Folding Forum

I'm a bit late to the game on this one, but I just finished a review of an XFX RX 480 8GB.

https://greenfoldingathome.com/2019/02/ ... me-review/

My PPD average came out a bit lower than what I was expecting at 245K, but it might just be the work units I had over the three days. Basically, anyone looking to get into Folding@Home for cheap should go with an Nvidia GTX 1060 instead for both performance and efficiency reasons. But, it was an interesting card to play with. I like that AMD includes a compute mode on their windows graphics drivers. If only Nvidia would do the same!

An RTX 2080 is even better, if you can afford one!

I wished Nvidia or AMD would make stream processor cards for folding.
2GB of DDR6 ram, and 3000-5000 stream processors, should be cheaper for them to manufacture,than an entire graphics card!
Plus, these cards could be used as coprocessors for gamers.

Interesting,

I’m just getting my PPD monitoring working and my initial numbers show 3.83kPPD/W for one of my 1060 6GB and 3.96 for the other.

In comparison:
1070Ti 5.78kPPD/W
RTX 2060 6.81kPPD/W
RTX 2070 7.39kPPD/W

This is using the smoothed average of PPD reported by the FAHclient divided by the Power Draw reported by nvidia-smi sampled once a minute over 20 hours so far.

All cards are running Ubuntu 18.04.1 LTS Server and using the current Nvidia 410 or 415 drivers and these values are consistent with what I’ve observed.

I’ve also been capturing overall System Efficiency by sampling total PPD as reported by the FAHclient and dividing it by the output wattage reported by the SNMP card in the dual-conversion UPS two of the folding systems are connected to.

The system with the 2 RTX cards is averaging 6.33kPPD/W over the last 24 hours and the system with the 1070Ti and one of the 1060s 3.97kPPD/W but that system also seems to report about a 10W higher system overhead (UPS o/p less sum of reported GPU draw) and I suspect that the UPS needs it’s electrolytic capacitors replaced on its inverter (they are both over 10 years old and were being trashed at my workplace)

I have confidence in the UPSes reported power draw as it is consistent with results from my Kill-a-Watt and they (PowerWare, now Easton) are well engineered and I’ve been using their big brothers in data centres and telecommunications closets for many years.

Googling (yes, that is a verb in our household) to find an post in the Nvidia Developer Forums where the accuracy of the values reported by nvidia-smi was stated (+/- 5W) I also found this presentation which also cites this source.

This might help explain why MSI seems to report a lower wattage than your Kill-a-Watt