RTX 3070 Laptop hard-capped at ~55W when SysPwrLimitGetInfo fails on Lenovo Legion H1CN — no fallback to static vBIOS TGP

[blvcksenpai]

RTX 3070 Laptop GPU hard-capped ~55W: SBIOS power table negotiated as "unsupported", driver never applies an operating power limit

##THIS IS REPORT IS GENERATED BY OPENCODE

Environment

• Laptop: Lenovo Legion 5 15TH6H (Type 82JH), BIOS H1CN58WW (newest), 300W proprietary AC adapter, AC present (ADP0/online=1, battery Full)
• GPU: NVIDIA GeForce RTX 3070 Mobile / Max-Q (GA104M 10de:24dd), VBIOS 94.04.4F.00.1C, Lenovo subsystem 17aa:3f91; dGPU mode (hybrid off)
• OS: CachyOS, kernels tested CachyOS 7.2.2, 7.2.3 and 6.18.40/6.18.48 LTS
• Driver: tested open modules 610.43.03 and 610.57.04 (same results both, matched userland)
• Wayland/X11 GNOME, nvidia-powerd enabled (identical failure with it masked)

Symptom

GPU never draws above ~55 W regardless of workload, util, clocks, pstate or power source mode. At 100% util in games it downclocks to ~1400 MHz to honor the implicit budget (nvidia-smi P3, ~13–30 FPS depending on title). No thermal (60–65 °C), no EC brake, no eco/quiet mode.

Representative measure under NVENC H.264 4K60 (clocks unrestrained: 1935–1950 MHz, P0, VRAM 7001 MHz):

54.48 W, 1935 MHz, 2100 MHz, 23 %, 7001 MHz, P0
54.60 W, 1935 MHz, 2100 MHz, 16 %, 7001 MHz, P0
55.21 W, 1920 MHz, 2100 MHz, 23 %, 7001 MHz, P0

Power pins at ~55 W at only ~20 % utilization → this is a configured operating budget, not an active throttle (no throttle counters tick during it).

Relevant nvidia-smi state

GPU Ceiling Power Limit
    Current Power Limit    : 115.00 W
    Requested Power Limit  : N/A
    Default Power Limit    : 115.00 W
GPU Base Power
    Current Base Power     : N/A
    Requested Base Power   : N/A
    Default Base Power     : N/A
Min Power Limit           : 1.00 W
Max Power Limit           : 130.00 W   # EC allows this; RM refuses to use it

nvidia-smi -pl and NVML SetPowerManagementLimit: unsupported. Setting EC cTGP to 130 W (Lenovo toolkit/WMI) moves only the ceiling, never the actual draw.

nvidia-powerd (identical every boot, every driver/kernel tested)

nvidia-powerd version:2.0 (build 1)
DBus Connection is established
ERROR! DC power limits table is not supported
ERROR! Failed to get SysPwrLimitGetInfo!!
ERROR! Client (presumably SBIOS) has requested to disable Dynamic Boost DC controller

Hypothesis

The NVIDIA power-management framework reads the AC/Dynamic-Boost power table from the platform (NVPCF communicate) via SysPwrLimitGetInfo. This OEM SBIOS responds so the request fails ("DC power limits table is not supported" / "disable Dynamic Boost DC controller"), and the RM silently falls back to a conservative ~55 W operating budget instead of the static vBIOS base TGP of 115 W. The same failure signature is reported in Bazzite #4384 (MSI GP66, RTX 3070 Laptop) and CachyOS-Settings #203 (Lenovo Legion 82JU). Note: in Bazzite #4384, disabling powerd restored full TGP; on this Legion H1CN platform disabling powerd does NOT restore power (cap persists ~55 W on all stacks above).

Impact

Games run below half the GPU's intended performance. Mitigation nvidia-smi -lgc 2000 (lock clocks) does not raise the budget; power still settles ≤ ~55 W.

Request

• When power-table negotiation with the SBIOS fails, fall back to the static/vBIOS base TGP (115 W here) rather than a fixed reduced budget; or
• make the power-table fetch resilient to SBIOSes that answer "unsupported" (e.g. re-negotiate via the standard platform interface, retry, or accept the EC ceiling).

