Rep Rap vs. Klipper
-
2 years ago, Bambu Lab stunned the 3D printing community with their radically new, game-changing tech. Following this, Creality and Elegoo introduced printers running Klipper, with stepper motion divided into smaller divisions, delivering higher quality prints at much faster speeds, leaving older tech printers in the dust.
Take for example, Elegoo's Neptune 4 Plus, which runs Klipper with a 64-bit, quad core processor and 8 Gigs of storage on its control board. With its print head's accelerometer sending feedback to its control board, this printer provides advanced features like input shaping, pressure advance, auto bed leveling, and much more for a price of only $350 US. Gasp! Can anyone doubt the magnitude with which the 3D printer manufacturing industry is being disrupted?
Recently, I've been looking at Modix, because a larger printer is on my short list. I like Modix, because it uses components known for very high quality, like Hiwin Rails, MeanWell power supplies, and Duet boards. However, from what I can see, Duet boards use older tech, 32-bit processors, and Modix printers aren't running Klipper. Still, I'm unsure if Klipper on 64-bit processors can produce higher quality, finer stepper motor divisions than RepRap firmware can with Duet's 32-bit boards.
Can anyone tell me if Duet boards need more powerful 64-bit processors, to compete against the disruptive tech coming from China? Disregarding the issue of speed, I'd like to know if Duet's current tech and RepRap firmware are sufficient to match the print quality we're seeing from printers like Bambu Lab's X1 Carbon.
-
@Ebbesendavis The microstepping between klipper and RepRapFirmware is the same since that's part of what the stepper drivers deliver, and apart from Bambu's proprietary solution most if not all printers use Trinamic drivers today.
There is literally no upside from using 64bit processors. The 6HC or the STMH723/743 printer boards supporting RRF are perfectly capable of producing high enough steprates comparable to klipper setups, without necessarily introducing the dependency on a USB connection and an operating system lacking real-time control, which leads to a number of question marks about operational security when (not if) the 3d printer has some controls failure.
In terms of print quality, the "magic sauce" in klipper based printers mostly is Input Shaping. RepRapFirmware supports that too, but the results don't match klipper's and Marlin's implementations today (it's getting better though with some improvements in the latest 3.5 prereleases and future changes in 3.6 that will match klippers approach where speed is attained in exchange for lack of sharp corners).
I cannot speak for Modix' setup because I don't know about it, but in a commercial machine I would prefer a fully supported open source hardware solution over the race-to-the-bottom boards we find in most cheap offerings, often with bad or even unsafe klipper configs and not necessarily an easy way to modify those configs without losing vendor support.
Bambu provides 3d printing appliances where you relinquish control over your machine mostly to a third party. I won't judge you if you go for that, but it's not how I want my 3d printers to work.
-
[Sorry for the rant. It is mostly addressing the RRF devs]
@oliof I'm with you, but it lately feels like we're dinosaurs.
Over the last few month I felt with growing concern, that the Duet-forum has lost many regular contributors.
The Klipper (r)evolution might be part of it? Or maybe the "goldrush" times for DIY printers are over?
People want to buy plug&play printers instead?I'm not saying RRF should provide auto-setup configs, but TBH the config tool development has been neglected for a long time and it was never up to date with RRFs rich features.
The forum was a place to present and discuss (sometimes crazy) ideas and mods for our DIY printers.
Now the majority of threads is about problems with configuration or lost connections and odd stuff happening with RRF3.5RCx.My last big contribution was the hashPrinter. It received great interest in the community, but it's more than two years ago.
Wouldn't it be the right time to implement a unique feature like "4-head simultaneous" printing?
Sure, some of the problems are in the non existent slicer SW. But there must be slicer-devs who want to collaborate?
I'd be happy to help building a market-ready hashPrinter. -
@o_lampe it's not a rant, it's a legitimate concern. Compared to Klipper (and even Marlin), the hurdle to get to a dev environment for Reprapfirmware for Duet boards is high (and for Linux reasons I can't get one installed right now at all, even though I contributed a write up how to do so in the past). The build environment for the STM fork is way more accessible. The Klipper approach attracts people with a 'modern' view on tech stuff ('move fast and break things' and other bad ideas), and those people also are the ones with fungible income. So that's where the hobbyist/enthusiast market tends.
With the advent of cheap SBCs, Klipper has become an option for the race-to-the-bottom market, so enthusiasts won't come from a botched Marlin build and then decide whether to go door 2 (RepRapFirmware) or door 3 (Klipper) when they try and fix up their printer. They will try to fix up their door 3.
RepRapFirmware still has a number of technological advantages (standalone mode, CAN FD, more concise macro language, actual real time support, better support for more kinematics and experimentation), but those aren't relevant for people where Klipper is "good enough" for a ton of use cases. Klipper's main advantage again is accessibility for software people, and software always wins over hardware.
