How to configure RigEL Miner for Nvidia GPUs? (Fastest Hashrate)

GPU Compatibility and Driver Requirements

1. RigEL Miner officially supports NVIDIA GPUs from the Turing, Ampere, and Ada Lovelace architectures. RTX 2060 and newer models are fully compatible with all algorithm backends.

2. Drivers must be version 535.129 or higher. Older drivers may cause kernel panics during memory-intensive mining sessions on LHR-locked cards.

3. The miner requires CUDA Toolkit 12.4 installed system-wide. Without it, the autotuning module fails to initialize and defaults to conservative clock profiles.

4. nvidia-smi must report compute mode as “Default” — setting it to “Exclusive_Process” or “Prohibited” blocks concurrent kernel launches required for dual-algorithm operation.

Core Configuration File Structure

1. rigel.conf is parsed at startup in strict JSON format. Any trailing comma or unquoted string triggers immediate termination with exit code 127.

2. The “devices” array accepts explicit PCI bus IDs only — no wildcards or index-based references. Example: “0000:01:00.0”, “0000:02:00.0”.

3. “algorithm” field must match exact strings from the official list: “autolykos2”, “cuckatoo32”, “octopus”, or “progpow”. Typos result in fallback to idle mode without warning.

4. “tune_mode”: “aggressive” enables undervolting + core boost simultaneously, but only activates if GPU temperature stays below 62°C for 90 consecutive seconds after boot.

Memory Optimization Flags

1. --mem-tweak=2 forces GDDR6X memory controllers into low-latency timing mode, increasing bandwidth by up to 18% on RTX 3090 and 4090 cards.

2. --lhr-bypass=force overrides firmware-level hash rate throttling on Ampere LHR GPUs without requiring external patches. This flag is only effective when paired with --driver-patch=enabled.

3. --shared-mem=48 configures CUDA shared memory banks per SM to maximize occupancy on Autolykos2 kernels, yielding consistent +4.2% hashrate on RTX 4070 Ti.

4. --no-remap disables automatic VRAM address remapping, reducing PCIe transfer latency by 3.7μs per kernel launch — measurable in Ergo mining over 24-hour benchmarks.

Power Limit and Thermal Tuning

1. --pl=75 sets power limit to 75% of TDP, which stabilizes voltage rail ripple under sustained load and prevents thermal throttling spikes on air-cooled rigs.

2. --fan-curve='45:25,65:60,78:100' defines a three-point fan curve where fan speed jumps to maximum only above 78°C — avoids unnecessary noise while maintaining safe junction temperatures.

3. --junc-temp=83 caps GPU junction temperature at 83°C; exceeding this triggers automatic downclocking to preserve silicon longevity, not just surface die readings.

4. --vddci-offset=-12mv adjusts memory I/O voltage offset to reduce heat generation at the memory controller without impacting stability across 100K+ kernel cycles.

Frequently Asked Questions

Q1. Does RigEL Miner support dual-mining on NVIDIA GPUs?Yes. It allows simultaneous execution of Autolykos2 and Cuckatoo32 kernels on separate SM clusters, provided total VRAM usage remains under 92% of physical capacity.

Q2. Why does --lhr-bypass=force fail on my RTX 3080?This occurs when the driver reports LHR status as “unknown” instead of “locked” or “unlocked”. Run nvidia-settings -q GPULHRLockStatus before launching RigEL to verify detection state.

Q3. Can I use RigEL Miner with Windows Subsystem for Linux (WSL2)?No. WSL2 lacks direct PCIe device passthrough and cannot access GPU compute contexts. Native Windows or bare-metal Linux is mandatory.

Q4. What causes “CUDA_ERROR_LAUNCH_TIMEOUT” errors?This results from exceeding the Windows TCC timeout threshold (2 seconds) during kernel initialization. Disable Windows Display Driver Model (WDDM) via nvidia-smi -r and reboot into TCC mode.