The Original Prusa MMU3 is our efficient solution for the MK3S+ and MK4, enabling these single-nozzle 3D printers to print using up to 5 filaments. While the MMU3 for the MK3S+ has been out for more than three months and is receiving excellent feedback, we keep postponing the MK4 variant further away. In this development diary, we’d like to talk about the technical problems we’ve encountered, how we’ve solved them, and what’s left to figure out before we can start shipping the production units.

Reliability of the MMU3 for the MK3S+

The primary reason for creating a successor to our MMU2S was to enhance its reliability and ease of use without the need for lengthy fine-tuning and calibrations. We’ve been shipping the MMU3 upgrade for the MK3S+ for over three months now (since July) and we’re thrilled to see that the reliability observed in both our internal and external beta testing is also manifesting in customer experiences. Achieving huge 70+ hour 5 color prints with zero interventions is exactly what we were aiming for.

Naturally, we want the MMU3 for the MK4 to be just as reliable and efficient, if not even more. However, we’re having a hard time matching the MK3S+ performance. Why?

Understanding MMU3’s reliability drop on the MK4

Both MK3 and MK4 use the identical MMU3 unit – there’s no difference between them. So one has to wonder, if the MMU3 works fine on the MK3S+, why is it not working just as well on the MK4? The differences between them are in the print heads. Each printer uses an entirely different extrusion and hotend system. They are also based on a different motherboard and firmware architecture of the printer itself.

When we first announced the MMU3 in March 2023, we had everything ready and working reliably with both the MK3S+ and the MK4. We were putting together units for external beta testers and we expected the production units to start shipping in late June. The only problem at the time was our manufacturing capacity. However, as soon as summer came and the ambient temperatures in all our test farms increased by 6 to 10 degrees Celsius, we started encountering a drop in reliability of the MMU3 on the MK4 (but not on the MK3S+). This was caused primarily by thin strings of filament that started to accumulate near the drive gears.

On the MK3S+, these strings would also appear, but to a much lesser degree. Plus, the extruder on the MK3S+ has a fairly open design and the filament path is completely straight, so we did not see any problems with them in our testing. On the MK4, however, these wisps of filament appeared much more frequently and caused a lot of issues with repeat loads and unloads.

An extreme case of the filament string forming on the tip

Troubleshooting the cause of these thin strings of filament led us down a rabbit hole of investigating what changed and researching ways to make the tips perfect again. We were changing nozzles, print and ambient temperatures, and speeds. The result of this complex research is a new logic for filament changes.

Completely new filament change strategy – elimination of strings

First, let me summarize how the filament change works on the MMU3 with the MK3S+.

  1. The print head moves above the wipe tower
  2. Quick ramming – rapidly extruding a tiny bit of filament to form a nice tip
  3. The filament is unloaded all the way back to the MMU unit
  4. The selector moves to the new position and loads the filament into the print head
  5. The print head starts extruding into the wipe tower/infill/wipe object and the color of the extruded filament gradually changes

In this process, the nozzle is full of the previously used filament until step 5. When we use this process on the MK4, the retraction in step 3 causes a tiny amount of the melted filament to be pulled out, which creates this problematic thin wisp of plastic. In step 5, the new filament needs to push all of the melted plastic out. So step 5 is also where the majority of the waste is generated.

To eliminate the tiny wisps or strings of filament on the MK4, we created a whole new strategy for changing the filament, specifically designed for the Nextruder geometry. It is loosely similar to how a cold pull is performed.

  1. The print head moves above the wipe tower
  2. Rapid ramming happens: the printer is moving at top speed and all of the melted filament is pushed from the nozzle into the wipe tower. The rapid extrusion also cools down the nozzle a bit.
  3. “Stamping” – by now, only partially heated filament is quickly pressed against the nozzle’s inside to improve the shape of the tip (a negative of the nozzle).
  4. The selector moves to the new position and loads the filament into the print head
  5. A tiny extrusion is made into the wipe tower to stabilize the flow, but the color is almost immediately clean
We would like to highlight two community projects that helped us develop our Stamping step and saved our developers and testers a significant amount of time – Skinny Dip post processing script by Erik Bjorgan and Dribbling by Antimix. While we ended up with a different approach, we would like to thank both authors for all the effort invested into the project and for making it open-source!

This way, there is little to no melted plastic in the nozzle after step 3, so we can achieve nice tips of filaments with the MK4’s Nextruder, even in highly elevated ambient temperatures. We are continually testing this in a special climate-controlled test chamber, which can run a small print farm. We spent months perfecting this new filament change routine, cataloging the effect every small difference has on the tip of the filament.

The tip of the filament with the new unload strategy

A welcome side-effect of this new strategy is a reduction in waste. Most of it now happens in the ramming stage (step 2), but since there is very little color mixing, the overall waste is reduced.

