or auto route. Thanks for sharing DJ

[@Darryl603]

Maybe the best approach is to auto route to get most of the rough in taken care of and then manually tidy up any bits you just can't live with. Still save you a bunch of time...

[@EEVblog]

Nope, already at maximum spread.

[@neogeo8267]

It's been nearly a decade since I did this professionally, but Mentor Graphics PADS did a pretty decent job when working w/ 4 layer boards (pwr/gnd) and proper rules were setup on each net. Of course, manual routing the tricky bits is pretty important, but auto-routers do have their place.

[@DrTune]

In a shift of focus from mocking component vendors, engineers, products or projects, today Dave mocks an Autorouter. Tune in next week for a withering critique of a paper clip. ;-)

[@nskvasov]

Dave, you should try autorouter in Diptrace (in that case there is no need to re-create the project: it easily exported in Altium Designer / PCAD200x and then imported to Diptrace). Indeed, in the next few years autorouters should become self-studying neural nets! (one part studies and routes, another one - checks and criticizes)

[@VEC7ORlt]

Oh come Dave, you could've at least set the autorouter properly if you want to show it - everyone pressed 'route all' on theirs but seldom do you see any advice on 'how', besides that one was abysmally slow, Diptrace would have done that ten times over (it is surprisingly snappy), and why not showcase something like TOPOR - with its organic flowing traces.

[@unaliveeveryonenow]

Altium's autoruter is so bad it always performs worse when you give it a subset of problem, TOPOR crashes when you are operating it's UI. Use Mentor Graphic's Pads if available.

[@VEC7ORlt]

I do PSU or simple(ish) stuff(up to 1k nodes or so), so I don't need autorouter, so far Topor left the best impression. i.imgur.com/ACUDG.png i.imgur.com/FtrGLCN.png

[@rafflesmaos]

Topor's fantastic, but I wasn't a fan of their recent node locked licensing changes using somewhat sketchy looking DRM. I suppose I could put it into a VM, but eh, previous version still works well enough. Altium autorouter is stone age compared to it, so Dave generalizing all autorouters is definitely unfair.

[@alexeykokh8240]

rafflesmaos And what about Eagle CAD autorouter?

[@nickh5937]

Auto router is a tool. Altimum is a sub par PCB layout tool just like its auto router. If you want a real auto router you should try specctra, just remember it’s a tool and you must learn how to use it

[@aaro1268]

Also worth noting is that altium offers the capacity to use plugins, allowing the use of better routers.

[@techy4198]

Hey Dave, what's your preferred tool and/or standard for calculating minimum PCB trace widths based on current (and temperature rise etc)?

[@RuneInternational]

how about giving the program free hands instead of starting with your final component layout that were changed 100 times during your manual routing.

[@michaelhawthorne8696]

Dave, I've been using Eagle 7.5 for the last year and have found it to be excellent. The auto router has worked very well when I did use it. It also seems faster than your AR too and is a lot more animated. I love using it and when a board is complete with copper pour included on both top and bottom layers, you really feel a sense of achievement. It's even better if you had to design the circuit as well. I notice there is a lot more information on your board tracks than is in Eagle, in fact, Eagle displays nothing unless you highlight a track and look at its properties and I don't think Eagle can do selective auto routing. There is a big price difference though between the two programs.

[@henryD9363]

I wish you turned off silk screen. It's hard to see the blue. Always enjoy your videos thanks

[@TheBdd4]

Thanks, good perspective on autorouting.

[@MAYOJAMESD]

well, I stumbled on this video a few months late so i probably missed my chance at a response, but... I recently embarked on my first home-made pcb project. I went with the auto router in eagle because making 250 connections myself on a 70mm x 100mm pcb while keeping trace widths and clearances relatively high (since I've never done this before) was much too daunting of a proposition. The best auto-route I got out of it has 70 vias which, when added to all of my through hole components (which is all of them), means a crap-ton of drilling and filling in to complete the electrical connections. All of this to come to my actual question, to wit: would hand routing potentially lead to less vias?

[@shallowfakes593]

is it possible to make the nixies show how many traces are left to make?

[@ColinMacKenzieRobots]

Great bake off. I guess nothing has changed since I last tried auto routing. Sooo, do you ever prototype on prototype PCB board, DIY boards or breadboard first? I'd be interested in overview of how you design your first prototypes from a check design to commercial prototype to commercial, etc. I just finished a fairly simple proto board using op-amps and by the time I finished I was like frig that was much longer than just designing it in EDA then DIY etching it and of course it looked like shit but it worked.

[@DreitTheDarkDragon]

Reminds me how my classmates on high school ran autorouter for hours because they were lazy to think. When autorouter failed, they just moved parts a little and tried again. Few hours....on 555 timer circuit!

[@MxWintersAFOL]

Just wondering, is there any design standards with regard to the number of via's on a trace? Does the more via's increase resistance ect by any substantial value that it would effect serial bus lines and the like?