<>RatRig V-Minion Fly Super5Pro RRF<> V-Core 3.1 IDEX k*****r <> RatRig V-Minion SKR 2 Marlin<>
-
@oliof said in Rep Rap vs. Klipper:
and software always wins over hardware.
I agree with everything you say apart from that one final sentence.
Silk purses and Sow's ears come to mind
A couple of examples. Multi point bed leveling using multiple lead screws. It isn't difficult to build a bed that is level and stays that way without any adjustment, driven by 3 screws, a single motor and continuous belt. Yet on the basis that "software is always better than hardware" people continue to adopt and use multiple screws and let software take care of things. Then whenever power is cycled the motors jump in different directions by up to 4 whole steps, meaning that the bed leveling procedure has to be run every time the machine is turned on, which is just a waste of time.
It also isn't difficult to build a bed which is flat, tram, and stays that way. Yet people insist on using (software) mesh compensation instead. The real crazy ones buy expensive tooling plate then because they don't know how or are too lazy to adjust the gantries such that they are tram with the bed, still use mesh compensation. A quick glance at some of the forum posts will reveal a plethora of issues that people have getting the software to work. Even if they manage to get it working, it is seldom consistent because the mechanical issues that it's supposed to compensate for can be variable. So the mesh has to be redefined. Having to constantly adjust the Z height while printing a single layer, also leads to unnecessary wear on the lead screws.
It took my less that 2 hours to get my bed flat, level and tram within less than 0.03mm over the entire 400mm x 400mm area and I've never had cause to make any adjustments since I built the machine. From what I've seem from posts on this forum, the "software always wins over hardware approach" would take up much more time over the life of the machine.
IMO, if you build a machine that is mechanically crap, it'll always be a crap machine, regardless of the software (or indeed the computing power of the control boards).
-
@deckingman said in Rep Rap vs. Klipper:
A couple of examples. Multi point bed leveling using multiple lead screws. It isn't difficult to build a bed that is level and stays that way without any adjustment, driven by 3 screws, a single motor and continuous belt. Yet on the basis that "software is always better than hardware" people continue to adopt and use multiple screws and let software take care of things. Then whenever power is cycled the motors jump in different directions by up to 4 whole steps, meaning that the bed leveling procedure has to be run every time the machine is turned on, which is just a waste of time.
I want to point out that I said software wins over hardware, not software is better than hardware. I am totally with you that proper builds requiring less algorithmic goo to make things work are inherently superior, but in a market that is governed by race-to-the-bottom economics, better does not win, cheaper does. And software is (seemingly, at first) cheaper than proper hardware solutions (until people forget how to engineer).
-
@o_lampe @oliof The thing with projects like RRF and Klipper is that innovation tends to happen early on in a products life-cycle. At that stage it is easy to make big changes that may stop existing features from working or that cause problems for one or two of the folks that are using the software. Early in a product's lifespan the users tend to be enthusiasts and will put up with changes, they are also knowledgeable and can contribute to fixes and new features. However as time moves on if a product is successful then there are more and more users, using the product for more "important" things and those users tend not to like having things not work. It gets more and more difficult to make those big changes. More and more developer time is spent identifying and fixing issues and less on those new features.
I'd say that RRF on the Duet ecosystem is well into that later phase these days and to some degree RRF on other hardware is getting there. Marlin has probably been there for some time (and also suffers from supporting a huge range of diverse hardware). Klipper I'd say is now also now entering this phase of its lifetime. I think you can see evidence of that in terms of the main klipper repository. Getting new features and changes accepted by Kevin has become increasingly hard as he concentrates more on the bigger range of systems he has to support and is prioritising stability. You can see evidence for this in the rise of "danger-klipper" and to some degree in the number of pleas for what the route is to get new code accepted. It will also be interesting to see what happens to Klipper now we have a lot more commercial products shipping with it as standard, will the makers of those products support the firmware going forward, what do they have in place to support some of the forks they have created? It's a tricky thing to fork an open source project and keep things up to date with fixes etc. but it is also not easy to support customers who expect a product to "just work" if those customers are also pulling new updates from a project over which you have little input or control.
-
@deckingman So you spent less than 2 hours to achieve what you feel is perfection. If I remember right your bed is not a flimsy stamped sheet of aluminum . I as one of the crazy ones purchased a cast bed and went to a three motor hydra arrangement. Odd that I am not having issues I suppose.
I do run an initial Z endstop along with finding A and B endstops and that is it. I can sit flipping the power off and on over and over and I am not seeing this motor jumping in different direction issue. Only motor that will run briefly is the hotend fan. And that could be the toolboard booting up that does that. Most who have issues with bed being knocked out of adjustment to me are having it due to multi start lead screws.
