Tesla's 85 kWh rating needs an asterisk (up to 81 kWh, with up to ~77 kWh usable)

[MarcG]

I own an early LR 3... and I've tried it. I can't even get 70 to show up, let alone 74, in the best conditions. I get the range (in the spring) without a problem, surprisingly, and the car still charges to ~312 or so miles at 100% after ~21k miles.

Bjorn did a couple of tests and got 70 kWh both times (shown in the "Since Last Charge" screen), though he started at 99% and ended at 3% SoC in one and 5% in the other.
I think he could've hit 73kWh if he went all the way to 0% -- come on Bjorn, try again!
This is all making me want to give it a shot myself, hmm...

 

[wk057]

Bjorn did a couple of tests and got 70 kWh both times (shown in the "Since Last Charge" screen), though he started at 99% and ended at 3% SoC in one and 5% in the other.
I think he could've hit 73kWh if he went all the way to 0% -- come on Bjorn, try again!
This is all making me want to give it a shot myself, hmm...

This is pretty good stuff. Keep in mind that displaying in percentage rounds down to the nearest percent. So, 99% could be 99.999999% or 99.0%... about a 0.7 kWh window. The best way to get accurate measurements would be to put the display in km's for the most granularity. Don't have to leave it that way, just for the start and finish.

Since we have the added granularity, we can see he's at 70.084 kWh in the first video, 70.038 kWh in the second video.

For the first video, starting at 99% and ending at 3% would mean large window of 99.999% to 3.0%, or 72.25 kWh. Smallest window, best case, 99.0% to 3.999%, or 73.77 kWh. I think we can agree his 99% was on the higher end, though, given that the charge stopped there. My guess would be closer to the high end of that 99% window with a very minor pack imbalance causing an early cut off, but no way to really know for sure since he's got it in percent.

For the second video, he was down to 1 kW input power on the charge, with 10 minutes left. So, max of about 0.167 kWh left to put into the pack... probably less, since some of that would be accessories. I'd say he was definitely about to hit 100% within the next couple of minutes, but we'll just round up and say 0.2 kWh away from 100 and add that in to the total to be more than fair. The end of 5% could be between 5.0 and 5.9999%. Would be 74.33 kWh best case, 73.54 kWh worst case, based on those numbers. Pretty close! I'd say the second video probably suffered from the error in the power measurement, though, due to the elevation changes. This system isn't designed to be 99.99% accurate, it just needs to be accurate enough so that you don't get stranded.

The kludge "buffer" code illustrates the margin of error in measurement quite well, with a best case being about a 2.7% margin of error accounted for in measurements based on that. Again, measurements at lower powers have higher error... so changes from uphill to downhill will cross through this high error point often, as will stop and go traffic and such.

Anyway, close, but not quite there.

If someone plans on doing this, I'd say they should charge to 100%, snapshot this with the car set to km's, drive to 0% (or as close as possible, or until you see that 74 show up) on flat roads, avoiding stops, slowdowns, regen, etc.

I'll point out that 74 would still only put total pack capacity at 76 kWh, so 76 is actually the number that would need to be hit to prove the 78 kWh pack claims. I said 74 because that's still basically unattainable without driving through the full buffer, which the car shouldn't allow you to do, but may in some cases.

Good luck.

 

[MarcG]

Thanks! So to summarize, the total pack capacity of long-range Model 3s is “about 75kWh”, just as Elon stated yesterday

 

[supratachophobia]

Yes, this makes a lot of sense - as mentioned, anything that does not come from @wk057 research and does not conform to his claims is fwaud.

To be fair, there aren't many others who have his experience with the inner workings of the car and this leaked fleet data, soooo..... whether you think he's right or wrong, he's still an "expert" in this field.

 

[supratachophobia]

It most certainly does roll backwards. I start my morning drive every day with a ~3 mile steep descent - my kWh used since last charge not only rolls back but goes negative.

That said, I think it's perfectly likely that number, over a long trip with lots of elevation change and acceleration/deceleration/regeneration is far from the most accurate measure of how much energy was actually consumed and being off by a few kWh is a very reasonable expectation.

I find no reason to accuse the individual of "cooking the books" or otherwise deliberately attempting to misrepresent the test. That's just silly. For the same reason, I also believe that Tesla has zero incentive to misrepresent the actual capacity of their packs in their own internal documentation and words from the CEO spoken to shareholders.

Wait, what? Tesla has every incentive to not give the full story on their battery packs. They have something called a stock price. If the public knew that 90kwh owners were getting screwed in totally capacity, and EPA mileage numbers on the dash were unachievable, they'd have dozens of lawsuits overnight from irate owners.

 

[ord3r]

Jack Ricard's testing of the 2170 cells line up with wk407's findings.

 

[brianman]

@wk057 - Would a climate control consumption based test be useful here?

