How to configure Team Red Miner for AMD GPUs? (Stable Overclocks)

Understanding Team Red Miner Compatibility

1. Team Red Miner supports AMD GPUs from the GCN 4.0 architecture onward, including RX 400/500 series, Vega, and RDNA-based cards like RX 5700 XT, RX 6000, and RX 7000 families.

2. Older Polaris chips require driver version 20.40 or newer for full stability with TRM’s memory timing control features.

3. RDNA 3 GPUs such as RX 7900 XTX benefit from explicit VRAM timing overrides introduced in TRM v0.9.5, enabling tighter timings without kernel panics.

4. Integrated Radeon graphics are not supported — only discrete AMD GPU devices with dedicated VRAM are recognized by the miner.

5. The miner detects GPU topology automatically but may misreport dual-GPU cards like Radeon Pro Duo unless PCIe bifurcation is disabled in BIOS.

Core Configuration File Structure

1. TRM uses a JSON-formatted configuration file named config.json, which must reside in the same directory as the executable binary.

2. Each GPU is assigned an index starting from 0; entries under 'devices' array map settings per-index rather than by PCI ID.

3. The 'pools' section defines mining endpoints, supporting multiple failover URLs and automatic algorithm switching based on profitability API responses.

4. Memory clock offsets are specified in MHz using signed integers, while core clocks use absolute values in MHz — a critical distinction affecting stability.

5. Fan speed is defined via a 0–100% linear curve mapped to temperature thresholds, with interpolation enabled by default between defined points.

Stable Overclock Profiles for Mining

1. For RX 6800 XT: Set core clock to 2250 MHz, memory clock to +1200 MHz, and voltage to 950 mV — this yields consistent 62 MH/s on KawPoW with sub-70°C thermals under continuous load.

2. For RX 7900 GRE: Apply +150 MHz core offset, +2000 MHz memory offset, and lock VRAM timing to tRFC=360, tRP=24, tRCD=24 — achieves stable 112 MH/s on Etchash without reboots over 72-hour stress runs.

3. For RX 6600: Use fixed core at 2450 MHz, memory at +1800 MHz, and disable boost entirely via 'boost': false — eliminates thermal throttling spikes during Kaspa mining sessions.

4. For Vega 56: Lock VDDCI to 925 mV and SOC voltage to 850 mV while setting memory clock to +1000 MHz — prevents ASIC-like memory controller crashes common in older firmware revisions.

5. All profiles require 'auto-tune': false to prevent runtime frequency adjustments that destabilize hash submission intervals.

Thermal and Power Management Tuning

1. TRM reads GPU temperature directly from the on-die sensor, bypassing SMU polling delays — making fan curves more responsive than legacy tools.

2. Power limit is enforced via PCIe power budget negotiation; exceeding the board’s TDP rating triggers hardware-level current limiting, not software throttling.

3. The 'temp-limit' parameter halts mining threads if die temp exceeds the value, but does not reduce clocks — it pauses work entirely until cooldown.

4. VRAM junction temperature is monitored separately on RDNA 2+ cards; exceeding 105°C triggers immediate shutdown, regardless of core temp.

5. Undervolting must be applied incrementally: reduce voltage in 10 mV steps while verifying zero stale shares across three consecutive 10-minute intervals.

Frequently Asked Questions

Q: Why does TRM report “GPU not found” even though Radeon Software detects the card?A: This occurs when the GPU is bound to the amdgpu driver but lacks compute capability exposure — verify amdgpu.compute=1 is set in kernel boot parameters and that no conflicting drivers like radeon or amdkfd are loaded.

Q: Can I mix different AMD GPU models in one TRM instance?A: Yes, but each device requires individual tuning — TRM does not auto-balance settings across heterogeneous hardware, and mismatched memory timings can cause pool disconnections on slower cards.

Q: What causes “stale share” spikes after applying overclocks?A: Stales increase when memory latency variance exceeds ±5ns between read/write cycles — tighten tCL and tCWL in VRAM timing strings, or reduce memory clock offset by 100 MHz increments until stales drop below 0.3%.

Q: Is it safe to enable “--enable-restart” on unstable overclocks?A: No — automatic restart masks underlying instability; it may lead to repeated PCIe bus resets that corrupt shared memory buffers used by watchdog daemons.