Best low power SFP+ cards
- Thread starter Mithril
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
I just realize that the TDP rating is not the only important factor when looking for low power cards.
It's also important to make sure that the card does not prevent the system from entering deep power saving state.
I run into a situation that my CX3 card prevents my CPU to enter PC8 package C-state.
As a result, the measured power consumption at outlet is around ~7.5w more, instead of ~3.8W more as listed in the spec.
See here for more details...
It's also important to make sure that the card does not prevent the system from entering deep power saving state.
I run into a situation that my CX3 card prevents my CPU to enter PC8 package C-state.
As a result, the measured power consumption at outlet is around ~7.5w more, instead of ~3.8W more as listed in the spec.
See here for more details...
Agreed. However, some extra notes:
That 3.8W spec is for the CX3, not total system power. Even with ASPM (etc), the total system's power increase will be theoretically higher than 3.8W (loss in PSU, more SoC/NB resources to support maintaining a PCIe link, etc). Mellanox/Nvidia specs usually list power consumption with passive cables (implicitly DACs), which can be a best case scenario (not always? afaik?). Any optic is likely going to draw more power (especially LR optics), with RJ45 adapters drawing even more power.
That 3.8W spec is for the CX3, not total system power. Even with ASPM (etc), the total system's power increase will be theoretically higher than 3.8W (loss in PSU, more SoC/NB resources to support maintaining a PCIe link, etc). Mellanox/Nvidia specs usually list power consumption with passive cables (implicitly DACs), which can be a best case scenario (not always? afaik?). Any optic is likely going to draw more power (especially LR optics), with RJ45 adapters drawing even more power.
I agree with you. But anyway the measured number is still very disappointing to me.
Note that the number I reported is in a testbench setup without any cable plugged. It's absolutely wasting power for nothing...
Once I read a suggestion from FreeNAS forum saying that having 10GE connection directly to the only high-speed client instead of the switch may save some power when the client is off. This turns out highly dependent on if the NIC in question can properly idle.
In a home usage scenario where the server could be idle for a fair amount of time, the idle power consumption might be more important than the TDP rating. But unfortunately the idle power consumption is usually not available from specs...
I really wish I can find an affordable card that can Idle within 5W...
The best option is still to buy a dual or quad 10G card and avoiding a 10G switch. Instead connect the server directly with the clients. By that the port itself powers completely down when the client is switched off. In addition you save 10W or even more for the switch. Of course this works only if your client has two network connections like 1G and 10G or Wifi and 10G.
The saving on switch is definite.
However the saving from 'the port itself powers completely down when the client is switched off' could be marginal.
I think we really need a good card with good driver to approach this part of the saving...
I missed to measure a fully disconnected port, but between 10G and 100M are 2 watts (RJ45):
Strom sparen mit powertop / Stromverbrauch von UnRaid verbessern
Not sure how huge the effect would be with active DAC cables.