FurMark GPU Stress Test Tool

FurMark GPU Stress Test Tool | Generated by AI

Home 2026.07

Question: what is FurMark 1.39 GPU test tool?

Answer:

FurMark is a GPU stress-test / burn-in tool. It renders a very heavy “furry donut” OpenGL scene to push the GPU close to maximum load, mainly to test:

GPU cooling system
thermal behavior
power stability
GPU core stability
overclock/undervolt stability
used GPU condition when buying second-hand cards

FurMark 1.39.x is one of the older generation versions from Geeks3D. Version 1.39.0 was released in 2025; 1.39.3.0 is a later 1.39 branch build. (MajorGeeks)

The basic idea:

FurMark
   |
   v
generate extreme GPU workload
   |
   v
GPU core -> 99%
VRAM -> active
Power -> near TDP limit
Temperature -> maximum
   |
   v
observe:
  - crashes?
  - artifacts?
  - overheating?
  - clock throttling?

It is called a “GPU burner” because it intentionally creates an unusually heavy thermal load. (MajorGeeks)

For buying used RTX 3090 / 3060 / 7900 XT (your case)

FurMark is actually useful.

Typical second-hand test:

Run GPU-Z first:
- check GPU model
- VRAM size
- BIOS
- PCIe link
Run FurMark:

Resolution: 1920x1080
MSAA: 0x
Duration: 10-20 minutes

Watch:

Good RTX 3090 example

GPU usage:       99%
Power:           350W+
Temperature:     70-85°C
Clock:           stable
Artifacts:       none
Driver crash:    none

Bad card signs

temperature:
95°C+

clock:
1800MHz -> 900MHz

screen:
colored dots
flashing polygons
black screen

driver:
"NVIDIA driver stopped responding"

But FurMark is not a real AI workload test

For your LLM/training use case:

FurMark tests:

graphics pipeline
OpenGL
shader load
thermal limit

It does not test:

CUDA
Tensor Core
VRAM ECC behavior
PyTorch stability
long training stability

For AI GPU checking, I would combine:

1. VRAM test

python -m pip install gpu-burn
gpu_burn 600

memtest_vulkan

2. CUDA test

import torch

x = torch.randn(
    (20000,20000),
    device="cuda"
)

for i in range(100):
    y = x @ x
    torch.cuda.synchronize()

print("OK")

3. Real AI workload

Example:

vllm serve Qwen3-32B

torchrun train.py

Run several hours.

For your second-hand GPU shopping:

FurMark + GPU-Z + CUDA test is a good minimum.
For RTX 3090 specifically, also check:
- memory temperature (GDDR6X can hit 100°C+)
- fan noise
- power connector condition
- whether BIOS was modified for mining

FurMark only tells you “the card can survive a torture load”; it does not guarantee healthy VRAM. (TechPowerUp)

References:

FurMark GPU stress testing overview (MajorGeeks)
FurMark version history / downloads (TechPowerUp)

Back Donate