How to use LolMiner for Nvidia/AMD mixed rigs? (Multi-Vendor Setup)

Mixed GPU Architecture Considerations

1. LolMiner officially supports both AMD and NVIDIA GPUs but requires distinct driver and runtime environments for each vendor. NVIDIA GPUs rely on CUDA 11.8 or higher, while AMD cards require ROCm 5.4.2 or newer for optimal compatibility.

2. Mixing vendors on a single mining rig introduces kernel-level conflicts when drivers coexist. Kernel modules like nvidia-uvm and amdgpu may compete for memory mapping resources unless properly isolated via cgroups or boot-time parameter tuning.

3. BIOS/UEFI settings must enable PCIe bifurcation and disable CSM to ensure stable enumeration of heterogeneous devices. Some motherboards fail to recognize AMD GPUs when NVIDIA drivers load first during initramfs.

4. Power delivery becomes asymmetric—NVIDIA cards often draw peak power in short bursts, whereas AMD cards sustain higher baseline consumption. PSU capacity must exceed the sum of TDPs by at least 30% to avoid brownouts during algorithm switches.

Driver Coexistence Strategies

1. Install NVIDIA drivers using the --no-opengl-files flag to prevent overwriting Mesa libraries used by AMD OpenCL stacks.

2. Deploy AMD drivers through official ROCm packages rather than amdgpu-pro; the latter includes closed-source components that interfere with CUDA context initialization.

3. Use LD_LIBRARY_PATH isolation per mining process: assign separate environment variables for each vendor’s compute runtime to avoid symbol resolution collisions during kernel launch.

4. Disable automatic driver updates via package manager holds—apt-mark hold nvidia-driver-535 and apt-mark hold rocm-dev prevent silent incompatibility after system upgrades.

LolMiner Configuration Syntax for Dual-Vendor Execution

1. Launch two independent LolMiner instances—one for AMD, one for NVIDIA—using distinct config files and process affinity masks. Example AMD command: ./lolMiner --coin ETHW --pool ethw.2miners.com:2020 --user YOUR_WALLET --devices 0,1,2 --lhr 0.

2. NVIDIA instance uses different pool endpoints and device indices: ./lolMiner --coin ETHW --pool ethw.2miners.com:2020 --user YOUR_WALLET --devices 3,4 --lhr 1.

3. Device numbering follows PCI bus order, not physical slot position. Verify with lspci | grep VGA and cross-reference with clinfo (for AMD) and nvidia-smi -L (for NVIDIA).

4. Avoid shared logging paths. Each instance writes to its own log file: --logFile amd_miner.log and --logFile nvidia_miner.log.

Thermal and Clock Management Separation

1. AMD GPU clocks are controlled via rocm-smi --setclock --level 0 --freq 1200, while NVIDIA uses nvidia-settings -a [gpu:0]/GPUMemoryTransferRateOffset[3]=1200. These tools cannot be applied interchangeably.

2. Fan curves must be managed separately—ROCm exposes fan control only on select Polaris and Navi chips, whereas NVIDIA supports PWM adjustment across all Turing+ architectures.

3. Thermal throttling thresholds differ: AMD cards trigger downclocking at 95°C junction temperature, NVIDIA at 89°C GPU hotspot. Monitoring must use vendor-specific utilities to avoid false negatives.

4. Undervolting requires hardware-specific profiles—AMD relies on rocm-smi --setvoltage with precise millivolt offsets, NVIDIA uses nvidia-smi -r -g 0 && nvidia-settings -a [gpu:0]/GPUVoltageOffsetAllPerformanceLevels= -125.

Frequently Asked Questions

Q: Can LolMiner auto-detect and balance workloads between AMD and NVIDIA GPUs?A: No. LolMiner does not implement cross-vendor workload distribution. Each GPU vendor requires a dedicated process with explicit device assignment.

Q: Does using both AMD and NVIDIA GPUs increase hashrate proportionally?A: Not necessarily. Algorithm efficiency varies—Ethereum Classic favors AMD’s high memory bandwidth, while KawPow shows better NVIDIA throughput. Raw summing ignores algorithm-specific bottlenecks.

Q: Is it safe to run lolMiner with --lhr 1 on AMD cards?A: No. The --lhr flag applies only to NVIDIA LHR-capable GPUs. Enabling it on AMD triggers undefined behavior and may cause immediate process termination.

Q: Why does nvidia-smi show zero utilization when AMD mining is active?A: This occurs when the NVIDIA driver fails to initialize CUDA contexts due to AMD’s ROCm kernel module locking PCIe configuration space. Rebooting with amdgpu.si_support=0 temporarily disables AMD compute features to test isolation.