
Guide for PC hobbyists and repair technicians on GPU memory testing, covering Kings_Overkill USB setup, BIOS tweaks, and interpreting test logs.
GPU Memory Testing Made Simple: Test RTX 3000 & AMD RX GPUs with the Kings_Overkill USB Drive
Published by Brav
Table of Contents
TL;DR
- Understand why GPU memory faults matter for performance.
- Set up a bootable USB with partitions and extract the test images.
- Disable Intel i3 graphics and UEFI in BIOS.
- Run the test, review the report.log and identify failing memory banks.
- Troubleshoot common pitfalls (partition loss, BIOS auto-run, large download timeouts).
Why this matters
I once spent hours chasing a mysterious black screen on a fresh GTX 750 Ti. Turns out the culprit was a single faulty VRAM chip. GPU memory errors can show up as artifacts, crashes, or a steady drop in frame rates, and they’re invisible to a quick hardware check. For a hobbyist or a repair tech, knowing whether the card’s memory is healthy is often the first step to deciding whether to replace or rework a GPU. MODS and MATS are the “leaked” Nvidia tools that have been used by repair shops for years to pinpoint such faults. According to the official guide, they’re powerful tools that test Nvidia cards for different kinds of faults, especially memory (MODS — Nvidia GPU Memory Testing Guide (2024)). While the early releases targeted GTX 1900-series cards, newer builds (400.xxx for Turing, 455.xxx for Ampere) now cover RTX 2000 and 3000 series, older GTX 16xx and 10xx cards, and even older GTX 700/600/500 series. The same guide shows a compatibility table: • 367.xxx – GTX 1000 and below • 400.xxx – RTX 2000 and below (incl. GTX 16XX) • 455.xxx – RTX 3000 and below However, MODS/MATS are Nvidia-only; they do not support AMD GPUs (MODS — Nvidia GPU Memory Testing Guide (2024)).
Core concepts
GPU memory architecture
GPUs have memory banks labeled a1, a0, b1, b0, …, X1, X0 – a simple clockwise numbering of paired memory chips on the GPU die. The same Nvidian guide explains the channel labeling and shows how to interpret error reports (MODS — Nvidia GPU Memory Testing Guide (2024)).
MODS vs. MATS
MODS (Modular Diagnostic Software) runs the whole card, while MATS (Memory Automated Test System) focuses on the VRAM. Together they form a firmware-level diagnostic suite.
Alternative tool – Kings_Overkill
A lesser-known bundle of three raw images (two 1.8 GB and a 31 GB PMF) plus a bootable Linux environment is the Kings_Overkill drive. It supports RTX 3000/2000, older GTX 16xx/10xx/9xx/7xx/6xx/5xx series, and all AMD RX 400/500/700/900/Vega/HD 7000 GPUs. The full test stack is described in the “Run nVidia/AMD GPU graphics memory tests” blog post (Run nVidia/AMD GPU Graphics Tests (2022)). The test runs from a Linux shell, logs into report.log, and will list failing memory banks such as a1 or b0 if errors are detected. The test can take up to an hour if many errors are found.
How to apply it
- Grab a 320 GB USB drive – the raw images total 31 GB, so an 80 GB minimum is needed; a 320 GB drive gives you a spare partition for backups.
- Create the four partitions – use DiskGenius to make four primary MBR partitions: 5 GB, 5 GB, 60 GB, and a spare. The manual explains how to create a new partition and set its size (DiskGenius — DiskGenius Manual (2025)).
- Download the three raw files – they’re password-protected; the key is “kings_overkill” (the blog notes this). Use a tool like JDownloader to resume the 31 GB file if the connection drops.
- Extract with WinRAR – WinRAR will unpack the PMF images into a folder. The extraction is simple; just run the installer and follow the prompts (WinRAR — WinRAR Official Site (2025)).
- Restore the images to the partitions – the 5 GB partitions hold the 1.8 GB images, the 60 GB partition holds the 31 GB PMF image. The blog explains that the restoration of the 31 GB image takes about 45 minutes.
- Prepare the BIOS – boot into BIOS, disable the Intel i3 integrated graphics (set IGPU off) and disable UEFI boot; the test requires legacy CSM mode (Run nVidia/AMD GPU Graphics Tests (2022)).
- Boot the USB – in legacy mode, the Linux menu appears. Choose the GPU you want to test (RTX 3000, RTX 2000, or AMD RX). If you don’t see the menu, the BIOS might have auto-run the test; ensure “Legacy boot” is the first option.
- Run the test – the Linux console will launch the MATS suite automatically. After completion, a file called report.log is written to the root of the USB. Use “less report.log” or “cat report.log” to view it.
- Interpret the results – look for lines like “FAIL a1” or “FAIL b0”; those indicate bad memory banks. The guide explains how to correlate the failing bank to the physical chip.
Pitfalls & edge cases
- Partitioning risk – an accidental format wipes your data. Always double-check the drive letter before saving.
- BIOS auto-run – if UEFI is still enabled, the test might launch without the menu, leaving you no chance to pick the GPU.
- Large file timeouts – the 31 GB download can stall on slow links; use JDownloader or resume capability.
- Older GPUs – the test may fail on very old cards like GTX 960; the guide warns that the software may not detect them.
- Runtime – a noisy card can push the test to the 1 hour ceiling; the log will still finish but it may seem stuck.
Quick FAQ
Q: What’s the difference between MODS and MATS? A: MODS runs a full card diagnostic; MATS focuses on the VRAM. They’re bundled together in the official suite. Q: Can I test an AMD GPU? A: Yes, the Kings_Overkill drive includes an AMD partition; the test environment supports RX 400/500/700/900/Vega/HD 7000 series. Q: Do I really need a 320 GB USB? A: The raw files total 31 GB, but you’ll need spare space for the boot files and potential re-writes; 80 GB is the minimum. Q: Why does the test sometimes run without showing the menu? A: That happens if UEFI is still enabled; disable UEFI or ensure legacy mode is first in the boot order. Q: How long will the test take? A: Roughly 30 minutes for a clean card; it can reach an hour if many errors appear. Q: What do the a1 and b0 labels mean? A: They are memory bank identifiers on the GPU die – each pair (a1/a0, b1/b0, etc.) corresponds to two VRAM chips.
Conclusion
If your card shows artifacts or crashes, run the Kings_Overkill test. A clean report.log means the memory is healthy; a failing bank means the chip is bad and the card likely needs a rewrite or replacement. For hobbyists, the process is straightforward once you have the right USB and BIOS settings. If you’re a technician, the same steps let you quickly triage a stack of cards in a shop. Happy debugging!