Is this position drawn?

I think black should win. I tried it and was able to blockade the d pawn with the king and keep one pawn with the knight.

The tablebase disagrees. F5 is indeed pointless but if you play it out on the analysis board, it is hard to find better. All moves weaken white's position.
Thanks, everyone, for the input. In the actual game, my opponent never played Kd5 for fear of pulling his king away from the h-pawn, and the actual continuation was the sequence above.
It seems like h5 before f5 or g5 is essential (as I thought it was at the time), but as you can see here because black has the opposition, I couldn't push the h-pawn through, and I couldn't kick the knight coming to g6 with my king. Black had just enough resources to hold.
If black has a win (and he seems to), Kd5 appears to be essential to it. If the white king gets to e4 as I did, both black pawns have to come off, and white is drawn at worst), but in an actual game (and black was under some time pressure), it's hard to fight the instinct to bring the knight back into the action immediately.

Looking at the ideas from DSmith in the top diagram and JamieDelarosa in the bottom. Black still wins except after the blunder in the actual game of 5...fxg5
After 5. ..Ne7 6. gxf6 gxf6 7. f5 Ng8 (only move to stop the h-pawn) 8. Ke3 looks better than h6, though I think black still wins. Thanks!
Indeed, knight = 3.5 points; 2 pawns = 2 points.
Much better for black.
A common misconception is pawns get stronger in the endgame.
Usually, this is not so, a piece is always a piece.
Quality pawns - advanced, connected - do matter.
Many GMs have it too.
My eyes opened only after I have been watching a large number of top engine games.
3 pawns vs minor is mostly draw or even the pawn side has slight edge, but minor plus 3 pawns vs 2 minors already favours the 2 minors side, generally.
The trend is amplified the bigger the number of pieces.
So, even in the very late endgame pawns are generally weak.
Advanced pawns are strong, but that is a different thing.