Steam Deck OLED Undervolt Efficiency and Performance Scaling

thubn@discuss.tchncs.de to Steam Deck@sopuli.xyz – 87 points –

Hello, today I tested the performance and efficiency gains of my Steam Deck OLED at different TDP settings. As a benchmark i used 3DMark Firestrike. Which might not be the best benchmark for the Steam Deck, because it renders at 1080p. But you can still get good information from the results.

Link to Spreadsheet: https://docs.google.com/spreadsheets/d/1aOf00eA7WACqq1FZA1lJ067E8RFica5YpKWVEwCyB3M/edit?usp=sharing

11

Interesting stuff.

I'd suggest reducing the decimals (significant figures) to a more readable amount (like 1 or 2). Additionally, inconsistent number of decimals makes it harder to compare down a column. Ex: 2.23 instead of 2.23758366384763.

Agreed. The data is interesting, but that level of accuracy likely doesn't mean anything, and it's really hard to read. Plus, humans are really terrible at conceptualizing large numbers (even if it's ultimately a representation of a fraction).

Accuracy may be important in calculations, but Excel will handle accuracy apart from forcing it to show only two decimal places.

Thank you for the feedback. I decreased the amount of decimals to 1 and changed the percent values to the percent format.

And thank you for accepting critique! A rare quality indeed.

Very interesting! Could you provide some summary of what the trends are or may be visualize the results? I am new to this kind of data

  • The stable undervolt for my deck (20/50/50) has roughly the same performance at 9W as the default 0/0/0 at 10W.
  • an easy 10/10/10 undervolt can give you 2% more performance
  • the 20/50/50 undervolt nets a 4% performance increase
  • firestrike on steam deck doesnt scale well with more than 10W APU power
  • if you want to undervolt your steam deck you should always check if its stable and increases the performance

Thanks for doing the work!

When you put the results like that, with 2 to maybe 4% gains, I'm definitely going to pass on undervolting my deck. Back when I had a C2D Macbook - yeah you had to undervolt that sucker to make it usable. But now I will happily leave a 4% boost on the table so that I never have to consider whether a crash came from the undervolt.

All I see is higher tdp means bigger numbers, unless I'm missing something else?

The 20/50/50 UV gives 4% more performance compared to stock when both run at 10W TDP (column C and column J). And in this benchmark you don't gain much performance past 10W (you can see this in row 27 on the right side)