How to increase hashrate on Linux miners? (Kernel Optimization)

Understanding Hashrate and Its Dependencies

1. Hashrate measures the computational power a mining rig contributes to a blockchain network, expressed in hashes per second (H/s).

2. It depends on hardware capabilities—GPUs, ASICs, or CPUs—but also on how efficiently the Linux kernel schedules tasks and manages resources.
3. Kernel-level bottlenecks include interrupt handling latency, CPU frequency scaling policies, and memory allocation strategies.
4. Suboptimal kernel configurations can throttle device access, delay GPU command submission, or cause unnecessary context switches during intensive hashing cycles.
5. Mining workloads are highly deterministic and repetitive, making them sensitive to microsecond-level scheduling jitter and I/O wait times.

Kernel Parameter Tuning for Mining Stability

1. Disabling transparent huge pages (THP) prevents unpredictable memory compaction delays: echo never > /sys/kernel/mm/transparent_hugepage/enabled.

2. Setting the CPU governor to performance mode locks frequencies at maximum: cpupower frequency-set -g performance.
3. Adjusting vm.swappiness to 1 minimizes swap usage under memory pressure, preserving RAM bandwidth for mining buffers.
4. Increasing net.core.somaxconn and net.ipv4.tcp_max_syn_backlog helps maintain stable stratum connections during high-frequency job submissions.
5. Disabling NMI watchdog with echo 0 > /proc/sys/kernel/nmi_watchdog eliminates periodic kernel interrupts that interfere with timing-critical hash loops.

Real-Time Scheduling and Process Prioritization

1. Assigning mining processes to real-time scheduling class (SCHED_FIFO) reduces preemption latency: chrt -f -p 99 $(pidof lolMiner).

2. Isolating CPU cores using kernel boot parameter isolcpus=2,3,4,5 prevents background daemons from interfering with mining threads.
3. Binding GPU miner processes to specific NUMA nodes ensures consistent memory access latency when feeding kernels to VRAM.
4. Using cgroups v2 to allocate dedicated CPU bandwidth and memory limits prevents system services from starving mining workloads.
5. Disabling irqbalance and manually affining GPU-related IRQs to isolated cores avoids interrupt storms during high-intensity kernel launches.

GPU Driver and Kernel Module Considerations

1. Loading NVIDIA drivers with NVreg_RegistryDwords='PerfLevelSrc=0x2222' forces fixed GPU clock states, eliminating dynamic throttling artifacts.

2. For AMD GPUs, setting amdgpu.ppfeaturemask=0xffffffff enables full power management control required for stable overclocked hashrates.
3. Compiling custom kernel modules with CONFIG_PREEMPT_RT patches improves determinism in GPU DMA buffer synchronization.
4. Blacklisting unused modules like bluetooth, snd_hda_intel, or uas reduces kernel module load time and memory fragmentation.
5. Enabling CONFIG_CPU_FREQ_STAT and CONFIG_IRQSOFF_TRACER allows precise profiling of scheduler-induced latency spikes affecting hash loop throughput.

Frequently Asked Questions

Q: Can kernel optimization increase hashrate on ASIC miners?A: ASIC miners typically run proprietary firmware and do not rely on Linux kernel scheduling for core hashing logic; however, kernel tuning still improves stratum communication stability and thermal reporting accuracy.

Q: Does disabling THP affect memory-intensive algorithms like Ethash?A: Yes—Ethash relies heavily on DAG memory access patterns; THP can cause page migration stalls and inconsistent read latencies, directly lowering effective hashrate by up to 3% on large DAGs.

Q: Why avoid the 'ondemand' CPU governor even if it scales up under load?A: The ondemand governor introduces variable latency during frequency transitions; mining requires immediate, sustained execution at peak clocks without ramp-up delays between job submissions.

Q: Is it safe to disable kptr_restrict and dmesg_restrict for mining systems?A: Yes—mining rigs operate in controlled environments where kernel pointer leakage is not a threat vector; disabling these restrictions allows monitoring tools to read internal scheduler metrics needed for fine-tuning.