[blvcksenpai]

Cross-referencing: vendor-cross-cutting cluster, not a one-off

Additional data points confirming the same failure class across multiple vendors and GPU generations:

Issue	GPU	Symptom
#1295	RTX 4070 Laptop	"105W laptop stuck at 55W" — same 55W ceiling, powerd does nothing, -pl unsupported
#1310	RTX PRO 1000 Blackwell / ThinkPad P14s	PlatformRequestHandler fails at init (platform_request_handler_ctrl.c:2117), GPU locked 25W vs reported 35/50W; Windows = 38.9W / +47% FPS
#1162	RTX 5060 Blackwell / HP OMEN Slim	Power Management Object: N/A — Device(NPCF) at \_SB.NPCF (not a PCIe child of GPU) never bound by open module; static 80W cap; OS-independent
#966	RTX 5080 / ASUS	DB DC enabled-check happens only at daemon startup; errors irreversibly if unavailable at boot — root design flaw
#1059	RTX 4060 / Legion Slim 5 16APH8	Restarting powerd alone preserves PM; reloading nvidia_drm alongside breaks it — shows statefulness of negotiation path
#392	AMD 4000/5000 + NVIDIA (triaged NV-Triaged)	"Dynamic Boost doesn't work on AMD CPUs" — acknowledged 2022, fix claimed in 535, still reported broken in 545/550 per 2024 comments

This machine (additional context)

• Laptop: Lenovo Legion 5i Pro 16ACH6H (82JH), BIOS H1CN58WW
• GPU: RTX 3070 Mobile / Max-Q (GA104M 10de:24dd), VBIOS 94.04.4F.00.1C
• Driver: 610.57.04 open kernel module (also reproduced on 610.43.03)
• Kernel: CachyOS 7.2.3-1-cachyos, also 6.18.48-1-cachyos-lts

Key evidence:

• Power Management Object: N/A — same as #1162 (Blackwell/HP OMEN), but on Ampere — suggests the NPCF binding gap for Device(NPCF) at \_SB.NPCF applies across generations, not just Blackwell.
• No kernel-level PlatformRequestHandler assertion in dmesg — unlike #1310 (Blackwell), the failure is silent at module level on Ampere; surfaced only by nvidia-powerd userspace errors.
• Three identical powerd errors on every boot, every driver/kernel tested:

  ERROR! DC power limits table is not supported
  ERROR! Failed to get SysPwrLimitGetInfo!!
  ERROR! Client (presumably SBIOS) has requested to disable Dynamic Boost DC controller
  

• GPU hard-capped at ~55W regardless of load/util/pstate. nvidia-smi -q: Requested Power Limit: N/A, Current/Default Power Limit: 115W, Max: 130W. nvidia-smi -pl rejected. sw_power_cap active throughout load; no thermal throttle.
• Disabling/masking nvidia-powerd does not restore power — cap persists on all tested stacks. (Contrasts with #1295 and CachyOS #203 where disabling powerd resolves.)
• EC cTGP raise via WMI (Lenovo Commercial Vantage path) moves the Max Power Limit ceiling (115 → 130W) but the actual draw remains ~55W — i.e. the ceiling is adjustable but the RM's operating budget is not.

Request

1. When the platform power-mode query fails (SysPwrLimitGetInfo / PRH), please fall back to the static vBIOS base TGP rather than leaving the GPU governed by an undocumented conservative default (~55W in this cluster). The vBIOS clearly advertises 115W (Default Power Limit) — the driver should honour it when AC/Dynamic Boost negotiation is unavailable.
2. Investigate the NPCF binding path for pre-Blackwell GPUs where Device(NPCF) lives at \_SB.NPCF (not as a PCIe child of the GPU device). The Power Management Object: N/A on Ampere (GA104M) suggests the same binding gap identified in #1162 applies here too, which would explain why NVPCF sub-func#2 replies have no effect on the power budget.