Cook Political Report has introduced their "Swingometer" page, allow you to tinker with turnout levels for various demographic groups. Based on 2016 numbers, and without the 6% third-party vote, Joe Biden wins comfortably 307-231.
But it doesn't take much for Biden to lose if everything else remains the same from 2016.
For example, if all that changes from 2016 is white non-college voter turnout goes from 55% to 60%, Biden wins the popular vote by more than 3 million, but loses in a 306-232 electoral college rout that mirrors the 2016 map. Trump would win PA, MI, WI, and FL all by less than 1% again.
Even worse would be just a 4-point shift in the Black vote, with Trump going from 8% to 12%. Biden would win the popular vote by more than four million, but lose 276-262 as Biden would only pick up PA and WI, but Trump would keep FL, MI, and AZ. The same thing happens if Biden still gets 92% of the Black vote, but Black turnout is suppressed from 57% to 52%. A combination of a 10% Trump Black vote and a 2% drop in Black turnout also leads to electoral loss for Biden. It doesn't take much for Trump to win, folks.
Having said all that however, if Biden's share of white college voters goes up just three points from 54% in 2016 to 57%, he wins Georgia, NC, and Florida and he rolls Trump 350-188. Same happens if he does 3 points better with non-college white voters, going to 34% from 31%. If he does both, same electoral margin, but he wins the popular vote by almost 12 million. If Biden does five points better with white voters, he takes Texas and Ohio too.
The polls though have Biden regularly doing much better with white voters overall. This bodes really, really well for the blowout scenarios where Biden gets 350 or more electoral votes, where Trump simply can't dispute the loss anymore. Biden is doing much better than Clinton did in 2016, full stop.
And if Trump is doing well enough to tie overall with white voters?
Trump gets evaporated like a snowball on the sun.