1. park in enclosed garage
2. set car to charge to 100%
3. check periodically until the car says Complete
4. wait 10 minutes
5. unplug, replug, wait in the car until the car says Complete
6. unplug
7. open garage door
8. open car windows
9. set climate to auto, HI and then adjust fan speed to 11
10. turn climate control when parked to On (or Dog Mode)
11. check on car periodically waiting until car reaches the 20% SOC "keep climate control on" cutoff
12. reach through the driver window frame and tap the climate back on
13. repeat step 12 every 15 minutes (?) while cleaning the garage until rate range remaining is reported as 1km
14. make note of the Since Last Charge numbers
15. set car to charge to your normal preferred SOC value
16. plug in

 

[ran349]

@wk057 - Would a climate control consumption based test be useful here?

1. park in enclosed garage
2. set car to charge to 100%
3. check periodically until the car says Complete
4. wait 10 minutes
5. unplug, replug, wait in the car until the car says Complete
6. unplug
7. open garage door
8. open car windows
9. set climate to auto, HI and then adjust fan speed to 11
10. turn climate control when parked to On (or Dog Mode)
11. check on car periodically waiting until car reaches the 20% SOC "keep climate control on" cutoff
12. reach through the driver window frame and tap the climate back on
13. repeat step 12 every 15 minutes (?) while cleaning the garage until rate range remaining is reported as 1km
14. make note of the Since Last Charge numbers
15. set car to charge to your normal preferred SOC value
16. plug in

The only thing I agree with there is cleaning the garage.

 

[supratachophobia]

@wk057 - Would a climate control consumption based test be useful here?

1. park in enclosed garage
2. set car to charge to 100%
3. check periodically until the car says Complete
4. wait 10 minutes
5. unplug, replug, wait in the car until the car says Complete
6. unplug
7. open garage door
8. open car windows
9. set climate to auto, HI and then adjust fan speed to 11
10. turn climate control when parked to On (or Dog Mode)
11. check on car periodically waiting until car reaches the 20% SOC "keep climate control on" cutoff
12. reach through the driver window frame and tap the climate back on
13. repeat step 12 every 15 minutes (?) while cleaning the garage until rate range remaining is reported as 1km
14. make note of the Since Last Charge numbers
15. set car to charge to your normal preferred SOC value
16. plug in

That's crazy. Just take a 254 mile trip on level ground, with a steady mph, and track consumption. That's what I did.

 

[MarcG]

Jack Ricard's testing of the 2170 cells line up with wk407's findings.

Who is wk407?

 

[wdolson]

Wait, what? Tesla has every incentive to not give the full story on their battery packs. They have something called a stock price. If the public knew that 90kwh owners were getting screwed in totally capacity, and EPA mileage numbers on the dash were unachievable, they'd have dozens of lawsuits overnight from irate owners.

The EPA mileage is achievable if you drive carefully. I often get around 285 Wh/Mi running errands around the local area, which is the Tesla rating. I know there are a number of fudge factors cooked into the firmware, but the car reports 298 miles at 100% and when I get around 285 Wh/Mi, the miles remaining is often a little low (I may drive 60 miles and the miles remaining meter may drop 62), but within a few miles. But 298 miles at 285 Wh/Mi is 84.9 KWh, so the smaller pack size is cooked into the numbers. The advertised range for the refresh Model S with 19" wheels was 294 miles, I just got lucky and got a few extra miles.

But I do agree that if they were advertising 85 or 90 KWh batteries they should have delivered something closer to that.

 

[supratachophobia]

The EPA mileage is achievable if you drive carefully. I often get around 285 Wh/Mi running errands around the local area, which is the Tesla rating. I know there are a number of fudge factors cooked into the firmware, but the car reports 298 miles at 100% and when I get around 285 Wh/Mi, the miles remaining is often a little low (I may drive 60 miles and the miles remaining meter may drop 62), but within a few miles. But 298 miles at 285 Wh/Mi is 84.9 KWh, so the smaller pack size is cooked into the numbers. The advertised range for the refresh Model S with 19" wheels was 294 miles, I just got lucky and got a few extra miles.

But I do agree that if they were advertising 85 or 90 KWh batteries they should have delivered something closer to that.

False. The EPA mileage is achievable only if you drive *under* the EPA rated watts per mile.

 

[ran349]

False. The EPA mileage is achievable only if you drive *under* the EPA rated watts per mile.

He is basically saying that too. Using his numbers, if he drives 60 miles he uses 62 rated miles. So 60/62 x 285 Wh/mi = 276 Wh/mi to achieve rated EPA mileage. This is very typical. Even with getting 298 rated miles on a charge, his actual useful battery capacity is .276 x 298 = 82.25 kWh. Assuming he could use all of the 4 kWh energy buffer, his capacity would be around 86 kWh. But of course, we know you can't count on the buffer being available for driving.

 

[darth_vad3r]

Keep in mind that displaying in percentage rounds down to the nearest percent. So, 99% could be 99.999999% or 99.0%.

@wk057 were you just generalizing the fact that there is up to 1% error, or how did you determine that the % display actually rounds down (floor) instead of ‘normal’ rounding (to nearest integer) like the trip meter seems to do for kWh? It makes no difference in total errors when subtracting two percentages, but I’ve been assuming “50%” meant 49.5% to 50.49999%.

