Throttling with multiple instances

[xyven1]

When using multiple instances the overall gpu usage drops.
gtx 1070, i7-7700k, 32gb ddr4

Lc0 client version 22
2019/04/09 00:23:11 lc0_main.go:786: serverParams: [--visits=10000 --cpuct=2.5 --resign-percentage=4.0 --resign-playthrough=20 --temperature=1.2 --temp-endgame=0.45 --temp-cutoff-move=16 --temp-visit-offset=-0.25 --fpu-strategy=reduction --fpu-value=0.5 --fpu-strategy-at-root=reduction --fpu-value-at-root=0 --minimum-kldgain-per-node=0.000008 --resign-wdlstyle=true]
Downloading network...
Args: [C:\Users\matla\Documents\Applications\lc0/lc0 selfplay --backend-opts=backend=cudnn-fp16 --parallelism=32 --visits=10000 --cpuct=2.5 --resign-percentage=4.0 --resign-playthrough=20 --temperature=1.2 --temp-endgame=0.45 --temp-cutoff-move=16 --temp-visit-offset=-0.25 --fpu-strategy=reduction --fpu-value=0.5 --fpu-strategy-at-root=reduction --fpu-value-at-root=0 --minimum-kldgain-per-node=0.000008 --resign-wdlstyle=true --training=true --weights=networks\5c758506d57f02f06ef1910425d6d439705364fec08a077b883b53452d632012]
_
| _ | |
|_ |_ || v0.21.1 built Mar 23 2019
id name Lc0 v0.21.1
id author The LCZero Authors.
Loading weights file from: networks\5c758506d57f02f06ef1910425d6d439705364fec08a077b883b53452d632012
Creating backend [multiplexing]...
Creating backend [cudnn-fp16]...
GPU: GeForce GTX 1070
GPU memory: 8 Gb
GPU clock frequency: 1835 MHz
GPU compute capability: 6.1
CUDA Runtime version: 10.0.0
Cudnn version: 7.3.1
Latest version of CUDA supported by the driver: 10.1.0
2019/04/09 00:23:15 lc0_main.go:637: GPU probably doesn't support the cudnn-fp16 backend
2019/04/09 00:23:16 lc0_main.go:786: serverParams: [--visits=10000 --cpuct=2.5 --resign-percentage=4.0 --resign-playthrough=20 --temperature=1.2 --temp-endgame=0.45 --temp-cutoff-move=16 --temp-visit-offset=-0.25 --fpu-strategy=reduction --fpu-value=0.5 --fpu-strategy-at-root=reduction --fpu-value-at-root=0 --minimum-kldgain-per-node=0.000008 --resign-wdlstyle=true]
Args: [C:\Users\matla\Documents\Applications\lc0/lc0 selfplay --backend-opts=backend=cudnn --visits=10000 --cpuct=2.5 --resign-percentage=4.0 --resign-playthrough=20 --temperature=1.2 --temp-endgame=0.45 --temp-cutoff-move=16 --temp-visit-offset=-0.25 --fpu-strategy=reduction --fpu-value=0.5 --fpu-strategy-at-root=reduction --fpu-value-at-root=0 --minimum-kldgain-per-node=0.000008 --resign-wdlstyle=true --training=true --weights=networks\5c758506d57f02f06ef1910425d6d439705364fec08a077b883b53452d632012]
_
| _ | |
| |_ |_| v0.21.1 built Mar 23 2019
id name Lc0 v0.21.1
id author The LCZero Authors.
Loading weights file from: networks\5c758506d57f02f06ef1910425d6d439705364fec08a077b883b53452d632012
Creating backend [multiplexing]...
Creating backend [cudnn]...
GPU: GeForce GTX 1070
GPU memory: 8 Gb
GPU clock frequency: 1835 MHz
GPU compute capability: 6.1
CUDA Runtime version: 10.0.0
Cudnn version: 7.3.1
Latest version of CUDA supported by the driver: 10.1.0