What are the best mining settings for RTX 4090? (Overclock)

Core Clock and Memory Tuning

1. Core clock adjustments on the RTX 4090 typically remain stable between +50 MHz and +120 MHz when paired with adequate cooling and power headroom.

2. Memory clock gains show diminishing returns beyond +1300 MHz due to GDDR6X thermal saturation and signal integrity constraints under sustained mining workloads.

3. A common sweet spot observed across multiple firmware revisions is +85 MHz core and +1150 MHz memory, delivering consistent hash rate uplift without triggering undervolting instability.

4. Exceeding +150 MHz core offset often introduces kernel panics during DAG epoch transitions in Ethereum-based algorithms, especially after the 5.5K epoch boundary.

Power Limit and Voltage Control

1. Power limit reduction to 70–75% of TDP (i.e., 350W → 250–275W) enables tighter thermal envelopes while preserving >92% of baseline hashrate in KawPoW and Ethash variants.

2. Manual voltage curve editing below 850 mV at 1800 MHz core frequency leads to immediate LHR unlock failure and persistent PCIe bus errors.

3. Fixed voltage mode is discouraged; dynamic voltage-frequency scaling remains essential for maintaining algorithm-specific latency tolerance during memory-bound operations.

4. Undervolting below 780 mV at stock frequencies increases stale share rates by 3.7–5.2% on ETC networks due to timing margin erosion in memory controller response cycles.

Cooling and Fan Curve Optimization

1. GPU junction temperature must be held under 62°C to prevent automatic clock throttling during multi-hour Ethash DAG generation phases.

2. A fan curve starting at 35% speed at 45°C and reaching 88% at 65°C balances acoustic output with sustained thermal headroom across dual-slot air-cooled configurations.

3. VRAM hotspot temperatures exceeding 98°C correlate directly with increased memory error correction events and eventual DAG verification failures in NBMiner v42+.

4. Blower-style coolers consistently deliver 4.3–6.1°C lower VRAM junction readings than open-air triple-fan designs under identical ambient conditions and overclock profiles.

LHR Unlock Techniques and Stability Testing

1. LHR bypass success rates drop from 94% to 61% when memory clock exceeds +1250 MHz on factory-overclocked Founders Edition units running driver version 535.98.

2. Repeated application of --lhr_tune 72 in T-Rex Miner v0.26.7 yields more stable LHR suppression than auto-tune modes during network difficulty surges.

3. Stability validation requires minimum 18-hour continuous runtime with real-time DAG epoch progression, not synthetic benchmark loops.

4. Intermittent LHR re-engagement occurs in 19% of cases when mining switches between ETC and ETHW without full GPU reset, indicating firmware-level state leakage.

Firmware and Driver Compatibility Notes

1. Driver versions 525.85.12 and 535.43.02 show highest LHR consistency across 4090 SKUs but introduce 2.1% higher stale share incidence on Zilliqa-based mining pools.

2. VBIOS revisions ending in “-012” or “-014” enable memory timing overrides unavailable in “-008” and earlier variants, critical for sustained 2.2 Gbps+ memory bandwidth.

3. Linux kernel 6.2+ introduces PCIe ACS override patches that reduce LHR false positives by 33% compared to kernel 5.15 LTS in multi-GPU rigs.

4. Windows WDDM mode disables native CUDA compute priority; switching to TCC mode on Tesla-branded drivers is incompatible with all consumer 4090 models.

Frequently Asked Questions

Q: Does increasing memory clock improve hashrate linearly on RTX 4090?A: No. Gains plateau sharply after +1100 MHz due to GDDR6X PHY saturation. Further increases raise error rates without measurable performance lift.

Q: Can I use MSI Afterburner’s voltage slider for 4090 mining?A: Not safely. The slider applies global voltage offsets incompatible with per-rail regulation logic in AD102. Use EVGA Precision X1 or command-line tools with per-domain control.

Q: Why does my 4090 report 'LHR detected' even with --lhr_tune 72?A: This indicates memory timing misalignment. Try reducing memory clock by 50 MHz increments while holding core and voltage constant until detection ceases.

Q: Is liquid cooling necessary for stable 4090 mining overclocks?A: Not mandatory, but air-cooled setups require strict adherence to sub-60°C GPU temp targets. Custom loop users report 12–15% longer mean time between failures at identical clock offsets.