[@EEVblog]

More vias equals lower inductance and lower impedance. Almost always a good thing. Ground planes usually have lots of via to lower both.

[@justinschunick5189]

Hi - I see you are using Altium 17, but you still used the old legacy autoroute tool in this video... You should try the new ActiveRoute feature in Altium....Its actually pretty decent for selective sections of your PCB...anyways love your blog!

[@prashkd7684]

I use to place ground and power plains before running auto router.. then run auto router in selected area for complex zones .. It worked just fine

[@bjtaudio]

I never use auto router, it's too much of a haste to set up all the rules, only to find I'm not happy with the results and have to re work it again manually anyway.

[@jozsefdornyei1253]

This can actually be a lot better programmed. I bet there are auto routers out there what can do this task. Dave might be able to get his hands on a top autorouter. I bet the company would be eager providing him with a trial license - even a real one.

[@TheChipburner]

Will be interesting to try TopoR autorouter for this.

[@MagooHifi]

i am curius if the autorouted board actually works.

[@xd-fu6ry]

I mostly use the auto router todo the data lines and then route the ground and vcc. And fix where the auto router screwed up. It can save a lot of time.

[@ToTheGAMES]

I'd love to see you do a comparison between a few common used auto-routers.

[@vejymonsta3006]

The auto router is good for figuring out component placement. If it gets hung up at 98% or whatever, then you probably have a placement problem. I don't particularly like sitting there routing everything manually. If I'm designing for me, then just make a 4 layer board with ground and power. Then auto route the signals on top and bottom. If it fails, move components. Repeat until it has ZERO issue with auto-routing. But I do find manually routed boards better looking. No issue with that at all, but for low volume or personal projects it's fine to save time on autorouter.

[@iamdarkyoshi]

The work that goes into motherboards, especially server motherboards... Damn....

[@Stefan_Payne]

And since you mentioned your old Intel CPU: Maybe it's time for you to go for Ryzen? ;-) Should come out in about 7 Days or so...

[@sparkplug1018]

Or just go for a older Xeon that will still out perform it. Just a thought.

[@KineticWasEpicVideos]

Also consider: -Core i7 -Core i5 -Core i3 -Core 2 Duo -Intel Atom -Raspberry Pi -Raspberry Pi zero -TI-83 graphing calculator -AVR microcontroller -PIC microcontroller -Bag of transistors and diodes -Toaster -Light Bulb -Brick Any of these options should be much more capable than AMD ryzen.

[@theLuigiFan0007Productions]

+sparkplug2018 It really depends on the software. Some programs run absolutely horrid on Xeons, so horrid it's laughable. It's the same reason why more then 4-6 cores dosen't lineally improve performance in games. High clock speeds and fast memory and is better then loads of cores in those circumstances. For most people the Ryzen will easily outperform a Xeon, in pretty much every application. Especially as it has 8 physical cores, and can run 16 threads if necessary. But if you're running a server, data mining/processing or video encoding server that's specifically optimized for lots of cores, the Xeon could perform better. Especially the Xeons with around 50 cores, if it can effectively make use of every single one of (or at least most of) them.

[@theLuigiFan0007Productions]

+KineticIsEpic Nope. Go read some benchmarks. The mid range Ryzen CPU matches or exceeds the highest end consumer i7 processor, which costs just over $1K, in ALL current tests. The i5 and i3 don't stand the slightest chance of beating it, and the others you mentioned, you're just a troll. Please leave if you have no interesting information or valid counterarguments backed by facts to the conversation.

[@sparkplug1018]

+theLuigiFan0007 might want to rethink that. The Xeon E5-1660 is a 6 core 12 thread part that boosts to 3.9 Ghz and can be bought for around $150-170. Not sure what programs your referencing, but I don't have any issues with games giving "horrid" performance on my Xeon systems. Yes you are correct though, that games don't leverage the 16 cores and 32 threads in my dual E5-2670 system, but that's not its primary task anyway. Still runs games very well though so I'm not about to complain. Just a side note, benchmarks put the E5-1660 ahead of the 7700K, just thought I'd add that in there if you were curious.

[@alexmihai22]

Hy, let me ask you how much the CPU was charged by that?

[@over2seeyer]

Altum huh?, damn! how much is the license for that? like 100k ?

[@petercrary5411]

You should get the auto-routed board made up and compare the actual usage characteristics of both.

[@andruloni]

I don't think there's much to compare if the full-auto board is not fully connected.

[@Z-add]

When i was in university studying electronics engg. I looked up salaries of various electronic engg fields and pcb designers were the worst paid job. That is why i never bothered with pcb design.

[@demoscenes]

Hey, this was a very nice explanation and really good to know your methods compared to the autorouting methods. Even if Altium can do an okay-ish work it sometimes takes longer ways than supposed, which can cause some losses in the signal-flow. Really great you shred some light on this from many different views! Thanks for the follow up. Some say this was a long one? No way! This is really essential when it comes to get a efficient design that will work! Way to go Dave, this time you nailed it! :)

[@freesky776]

It would be interesting to see comparison and a review with Topor!

