Ocassionally GPU rig won't run at the full speed - why?

[cc4fun]

Hi,
The rig is: 7x RX 480 8GB, MSI Gaming 5,  15.12 drivers, Win 10
For last several days I observe a new issue:
From time to time this rig will only run at less than 50% of the usual speed (drawing some 70% of the usual power)
This is does not depend on:
- miner (happens with Claymore's ZCash 9.2 and Claymore's CryptoNote 9.6)
- miner settings (like voltages)
- pool (happens with flypool and dwarfpool)

The strange thing is that I can restart the rig many times during an hour or so and the issue is still there. Then magically, without any modifications to the settings or HW, the miner launches at the full speed.

Has  anyone encountered such issue?

edit: When this happens GPU and MEM clocks behave normally that is they run at the max speed.

[eckmar]

How close are your GPUs one to another (phisically) ? What are your temperature settings ? It could be that your cards are thermal throttling.

[cc4fun]

How close are your GPUs one to another (phisically) ? What are your temperature settings ? It could be that your cards are thermal throttling.

There's a GPU each 9cm along the rack.

When this issue occured just now I turned the rig off, cooled down the whole room, GPU temp dropped by 10C, started the rig and it... launched with the full speed yay!

So your guess may be the answer to my question! I will speed up the fans and see if that solves the problem as actually they spin very slow. Maybe I have to relax my low noise requirements...
Thanks

[m0niker]

The cards might be decreasing their clock rate, if worst comes to worst try plugging on a dummy monitor into the cards that run slow

[eckmar]

How close are your GPUs one to another (phisically) ? What are your temperature settings ? It could be that your cards are thermal throttling.

There's a GPU each 9cm along the rack.

When this issue occured just now I turned the rig off, cooled down the whole room, GPU temp dropped by 10C, started the rig and it... launched with the full speed yay!

So your guess may be the answer to my question! I will speed up the fans and see if that solves the problem as actually they spin very slow. Maybe I have to relax my low noise requirements...
Thanks

Since you want low noise, guess your cards are somewhere in your house. I put all my rigs in garage so I don't have to care about noise. It's always better to run fans as high as you can but not over 80% (only in case it's needed), since it's does not matter how fast fans spin (they won't last longer if they spin slower). What matters to fans is change in speed, so you should make your speed constant, not dependent on temperature.

[cc4fun]

How close are your GPUs one to another (phisically) ? What are your temperature settings ? It could be that your cards are thermal throttling.

There's a GPU each 9cm along the rack.

When this issue occured just now I turned the rig off, cooled down the whole room, GPU temp dropped by 10C, started the rig and it... launched with the full speed yay!

So your guess may be the answer to my question! I will speed up the fans and see if that solves the problem as actually they spin very slow. Maybe I have to relax my low noise requirements...
Thanks

Since you want low noise, guess your cards are somewhere in your house. I put all my rigs in garage so I don't have to care about noise. It's always better to run fans as high as you can but not over 80% (only in case it's needed), since it's does not matter how fast fans spin (they won't last longer if they spin slower). What matters to fans is change in speed, so you should make your speed constant, not dependent on temperature.

OK so after some time I'm 90% sure it was thermal throttling. Thank you again eckmar.
Strange though, that MEM and GPU clocks are not downclocked in such case.

And yes you're right I keep this rig in my house (and I can't move it elsewhere). So I will have to deal with extra noise in one of the rooms. Anyway I will try to add extra fans to this rig so that GPU fans can spin slower and still maintain a reasonable temp.

[joblo]

OK so after some time I'm 90% sure it was thermal throttling. Thank you again eckmar.
Strange though, that MEM and GPU clocks are not downclocked in such case.

Throttling is downclocking. What are the GPU temps? If they were too hot they would cool when you reboot.
If rebooting doesn't fix it the problem may be outside your rig like networking issues, pool issues etc.