alpha-miner v1.3.0

alpha-miner v1.3.0

Performance improvements for the 5000-series and Hopper datacenter cards (H100 / H200). Linux x86_64 build, NVIDIA Volta through Blackwell.

What's new

Improvements to the 5000-series and H100/H200 kernels.

Performance (Jonathan's numbers on reference hardware)

GPU	Hashrate
H200	~600 TH/s
H100	550–580 TH/s
5090	280–300 TH/s (thermal-throttle dependent)
5080	~unchanged from v1.2.0
4090	~160 TH/s (validated on internal rig)
V100 / CMP 100-210	unchanged from v1.2.0

Architecture support

--force-backend accepts: volta | ampere | ada | hopper | blackwell

Architecture	SM	GPUs
Volta	SM_70	V100, Titan V, CMP 100-210
Ampere	SM_80 / SM_86	RTX 30-series, A100, CMP 170HX/220HX
Ada	SM_89	RTX 40-series, L4, L40
Hopper	SM_90	H100, H200
Blackwell	SM_120	RTX 50-series, B100, B200

Turing (RTX 20-series, T4, CMP 30/40/50/90HX) is the only remaining gap — kernels pending from upstream.

Quick install (Linux)

curl -L -o alpha-miner https://pearl.alphapool.tech/downloads/alpha-miner
chmod +x alpha-miner
./alpha-miner \
  --pool stratum+tcp://pearl.alphapool.tech:5566 \
  --address prl1pYOUR_ADDRESS \
  --worker rig01

For static difficulty (recommended for high-end cards or HiveOS rigs that reconnect frequently):

./alpha-miner ... --password "x;d=65536"

Suggested d values: 3060Ti–3070 → 4096–16384, 3090/4080 → 65536, 4090/H100/H200 → 131072–524288, 5090 → 262144+.

Docker

docker run --gpus all \
  -e PEARL_ADDRESS=prl1pYOUR_ADDRESS \
  -e PEARL_WORKER=docker-rig \
  alphaminetech/pearl-miner:1.3.0

Set PEARL_DIFFICULTY=N for static difficulty.

Regional endpoints (lower latency)

Region	Endpoint
US East	pearl.alphapool.tech:5566
US West	us2.alphapool.tech:5566
Europe	eu1.alphapool.tech:5566
Asia	sg1.alphapool.tech:5566

Solo mode: same hosts, port 5567.

Assets

File	SHA256
alpha-miner (Linux x86_64)	ea1bc6c147fb35dacdd9d7ffc1d69a88511512c41788a048de99da46faeb9494
alpha-miner-hiveos.tar.gz	1b8600c31b88b5eadfbba3a5906a7ebb2b58d84348f6e8225dc5c62cb2ec30e3