My GY-521 actually uses 2.2K pullup resistors which seems low.
That's OK for 3.3V.
So should the Arduino's internal pullup resistors be disabled after Wire.begin() ?
No, the internal pull-ups are so weak, that shouldn't be necessary.
If the internal pullups are enabled will this help to achieve correct logic levels for a 5V Arduino ?
No really, but a 3.3V level should be OK. Just keep the wiring between the Arduino and the GY-521 as short as possible. Additional capacitance could ruin the voltage rise and you must decrease the I2C speed to get a valid result.
Is leaving them enabled any danger to the MPU6050 or the 3.3V regulator on the GY-521 ?
No. The voltage will increase only slightly, definitely inside the MPU6050 specs.
You are playing the game of sometimes maybe (sometimes it will work sometimes it won't) You need to do a voltage translation, all units on the I2C bus need to be at there logic voltage levels. Rough rule of thumb with CMOS the logic low is VCC * 0.4 and a logic 1 is VCC * 0.6, this gives you about a 0.02 volt window, guess what happens if you get noise or a power supply droop. This response is to help you get started in solving your problem, not solve it for you.
Good Luck & Have Fun!
Gil
@pylon thanks very much for addressing all my questions. You have cleared up my doubts.
@gilshultz How do you get 0.02V window ? 0.6*5V = 3V. The pullup is to 3.3V, so I might expect under 0.3V headroom, but not only 0.02V. I checked the specs for the ATmega328P used in the nano. The graph on page 372 shows logic 1 at Vcc=5V to be around 2.7V. That being said I still do not rule out I2C connection problems as the root of my misery !
I wish the designers of the GY-521 had thought to leave an adjustable 'link' for connecting the on-board pullup resistors to 3.3V or 5V (since both are available on the board). Open collector/open drain outputs are ideal for connecting devices with different logic levels.