I also assumed the reason charging doesn’t stop, e.g., when it ticked up to “50%” was because that meant it was only just at 49.5% and the actual target was 50.0% and not achieved yet (for example).

Thanks

 

[wdolson]

@wk057 were you just generalizing the fact that there is up to 1% error, or how did you determine that the % display actually rounds down (floor) instead of ‘normal’ rounding (to nearest integer) like the trip meter seems to do for kWh? It makes no difference in total errors when subtracting two percentages, but I’ve been assuming “50%” meant 49.5% to 50.49999%.

I also assumed the reason charging doesn’t stop, e.g., when it ticked up to “50%” was because that meant it was only just at 49.5% and the actual target was 50.0% and not achieved yet (for example).

Thanks

In C and C++, the two programming languages most common in Linux and embedded systems, the way integer division works, the entire decimal part is discarded rather than round off like you'd expect. So 29/5, which would be 5.8 and would be rounded up to 6 would be 5 without some extra work in C or C++. There are ways to write the code so it rounds up as you'd expect, but the language doesn't do it on its own.

I'm just guessing, but I would assume that's going on with at least some of the calculations.

 

[AWDtsla]

In C and C++, the two programming languages most common in Linux and embedded systems, the way integer division works, the entire decimal part is discarded rather than round off like you'd expect. So 29/5, which would be 5.8 and would be rounded up to 6 would be 5 without some extra work in C or C++. There are ways to write the code so it rounds up as you'd expect, but the language doesn't do it on its own.

I'm just guessing, but I would assume that's going on with at least some of the calculations.

Some Tesla calculations round from the lower end to the next integer. i.e. 4.2 becomes 5. Waiting for an explanation on that one.

 

[darth_vad3r]

In C and C++, the two programming languages most common in Linux and embedded systems, the way integer division works, the entire decimal part is discarded rather than round off like you'd expect. So 29/5, which would be 5.8 and would be rounded up to 6 would be 5 without some extra work in C or C++. There are ways to write the code so it rounds up as you'd expect, but the language doesn't do it on its own.

I'm just guessing, but I would assume that's going on with at least some of the calculations.

Yes, that's how integer division works. If you want to round to nearest, you add 0.5 first, it's only 1 extra op. They are certainly intentionally doing this already for kWh display.

e.g. my last trip reads: "38.4 km 6 kWh 144 Wh/km"
38.4 km * 144 Wh/km = 5.53 kWh ... shown above as "6 kWh" shows that they are choosing to round to nearest integer here.

I have no clear evidence either way for % display or km display for range, but I had just assumed the same rounding and hadn't thought otherwise until I read @wk057 's post I quoted above.

(BTW, I have lots of evidence of the intentional rounding-to-nearest for kWh on the trip ... I have almost 100 trips recorded, the average delta from the displayed kWh and calculated kWh (from the displayed km and Wh/km) is 0.22 kWh, which is pretty near what you'd expect for a random-ish sample size of data that is being rounded anywhere from 0 to 0.5 (avg 0.25), and not at all what you would expect if you were 'rounding down' (avg round of 0.50).

I could certainly see the logic in rounding down for range ... you'd prefer not to over-advertise available range, e.g. if rounding to nearest 50km, saying "50km" when you are at 80km actual range is better than saying "100km" since you'd be relying on that range to tell you if you can make your destination ... I just never thought of applying that logic to such small units as 1% or 1 km.

Anyways, @wk057 I'm curious how you determined this was the case if you did.

Thanks.

 

[darth_vad3r]

Some Tesla calculations round from the lower end to the next integer. i.e. 4.2 becomes 5. Waiting for an explanation on that one.

Really? Hmm. Can you give an example on that?

One thing I can think of is that the cumulative error from rounding on the distance and the displayed (calculated) consumption could lead you to think they've rounded the wrong way. This could be easier to exaggerate on a shorter trip. Most of my trips are not short, but even so I have one outlier in my data for a long trip that appears to show kWh rounded down:

e.g.

trip: "274.4 km 42 kWh 155 Wh/km"
274.4 km * 155 Wh/km = 42.53 kWh huh? rounded down to 42? ... well, in reality here the 155 Wh/km number has likely been rounded up. If I punch in 154.88 or lower I get 42.499 kWh and my faith in rounding is restored ​

 

[AWDtsla]

Really? Hmm. Can you give an example on that?

Check displayed charging speed in the mobile app vs what's reported by the API while charging on 120. I actually haven't personally checked in many software revisions, but it was "broken" for about 2 years before which it was correct. It wasn't an averaging problem, you could charge for 10 hours straight and nothing changed, somehow with the API reporting < 4.5 you could get a 5 displayed.

 

[wdolson]

Some Tesla calculations round from the lower end to the next integer. i.e. 4.2 becomes 5. Waiting for an explanation on that one.

The programmer can add code to force a round up for any fraction, but it takes a bit of extra code.