How accurate is Zwift’s power estimate for classic trainers?

Zwift world power vs real data
Power in Watts and Speed in Km/hr

I began using Zwift late last year for offseason riding. In a way, it is my winter group ride replacement. I use a classic magneto trainer with a Garmin speed, heart rate and cadence sensor, which do the trick for me. Because I know the speed I’m pedalling, and since the resistance of the trainer is consistent, I can tell very easily if I’m improving.

Zwift race starting line waiting for users

 

Less than one week in and I find out that the Zwift Academy has started. It looked like fun, so I joined. I believe one had to complete 2 races along with some group rides and their training workouts to graduate.

The races were fun. I enjoy the challenge of competing with other riders any day of the week. but I had no way of knowing if my power was accurate. So I did the best I could to compare Zwift’s power to CycleOps power graph, which I wrote about.

Still unsatisfied, I got my hands on some Favero Assioma dual pedal based power meters, and did some real tests. Here are the results.

Favero Assioma power meter duo box

 

All tests were done on the CycleOps Magneto trainer. Speed was recorded using the Garmin 520 computer. Power was recorded from the Assioma Duo pedals and Zwift’s estimate.

Kms/hr Zwift Estimated Power Assioma Power % Difference
23 160w 138w +14.75%
34 320w 272w +16.22%
40 412w 342w +18.57%
48 490w 414w +16.81%
57 550w 590w -7.02%
64 550w 690w -22.58%
Zwift world power vs real data
Power in Watts and Speed in Km/hr

Conclusion:

It’s pretty clear that Zwift’s estimates are off. In the lower power range zwift overestimates power by more than 30%. However, once it passes about 540 watts the pendulum swings the other way and then it begins to underestimate power by well over 22%.

I really enjoy Zwift and would encourage anyone to hop on the platform, with or without a power meter. Hopefully Zwift will be able to use this data along with data posted by other users to help them improve the accuracy of their software for all riders. Those who can afford power meters and the many who are just getting into the sport.

15 Comments on "How accurate is Zwift’s power estimate for classic trainers?"


  1. Thanks for your review!
    My Zwift shows 450 watts average and 4,9 to 5,6 kg/w, and I can mantain that for 30 or 40 minutes. My bpm is 175 average and when I sprint I go to 200bpm.
    I keep being “reported” for doing too much power and can’t finish a race.
    Is this normal? Or is it a problem with my setup?

    Reply

    1. Hi Andre, do you have a friend with a power meter that you can use to do a 30 minute test with? I did that before I had a power trainer to calibrate my numbers and see where I was at. Zwift has certain metrics they use to flag unexpected numbers. I’m no expert on their algorithm. Our HRs are very similar.

      Reply

  2. Thanks for the information. I’ve been riding with a speed/cadence set-up on a classic trainer and the power of Zwift was more or less matching the estimated power on the display of my trainer.

    Yesterday, for the first time, I rode with the Favero Duo power meter and my numbers were much lower than I had before, similar as your results. Did a few calibration tests thinking it was the power meter, but so it seems it’s actually Zwift overestimating.

    One thing though that keeps me wondering is the power reading of my Tacx display. I calibrate my trainer every ride for a good power reading but this seems now to be off by 15-20% with the reading of the Favero…

    I see more articles on Zwift about overestimating power with classic trainers, but it would also like my old Tacx was also not providing an accurate readout?

    Reply

    1. Hi Steven, I’ve tested the Favero Duo with the 2018 Wahoo Kickr (which I should publish on this blog as it was a really interesting story) and the Cyclops Hammer. Both of them showed the same reading +- 1%.

      I was riding on a new drivetrain. New chain and new sram red components. Wahoo shared an interesting graph with me showing the discrepancy created by worn drive trains. So that would be one thing to check as well. As your Tacx is reading power at the end of the drive train, and the Favero is reading power at the pedal (beginning of the drive train). And power can be lost due to friction of a worn drivetrain.

      Reply

  3. This is a lot of work going to a pointless conclusion.
    Get a modern, power accurate trainer is the answer – you can expect perfection from a feature that’s designed to begrudgingly allow luddites onto the platform.

    Reply

    1. Hi Eric, thanks for your comment. Having zwift estimate power is a great thing. Cycling can be an expensive sport to get into, but platforms like Zwift are lowering the barrier. Getting a bike, then a trainer with an accurate power meter comes with a big price tag. So the more accurate their estimates the better, especially for newbies getting into the sport. The more cyclists the better.

      Reply

  4. I have just entered the world of Zwift with a dumb trainer and Garmin Vector 3 pedals , I have been riding with a group that I trained for years and didn’t understand why I was having so much trouble following, so I pushed harder and things started to come together. One of 2 things will happen next summer on the road.
    1 they will beable to ride with me or 2 I will kick some serious ass….
    My feeling is option 2…
    Any idea if that will hold true.
    Thanks

    Reply

  5. Thank you for the information, very helpful as I have the same trainer and have remained skeptical about power readings. One important thing worth noting is that Zwift limits turbo trainer power to 550w, so any data above that is not inaccurate, just not represented to avoid further discrepancies when it comes to racing/competing.

    Reply

  6. Thanks for this data, Jonathan. It is clearly very helpful. I would completely agree with you that the Zwift estimation is not accurate but on a day to day basis it is consistent and will help you gauge progress. I train on a very old Computrainer. Unfortunately, I can’t sync it with Zwift due to mechanical problems with my controller. That said I can control resistance and the first thing I did was to estimate the appropriate resistance setting I needed to duplicate road resistance. I did this by testing it on a 0 grade part of the course and maintaining 90 rpm using the Wahoo speed and cadence sensors. I then set the resistance to the level that I know I would allow me to maintain the speed that I average when on the road. I used this resistance to do the FTP test on Zwift. My theory is that this will help overcome the inability of Zwift to assess drag without an accurate power meter and improve the accuracy of the FTP test. Does this make sense to you?
    Thanks for your help,
    Elliot

    Reply

  7. Thanks for the article. Your conclusion confirmed what I suspected. The overestimate might just be what makes people like Zwift. It allows them to see themselves as strong riders thus stroking the ego. One thing I noticed is that before I knew I could select my trainer, (CycleOp fluid 2) my power was really low. After selecting the proper trainer power seemingly shot up. And until I get a power meter I will not really know but it really does not matter. As long as there is some semblance of consistency I can gain good training.

    Reply

    1. Hi Doug,

      Completely agree with you, having consistency is what matters for benchmarking training and improvement. And Zwift power offers that.

      I think some trainers might be overestimated, some under. But I hope that Zwift finds a way of doing some sort of spin down test that calibrates the estimated zwift power and speed with a riders gearing. One thing I like about zwift power is that it opens up the gaming experience to any rider. Cycling can be an expensive sport, but the easier it is to jump on any bike and any trainer to join zwift, I think it will attract a lot more riders to cycling.

      Reply

Leave a Reply to Doug Cancel reply

Your email address will not be published. Required fields are marked *