I ran a Z tilt then bed mesh a month or so back and when needed can run them again. Not sure of total run time but do know 11+ reels of filament have fed the printer since last level. And it is still printing a perfect initial layer. Maybe I should flip the power switch off and on more?
I did spend time making sure as this printer was built that all measurements were right and it was square. So no playing with the gantry and squaring issues like the first that was created out of an Ender 5 plus. That also had a bed that could and would change shape depending on how hot it was and the air temperature. And that did need frequent leveling until it too went to a cast bed. But even with linear rails I finally was forced to tear it down and rebuild the frame . And finally decrapped it.
I prefer Duet boards because I do not need to add a SBC to them. Thus no need to maintain to me a somewhat fragile ecosystem on a SD-card setup. One that if I were to flip the power off and on again multiple times could do way more harm than simply knocking a stepper out of alignment.. So odds of me ever adding a SBC to the Duets is lsim to none.
-
@gtrider The reason you aren't having problems with motors getting out of sync when cycling power maybe due to printing smallish parts in the centre of the bed. I frequently print large parts almost edge to edge on the build plate so bed levelling is far more critical.
The thing that most people forget, or don't know, is that levelling a bed makes it a plane which is level with a plane described by the attachment points of the lead screws. However, the gantry's often run on frame members which form another plane, separate to the bed and lead screw planes. With the best will in the world, it's practically impossible to build a machine whereby a plane described by the gantries is parallel with a plane described by the lead screw attachment points within (say) 0.1mm by eye. So some adjustment of the frame is usually necessary to get the gantry plane parallel with the levelled bed. This process is called tramming and is often overlooked.
Usually what happens next is that people probe all 4 corners of the bed or attempt a print, after the bed has been levelled, only to find that one or more corners appear to be high or low. They then assume that the bed must be distorted and resort to some sort of software compensation when on fact all that is required is to do the tramming process.
Having extensively used machines which required software compensation and those that do not, I know which I prefer. As well as the time taken to run the software compensation, one must also take into account the time and filament wasted if the compensation is not regularly run and prints fail.
I agree with your comments about a separate SBC. -
@deckingman Your explanation is spot on.
But I wonder if it also happens to Klipper users, buying off the shelf printers?
Can we ship a factory calibrated printer across the globe, unpack it and expect it to still be in perfect shape? (literally)
I guess, when those Klipper users see imperfect 1st. layers they don't dare to touch the printer frame. But instead they are glad to have three (or four) bed-mounts with individual motors.
With regards to off the shelf printers, bed tramming and mesh levelling is a pro argument. It's just a band aid, but very welcome during warranty phase. -
Some of the latest Bambu innovations though do seem to be incredible. Like the latest one where they have moved beyond touch control to a thought interface. Just look at the printer and think of what you want and it will search through Makerworld for the item and start printing. I heard shortly it will be able to also search through Thingiverse and Printables and import the print into Makerworld and set it up for you by just thinking about what you want. That and the soon to be released replicator tied into the AMS means that it will be able to print models in any filament color and type you desire. Not sure how Duet or any printer manufacture is going to be able to complete.
Just avoid thinking about Stay Puft, that did not go so well for another group.
-
@gtrider Nice one - happy 1st April.
-
@gtrider Yes, indeed, and they will do all of these, with their new 600x600x800 enclosed, active heated, glass bed printer, wich I heard is on presale today. I think it will be over in 1 hr...
-
@deckingman The new AI controlled CAD program that was announced today should complete the Bambu takeover of the 3D printing industry. It will be seamless, for example you want to print a fuel cap for a 1947 Clinton single cylinder engine tank. First Makerworld will attempt to start the print using existing data from itself. Failing that it will then move to the competing sites and search. If it does not find it Makerworld will then use a Chinese supercomputer to create the desired fuel cap.
Since almost anything created or designed elsewhere is known by their supercomputer it will be able to complete most request. Asking for a model of something frowned upon by authorities though might be a problem. No replicas of the Goddess of Democracy for instance.
Gasp, can you imagine the disruption just these abilities alone will cause to the 3d printing industry? And that is not even taking into account upgrades. For instance it is rumored that the engineers at Bambu are working on a dimensionally transcendental workspace inside the printer. This will enable them to print objects larger than the printer. How they will manage to remove the prints is the sticking point at the moment. Similar to the person who builds a 40 foot sailboat in his basement.
But other upgrades such as the satellite system to communicate between China and your printer will be easier. It will print out the plug and play module that you will then replace the existing control system with.
-
@gtrider Sounds good, but I've heard the AI has already started to produce these wishes itself, and only sends us a link to TEMU, where we can buy it for 0.69 $ with free shipping and a ton of extra freebies on top.
