ACADEMY
MINING EDITION
MODULE 3
LESSON 2
Miner Performance Tuning for Qubic
Mining Qubic effectively requires not just good hardware but also smart tuning. Here we’ll cover how to squeeze the most performance out of your rig through thread optimization, balancing CPU/GPU usage, managing heat, and OS-level tweaks.
Thread Count Optimization
Unlike some simplistic miners, Qubic’s AI training can exhibit diminishing returns with too many threads or hyperthreading. It’s often not optimal to simply use 100% of your CPU cores with hyperthreading. Each additional thread increases work but also contention for resources like cache and memory bandwidth. Many miners find a sweet spot where using, say, 75% of logical cores yields higher iterations per second than using all cores with HT on . For example, a 16-core/32-thread CPU might perform best around 16–24 threads instead of all 32. Experiment with your thread count (CpuThreads setting) to see where your rig peaks – sometimes less is more, especially if running a GPU concurrently. Monitor the miner’s output or pool dashboard to compare solution rates.
CPU vs RAM Balancing
Qubic’s AI tasks are memory-intensive. Fast RAM and sufficient memory are crucial. Ensure you have at least 16 GB of RAM (preferably DDR4/DDR5 at high clocks) for a multi-threaded setup . If you allocate too many threads without enough RAM, your system may start swapping (which devastates performance). A key tuning here is enabling Huge Pages (large memory pages) on your OS . Huge pages allow the miner to access memory in bigger chunks, reducing overhead. The Qubic trainer will actually warn you if your system isn’t configured with enough large pages and tell you how many to enable . On Linux, this is as simple as running the suggested sysctl -w vm.nr_hugepages=<number> command as root before starting the miner. On Windows, enabling large page support requires granting the “Lock Pages in Memory” privilege to your user – an advanced step that can also improve performance in some cases. By matching huge pages to the miner’s needs, you can gain a noticeable uptick in iterations per second, especially on big multi-core systems.
GPU Considerations
If you mine with a GPU (Qubic supports both CPU and GPU mining, even concurrently), remember that VRAM and drivers matter. Qubic’s AI models can be memory-hungry, so GPUs with 8–12 GB VRAM or more are recommended . NVIDIA RTX 30/40-series and AMD RX 6000/7000-series cards are proven choices. Keep your GPU drivers up to date – driver optimizations can improve AI workloads. For multi-GPU rigs, ensure your power supply can handle the load and that cards are spaced for airflow. You may also limit each GPU’s power target (e.g., 80% TDP) to improve efficiency and thermals with minimal performance hit. If running hybrid CPU+GPU mining, watch your total system power and heat: a maxed-out CPU next to one or more busy GPUs in the same case can raise ambient temperatures, affecting both. In such cases, extra cooling or lowering one component’s usage (for example, using slightly fewer CPU threads when the GPU is also active) can prevent thermal bottlenecks.
OS-Level Tuning
Small software tweaks can yield performance gains. On Windows, set your Power Options to “High Performance” or “Ultimate Performance” so the CPU doesn’t downclock aggressively . Disable any “eco” or “cool & quiet” modes if you want maximum output. On Linux, use performance CPU governors (e.g., cpupower frequency-set -g performance) to avoid energy-saving throttling. If you’re comfortable with BIOS settings, you can try disabling SMT/Hyper-Threading to test if pure physical cores perform better for Qubic . Sometimes turning off SMT yields cooler temps and higher per-thread throughput, which can outpace having twice the threads running slower. Also, consider disabling CPU “Turbo Boost” or limiting the max frequency if your chip’s boost causes thermal throttling; a chip locked at a slightly lower GHz but running all cores consistently can beat one that oscillates due to overheating. Every system is different – the key is to observe Qubic’s performance metrics as you tweak. Keep an eye on your miner’s console output or pool statistics page when making changes, and adjust accordingly for the best stable output.