Your GeForce runs AI at half speed. This fixes it for 20/30/40 series cards. Up to 2× FP16 / BF16 / TF32 tensor throughput
C
/* SPDX-License-Identifier: GPL-2.0-only */
#ifndef GPU_UNLOCK_UNLOCK_610_H
#define GPU_UNLOCK_UNLOCK_610_H
#include "firmware_610.h"
/*
* Run the selected family recipe on a cold, unbound supported card, or one nvidia suspended
* before a suspend to RAM and resumes after this returns. The caller holds
* pci_dev_lock and supplies the context resolved during enumeration. Returns
* 0 when the card is unlocked and handed back as found;
* otherwise a negative errno. A failed transaction may have applied a prefix
* of the table: it must not be reported as ready. If quiescence cannot be
* established, DMA buffers and BAR ownership are retained until power off.
* After resume,
* -EALREADY means the card kept power and is still unlocked, and nothing was
* written. Unlocks of different cards may run concurrently with the same
* shared firmware.
*/
int gpu_unlock_610_unlock(const struct gu_device *device, const struct gpu_unlock_610_shared *shared,
bool resume);
#endif