XMR-STAK now printing INTERLEAVE NUMBERS

[rubantin]

[2019-01-18 10:17:12] : OpenCL Interleave 1|1: 18/1270.62 ms - 44.9
[2019-01-18 10:17:12] : OpenCL Interleave 0|1: 14/1349.53 ms - 45.0
[2019-01-18 10:17:15] : OpenCL Interleave 1|1: 15/1269.49 ms - 45.0
[2019-01-18 10:17:17] : OpenCL Interleave 1|1: 21/1271.78 ms - 45.0
[2019-01-18 10:17:17] : OpenCL Interleave 0|1: 38/1527.65 ms - 45.0

is now displaying as I run the miner... How do I get rid of those?

[Spudz76]

"reach a good delay" and they stop...

set interleave:45 or higher in amd.txt
default is 40 which is too small for your rig, which hits the auto interleave tuning and the message
eventually it should stop setting a new maximum and stop the message
whatever the highest number was, set at least that in the config

[rubantin]

Highest number is 560

[psychocrypt]

you should not increase it over 45 and it should not go higher than 50.0 than

[CryptoBroke550]

Does cpu mining cause issues with Amd interleave settings?
I can't seem to dial in my Rx 550 4GB cards. They count up like this and scroll like crazy.
If I go down the messages slow down but don't go away. Intensity change does not fix it.
I lose about 50H/s per card if I disable the interleave. So worth it if I can set it right.
Should I just set to 45 and forget it?
Right now mining Monero, set to 35 they count up to 40 and print to screen slowly.
Not sure what is better.

[Spudz76]

You should use affinity and ensure the CPU mining threads are always on CPUs that the GPUs are not on.
You can run the GPU threads on virtual CPUs to decent effect if you have all the regular real cores mining.

Also test with CPU disabled (--noCPU) to see most easily if that is what is bringing the AMD backend speed down (and timing problems may go away). Then adjust the affinities so nothing is on the same CPU, otherwise by default the OS just runs them wherever it feels like (usually wrong).

How many cards, and are each of them 2x threads?

[CryptoBroke550]

All (10) Rx 550 cards are using dual threads, I have three different cpu's right now, i7-2600 4 cores for mining 4 hyper threads with 4 cards on it. I have a i5-8600 4 cores mining and 2 not with 6 Rx 550cards and 2 Rx 580 cards. One i3-8300 4 cores mining running 6 nvidia cards.

Was thinking of putting all 10 Rx 550's on the i5-8600?
I have not mined without CPU's since interleave was added, easy to test out.

[CryptoBroke550]

Testing the i7 without cpu mining. --noCPU. 4 Rx550 cards all using 2x threads.
Started at 35, it just counts up and if I change the interleave higher as well it just counts up a little faster as I go. when I set it to 45 it counts to 50 and scrolls messages like mad.

So now lowering from 35 to 33, it scrolls for a bit, goes away, then comes back, then goes away and only one thread comes back counting up. I then change just that thread to 32.
Now I wait, in the past tests they would slowly come back and count up.

My next step is to affine_to_cpu. How many GPU threads should a CPU thread handle?
If CPU mining with 4 cores can 1 virtual core handle 2 GPU threads? or use two cores for GPU threads and 2 left for windows?