¡CURSO BÁSICO DE TARJETAS GRÁFICAS! POR FIN VAS A ENTENDERLO TODO.

Overview

• Hands-on tutorial explaining GPU power delivery (VRMs/phases), how to diagnose common faults, and how to repair a desktop graphics card (GTX 950 used as the example).
• Emphasis on measuring with a multimeter and using a simple oscilloscope to observe PWM/gate signals.
• Practical guidance on component replacement (MOSFETs, capacitors).

Key technological concepts explained

PCIe connector pins and rails

• First three PCIe pins carry 12V (power to VRM/phases); the motherboard also supplies 3.3V.
• The last pin (PEX enable) activates the PEX coil — enable logic sometimes implemented via a resistor.
• PCIe data pairs go directly to the GPU; continuity/open/short behavior of those pairs helps differentiate GPU vs PCB/VRM issues.

VRM / phase topology

• Each phase consists of high-side and low-side MOSFETs, an inductor (coil) and output capacitors. Together these convert 12V to the GPU Vcore, memory rails (e.g., ~1.3–1.5V), etc.
• MOSFET arrangement varies (1–3 MOSFETs per phase). Dual/integrated packages (DOR) are common.

Gate drivers and phase controller

• Each phase’s MOSFETs are driven by a gate-drive IC; the gate driver requires its own supply and a PWM input (IN).
• A phase-controller IC (example: NCP81174) coordinates PWM outputs (GA1..GA4) to the gate drivers and includes enable/config pins and its own VDC supply.
• Controller configuration matters: some controllers use 1.8V or 3.3V logic; enable thresholds are typically around ~2.4–2.5V. Always check the datasheet for specifics.

PWM behavior and smoothing

• High-side and low-side MOSFETs receive complementary PWM pulses. The coil plus capacitors smooth these pulses into a linear DC output (Vcore).
• Oscilloscope observations:
  • Gate pulses vary in width between phases.
  • The MOSFET output into the coil shows thin PWM; after filtering by the coil and capacitors the output appears as a steady DC voltage.

Diagnostic tips & measurement interpretation

Multimeter continuity/resistance

• Example: measuring ~7 Ω between 12V pins and negative can indicate GPU internal resistance (not necessarily a hard short).
• Very low resistances (≈0.2 Ω) on modern cards can reflect the massive transistor count and are not always a fault indicator.
• If the 12V rail measures near zero ohms to ground, that suggests a short.
• Open or shorted data pairs typically point to a GPU-related issue (e.g., reballing needed) rather than a VRM PCB fault.

Phase checks

• Measure continuity between phase outputs and ground to infer MOSFET/coil/GPU connectivity.
• Check MOSFET gate pin connectivity (example pin 4). Gate short/open behavior helps identify faulty MOSFETs.
• If one MOSFET in a phase is bad, replace all MOSFETs in that phase to reduce the risk of cascading failures.

Controller checks

• If no PWM pulses are present, verify:
  • Controller IC power (VDC pin).
  • Enable pin state.
  • Configuration bits.
• Measure gate outputs (GA1..GA4); sequential pulses are normal since phases are often fired in sequence.

Practical signs

• Overheated/burnt MOSFETs, bulging/bad capacitors, or smoke indicate serious failure.
• The card may run on reduced phases after repair but will have degraded or unstable behavior until fully restored.

Repair workflow (as shown)

1. Visual inspection for burn marks, overheated components, bulging capacitors.
2. Measure PCIe pins (12V, 3.3V). Check BIOS 3.3V and any 5V regulator presence.
3. Measure resistance/continuity on phase outputs, MOSFETs, and data pairs to isolate GPU vs VRM issues.
4. Remove suspected faulty MOSFETs/components from the affected phase(s).
5. Replace MOSFETs with compatible N-channel parts (match channel type, voltage, current and thermal ratings). DOR packages can be direct replacements.
6. Reflow/solder quickly — pre-position parts, then heat — to minimize thermal stress.
7. Power up carefully; check voltages (Vcore, memory rails) and use an oscilloscope to verify PWM pulses and smoothed outputs.
8. If the card shows an image but runs on reduced phases, continue replacing all components in that phase for long-term reliability.

Practical notes and cautions

• Replacing a single MOSFET in a degraded phase is risky — replace the whole phase’s switching components where possible.
• Use compatible N-channel MOSFETs; P- vs N-channel behavior differs and matters for replacement choices.
• Always consult the controller datasheet for pin functions, enable thresholds, and configuration details.
• A simple oscilloscope is sufficient to visualize gate/PWM pulses for diagnosis.
• Take care when powering up during testing to avoid further damage (use current-limited PSU or other protective measures).

Products, parts and examples mentioned

• GPU examined: GTX 950 (used as the example board). The GTX 950 was compared to GTX 960 for context; modern RTX cards were referenced to illustrate very low internal resistance.
• Phase controller IC example: NCP81174 (used to illustrate pin functions and behavior).
• MOSFETs: N-channel MOSFETs for high/low side switching. DOR (dual) MOSFET packages referenced as common replacements.

Tutorial / review elements

• Presented as a basic course/tutorial for GPU repair: step-by-step fault detection, component-level diagnosis, oscilloscope examples, and replacement guidance.
• Demonstrates measuring techniques and how to interpret readings (e.g., what ~7 Ω implies).
• Includes real repair process examples, showing mistakes (smoke, partial repair) as learning points.

Main speakers / sources

• Presenter/channel: the video author (referred to as “reprom”); the presenter demonstrates and narrates the course.
• Hardware references: GTX 950 board used for examination; RTX family referenced for comparison.
• Controller IC referenced: NCP81174 (phase controller; used as a datasheet/configuration example).