Nouveau vs. NVIDIA Power Monitoring Results

[phoronix]

Phoronix: Nouveau vs. NVIDIA Power Monitoring Results

In complementing the Nouveau vs. NVIDIA OpenGL benchmark results published earlier this week on Phoronix, here are the power consumption and performance-per-Watt metrics...

http://www.phoronix.com/vr.php?view=MTc1NDA

[halfmanhalfamazing]

You could have at least embedded the second table into the article.

[Marc Driftmeyer]

Sure looks like the binary blob is stomping the crap out of the FOSS binary.

[Calinou]

Sure looks like the binary blob is stomping the crap out of the FOSS binary.

It's not that far behind in Xonotic. Warsow always had bias towards proprietary drivers.

Sadly, this brings again the “ecology versus free software” issue. I would probably choose the later.

[blackout23]

Shouldn't it be possible to trace how the NVIDIA blob reclocks these cards and then to just replay the reads and writes to the registers in nouveau?

[DanL]

Sure looks like the binary blob is stomping the crap out of the FOSS binary.

[sarcasm]Hmmm, I wonder why...[/sarcasm]

[Luke]

If you want both ecology and free software

It's not that far behind in Xonotic. Warsow always had bias towards proprietary drivers.

Sadly, this brings again the ?ecology versus free software? issue. I would probably choose the later.

Get either an Nvidia card for which reclocking works fully or an AMD card and run the open drivers. On AMD the reclocking is essentially the same on open or blob, as AMD open-sourced their entire PM code. Nvidia dropped some code too, I don't know if they didn't release enough information or if due to their lack of earlier development help for Nouveau the driver was so different that their PM code could not be used without a shitload of modification.

When running any kind of card that only supports static reclocking, you can use scripts or other untilities to turn it all the way down for desktop (even on small cards!), to mid level for playing back big videos, and all the way up for games.

If you are running small or midsize cards and not huge gaming cards, look for electrical savings elsewhere. Get rid of any remaining incandescent bulbs, replace them with LED bulbs. Replacing just one 100W bulb with an LED will cover all the load of running a card like my Radeon HD6750 full power full time, assuming the light and the card get equal use. Three 100W incandescent bulbs literally use the power of the biggest single GPU cards in existance wide open and overclocked.

[Luke]

More details (edit limit)

Get either an Nvidia card for which reclocking works fully or an AMD card and run the open drivers..

For the GTX 650 I see almost the same power consumption in Xonotic (not weighted against FOSS drivers) as in the blob. Performance vs the blob at or near the results with AMD-and the GTX650 is a midsize card just like my Radeon HD6750. Like I said, this driver is catching up fast!

[My8th]

It be interesting to see what the frames per watt would be if the temps using both drivers were the same, would we get closer frames per watt.

[curaga]

No mention of why the blob failed to report wattage for the gt520?