Getting this new strategy dialed in was a long and tedious process, as every change needs at least 2-3 days of nonstop printing on the entire MMU print farm to get a reasonable sample of reliability increase or decrease. It took us until late August to get to a point where we were happy with the tips that this procedure generated in various ambient temperatures. And our developers are still cautious about calling it the “final version” of the filament change procedure on the MK4. For example, we would like to test a much larger sample of filament colors and filament brands.

You may wonder why we are sticking to the hard way of producing the stringless tips. Two main reasons. First, this system allows us to unload most of the material from the hotend and re-use it for the next time this color is needed. And second, as the nozzle is basically clean before the next material loading, there is much less inter mixing compared to filament joining or cutting it and leaving the rest in the extruder. It reduces both waste and time dramatically.

Of course, the ideal solution is to have completely separate tool heads, such as the Original Prusa XL. With that kind of setup, the need to empty the nozzle of the previous color for each filament change is completely eliminated. The efficient wipe tower on the MMU3 is the next best thing, allowing our single-nozzle 3D printers to print with up to 5 filaments.

Current problems of MMU3 on the MK4

With the new loading procedure in place, the reliability of the MMU3 on the MK4 increased significantly. But as it happens with major changes like this, changing one thing affects other parts of the system and things that were working perfectly need fine-tuning again.

The MMU3 can do upwards of 2,000 filament changes in big prints and all of them have to happen perfectly. Just a couple of failed filament changes are enough to be really annoying, even if the error is recoverable most of the time (the LCD will show you what went wrong and how to fix it).

Stress test of the MMU3 on the MK4 – 5 color print – 1500 filament changes

For example, we discovered that the filament sensor now triggers at a slightly different time with the new filament tips. This caused holes in the wipe tower and sometimes even the print. A similar thing happened when we enabled Input Shaping, Precise Stepping and Pressure Advance (that’s where the numerical error turned out to be) and started printing the Wipe tower at high speeds. We have now adjusted the algorithm and fixed both of these issues.

These are the three remaining things to solve – they are the most common causes of print problems. About 15% of the errors are caused by incorrect sensing of the filament entering the print head. Just like on the MK3S+, we have developed an alternative version of the printed parts around the filament sensor to solve this on the MK4. Rather than detecting the filament simply entering the area above the drive gears, the new lever is actuated by the filament pushing the idler doors slightly open. So the hall sensor is only triggered when the filament actually enters the drive gears. We are still iterating this part until we eliminate the problem, or at least make it extremely rare in occurrence.

With the new printed part, the lever is actuated by the filament entering the drive gear and pushing the idler doors open

Another slightly less common problem is the filament sensor on the MMU3 unit itself not triggering correctly. Since this sensor is working perfectly with the MK3S+, the problem is most likely caused by tiny pieces of debris still being sometimes generated by filament loads – this may cause the FINDA ball to get stuck on unload.

The most common issue right now (roughly 50% of all fails) is caused by the filament not going smoothly through the extruder and into the metal filament guide (the “tube” connected to the nozzle). This can manifest in several ways, such as the filament getting stuck on the edge of the tube. We are working on both mechanical and firmware solutions, e.g., an automatic filament reload that is initiated when the printer detects this situation occurred.

If you’ve changed your spool setup after slicing the file, the new pre-print screen lets you remap colors, set up spool-join, or reload different materials

When will we ship it?

Repeated delays are frustrating for you, our customers, and our team. With that said, we have decided to not start shipping the MMU3 for the MK4 until we’re sure the reliability is hitting the goals we set. It is extremely hard to estimate when this will happen.

It’s quite possible our testers will confirm that the latest changes solved the listed issues – and we will be ready to start shipping within a couple of weeks. But there is also the possibility that the solution won’t be sufficient and we will need even more time to get everything right. We honestly don’t know, and at this point, we rather say it out loud this way than give you a random estimate.

We would like to apologize for not shipping the MMU3 for the MK4 on time! I’m sure you’re curious about what’s going to happen with existing orders – please check the box below for detailed info.

Important information for current orders:
As a gesture of appreciation for your continued patience, we’ll send you an email with a $50 voucher next week. This applies to all MMU3 for the MK4 orders. The voucher can be applied to your future orders. Of course, you have the option to cancel your order at any time, without any fees or other charges. You can re-order it later, the only downside is losing your spot in the queue. Orders of the MMU2S to MMU3 upgrade for MK4 will get a $15 voucher (the same proportional value in regards to the respective product’s price).
If your order includes MMU3 for the MK4 together with other items, we will send you an email next week, giving you the option to split the order. If you choose to do so, we’ll send you all the other items in your order first, with no shipping charge. Once the MMU3 unit is ready, we’ll ship it to you separately.

You will also receive the $50 voucher. The vouchers will be sent to all customers who made the order before Friday, October 27, 2023, 16:00 CEST (10 AM EST).

We will keep you updated both with the good news and shipping being ready, but also in the case of ongoing troubleshooting and further development.