NOTE: I am leaving this original post intact, but look on page 2 of the thread for an update based on something somebody pointed out. I did 10 trials for each tire, but only pasted 9 of them into some of the original analysis. I’ve updated figures for the missing set of trials.
As promised, I am posting the results of some field testing I did comparing two different front wheel and tire setups. The results were both as expected and not as expected.
The two wheel setups were:
Zipp 404 (2001 model) shod with tubular Tufo S3, size 21. Glued with Continental glue about one year ago. The gluing job at that time involved a full cleaning of the rim, multiple coats, etc. That wheel+tire has been trained and raced on 4-5 times in the past year for a total of about 150 miles. It has been stored inflated. It does not have sealant in it. To the casual observer (me) the tire looks virtually new.
Zipp 808 (2005 model) sporting a clincher Michelin Pro Race, size 23, with standard weight inner tube (Specialized). That tire has about 1,500 miles on it.
Bike setup was my Yaqui DE, size 55 TT bike. Rear wheel was PowerTap hub on OpenPro rim and Michelin Pro Race size 23 clincher. No water bottles; two frame cages. Both tires at 115 psi. No adjustments made to any bike fit parameters during testing window.
I wore the exact same clothing for each trial – snug sleeveless tri-top, basic bike shorts, Bell X-Ray helmet, TriVent shoes. No arm warmers or leg warmers.
Trials took place in three batches in a six-day window. Course was Fiesta Island “short loop,” which is 2.5 miles, give or take 50 meters depending on exactly where I hit the interval button on the PT computer. I measured single laps, stringing together 1-3 in a row on each wheel before changing wheels. Each lap began with a flying start at about 20 mph, holding a given power output target for that lap. Each lap was completed at a constant power. For example, I would do a whole lap at ~175 watts, then start over for the next one at, say, 215 watts. A total of ten laps were completed for each wheel+tire.
Conditions ranged from a barely perceptible breeze to dead calm. I did each set at 5:30 to 6:30 am, and quit before the sun got up in the sky and kicked up the wind. Road surface there is fairly good – about what you’d expect on most N. American Ironman courses. Slight rise and fall of ~20 feet per lap.
I recorded the average watts and average speed off of the PT computer display. I was not able to complete downloads into CyclingPeaks on all of the runs, so I used the PT computer data. More on that point later.
The data:
404+Tufo
Power Speed (mph)
162 19.5
205 21.6
173 20.6
165 19.6
183 20.3
210 21.9
164 19.7
214 22
235 23
808+Michelin
Power Speed (mph)
173 21
211 22.7
222 23
215 22.9
190 21.8
188 21.6
160 20.5
242 23.9
245 23.9
A casual examination of these figures should convince anyone that the 808 setup is much faster (at the 99+% confidence level). It’s not even close.
The big question is, “Why?” Is it the tires or the wheels? Let’s first talk about the wheels. Zipp has published figures that indicate that the 2001 404 is about a half second per kilometer slower than the 2005 808. In terms of my data accuracy (tenths of a mile per hour) that is barely measurable. While I had intended to make an exacting adjustment to the data for the drag differences between the wheels, I think it is a waste of time to do so. The differences overwhelm an adjustment that small.
So, I start by assuming that the entire system of bike and rider had the same CdA for all trials. (If Josh at Zipp thinks this is a bad assumption, please feel free to say so…).
With full credit to technical assistance from my mate from the Great White North, rmur, assuming a CdA of 0.27 and some other constants, we would estimate the Crr’s to be:
404+Tufo: 0.0073
808+Michelin: 0.0050
This is equivalent to a 15-20 watt differential to travel the same speed. Note that these Crr estimates are for *both tires being the same. *In the case of the 404+Tufo setup, all of the difference is attributed to only one tire. In other words, these figures understate the Crr for the Tufo tire. Under our assumptions here about CdA, the Tufo is actually somewhat worse than 0.0073, and the wattage penalty for the Tufos is somewhat worse than 15-20 watts. Throw in a small speed adjustment for the 404, and maybe we get some of those watts back. But just a little. The difference here is 10-12 minutes in an Ironman, my friends. And that’s just one of two tires.
I could leave it at that and claim to have proved a point, but there is a puzzling anomaly that bears reporting.
In the method above, we made a (perfectly valid, IMO) assumption that the CdA was near-constant for both setups. It turns out that we can test that assumption. Via a method published by Allen Lim, one can analyze these types of trials at varying speeds and power levels and extract CdA and Crr separately. Kraig Willet was kind enough to crunch the numbers for me and show me an odd result.
According to the Lim method, the 404+Tufo trials exhibited a materially lower CdA than did the 808 setup trials. While we assumed above that they were the same, or that, at most, the 404 was a little slower, the data tells us the “slower” wheel was faster in terms of CdA. How much faster? A CdA of 0.24 versus 0.28. If you didn’t realize it, this is a shocking gap.
This has two serious implications. First, if it is true (and I don’t think it is), Zipp better go back to the wind tunnel! If it is true, it means that the Tufo tires performed far, far worse than rmur estimated. The Crr had to have been so bad as to overcome a profound aerodynamic advantage held by the 404 wheel. While I am convinced the Tufos performed badly versus the Michelins, they cannot be that much worse! Knobby mountain bike tires maybe, but not the Tufos.
The second implication is that the data has some flaw in either the way it was gathered or the way it was computed and recorded. The only “gathering” flaw would be if I had held the aero position while on the 404s and sat up while on the 808s. I assure you that was not the case, so we must look elsewhere.
Two possibilities: First, I discovered that the PT head unit reports average power and speed figures that are slightly different from the averages one gets if one downloads the data and analyzes the raw file. Average power and average speed are sometimes under-reported on the head unit, versus the data actually recorded on-the-fly. The differences are in the 0.4% to 1.1% range, and I cannot discern a pattern. So, it is possible that, by reading the results off the PT head unit instead of from a download, I have flawed data. It is unlikely to affect the basic conclusion from the first half of my discussion because the possibility of mis-calculated average figures would seem to apply equally to both wheel setups. While these changes might be enough to throw off the regression results from the Lim technique, they are quite unlikely to throw off the basic conclusions.
Second, Lim’s method argues for lots of data points. In particular, it begs for additional data points from low-power, low-speed trials. My slowest trial was 19.5 mph at 162 watts. Lim did trials starting at 100 watts in his paper. Kraig has suggested I go do a few more low-speed trials. I think I’ll do that and see if the Lim formula spits out results that are a bit closer to what we might expect. It is an open issue that needs resolution. It is, quite frankly, impossible that the 404 setup reduces the CdA of the entire bike + rider system by a whopping 0.04.
So, what do I conclude?
This and only this: A 2005-era Zipp 808 clincher front wheel mounted with a Michelin Pro Race 23 mm tire requires 15-20 less watts to push along at a given race-level speed than a 2001-era Zipp 404 tubular front wheel mounted with a Tufo S3 21 mm tire, on a bike set up as described, and piloted by me.
Of that much, I am quite certain. The data is overwhelming, and there is no room for the null hypothesis.
What I am less certain about is how much of that difference is strictly due to tires. I have an overwhelmingly strong hunch that it is all in the tires. But until I run some more trials, the Lim calculation leaves me scratching my head.
Questions?