[@Mriks007]

I think the autorouter tool is not created to route complete board from start to finish regardless what rules you set

[@soothcoder]

If you pre-place the fill zone (ground pour), will it us it/cut it as required or will just stuff it up further?

[@SianaGearz]

Routers usually ignore the fill, because the fill is irrelevant to DRC, it still needs to ensure that all signals have minimum widths and minimum distances, so whether it takes fill into account is irrelevant in this regard, as the net is the proof of DRC. However when fill corresponds to the net that is routed through it, it's merged so it visually looks neat and you're no longer disturbed by the dirtywork in the net.

[@Brutaltronics]

what if... your autorouter takes your manually routed boards as input and learns from those using deep learning algorithms?

[@Razor2048]

Why doesn't it do the full ground plane when it is so common to see a PCB with it? Can't they make auto routers that try to get results similar to human results?

[@49studebaker1]

Dave, Have you tried the Electra auto router? I think there is a trial version plugin for Altium.

[@midnightwatchman1]

you are a hard taskmaster, most students would have feared you

[@tyttuut]

I used Fritzing in my first few attempts at board design, and that autorouter was just horrendous.

[@thcoura]

I'm curious what an autorouter used for microelectronics would do in this example. PCB autorouter always had a bad reputation but in digital projects with billions of gates it is an essential project step. In order words, considering that IC autorouter gets more money and effort than compared with the PCB ones ( my assumption ) what would like the final result in this example

[@xbox360Rob]

Eagle autorouter is wayyyyyy better, maybe do a comparison with the same pcb?

[@noel1637]

Wayyyy better than this for sure.

[@aabb5283]

You have used "cheap Chinese-optimized" design instead of high-end one ? It's kind of flawed from the beginning then. The ideal design would be to use one microcontroller for each digit indicator and either MOSFET or other high voltage switching array (bipolar or other similar things can control the current, so potentially dac outputs can be used to control the brightness) or a few arrays for each tube to control each pin without multiplexing, to avoid flickering problem. And probably one more microcontroller to control the whole set of indicators. And the result would be a high-end version of indicator unit.

[@mcconkeyb]

Yup, the Altium auto-router is not good for doing a whole board from scratch. But then most struggle without a good setup. Many years ago I used the Mentor Graphics auto-router and it worked OK, but still not as good as a person.

[@Porama6400]

Is Eagle auto router better?

[@thenewfire]

Very surprised people aren't pointing out how much better Specctra Autorouter is. It is often used in industry and does an excellent job. Lots of discussions about it online.

[@PedrinbeepOriginal]

What about ELECTRA from Konekt?

[@bur1t0]

Unless the autorouter runs on multiple cores (*very* unlikely, pathfinding is not friends with parallelism), the screen capture wont affect the performance of the autorouter one little bit.

[@computerizedcat]

You should produce this as a one off and compare it to the board you routed, see if it actually causes issues with the way it works.

[@didiswiss]

Altium Designer? What has happened with KiCad? Have you abandoned it?

[@ytmmcosta]

I gave you thumbs up! I never trust auto-routers. They lack of aesthetics, some basic rules when dealing with power planes, isolation, acid traps, etc. Once the main items of the board is resolved manually, I might use auto-router to finish the obvious traces where my laziness don't allow me to do.

[@SidneyCritic]

I'm never going to use that info, but that was 2 interesting videos. An autoroute shootout between brands would be good, you just have to keep it to 100x100 because free Eagle is limited

[@suyashrajpure3562]

You are one of the best channel on KZbin.... but please I request you to make your videos short enough.... around 15 to 25 minutes...... This is my feedback

[@Neria2288]

Thank you thank you dark forces! I'm still using protel for win 3.11 and those 2 videos on PCB design gave me lots of tips, even for that dinosaur!

[@pratherat]

G'day Dave, What's your take on EasyEDA?

[@mustafayasiraydin]

I am curious too.

[@pratherat]

@@mustafayasiraydin Jump in! I's free as you aren't using it commercially. You can download the gerber files, or have your boards fabricated by them. They also have rapid assembly capabilities. My question was asked years ago, but my answer is based on recent successes. easyeda.com

[@mosfet500]

Proteus has a better autorouter. Sometimes I use it then go through the board and clean up what I don't like. I autorouted a complex board with a lot of redundancy and it saved me hours of work even going back and cleaning it up.

[@Gabriel-kz6kb]

Still amazing in my world! He did it in 7 Minutes, people would take hours for this...

[@tmd63]

'Acid Trap' I have always known them as "Undercuts"! But maybe that is a UK term.

[@uni-byte]

When I'm in a hurry I'll let the auto-route have the first go at it then spend a little time to fix-up the most egregious mistakes. This can be quite effective and efficient.