GTX 580: The AlexNet GPU

GTX 580: The AlexNet GPU | Generated by AI

Home 2026.07

Question: Intro to NVIDIA RTX 580, and did Ilya Sutskever and Hinton use it for AlexNet?

Answer:

There is no NVIDIA RTX 580. You probably mean NVIDIA GeForce GTX 580 (released in 2010). It was one of the GPUs that made the AlexNet breakthrough possible.

The GeForce GTX 580:

Architecture: Fermi
Release: November 2010
CUDA cores: 512
VRAM: usually 1.5 GB GDDR5
Memory bandwidth: ~192 GB/s
FP32 compute: ~1.5 TFLOPS
Power: ~244W TDP

By modern standards it is tiny:

RTX 4070: ~29 TFLOPS FP32, 12GB VRAM
RTX 4090: ~83 TFLOPS FP32, 24GB VRAM

But in 2012 it was a monster for deep learning.

Did Ilya Sutskever and Geoffrey Hinton use GTX 580 for AlexNet?

Yes, but the exact GPUs were not RTX 580. AlexNet was trained on two NVIDIA GTX 580 GPUs.

The famous 2012 paper:

“ImageNet Classification with Deep Convolutional Neural Networks”

by:

Alex Krizhevsky
Ilya Sutskever
Geoffrey Hinton

used:

2 × GTX 580
3GB total GPU memory (1.5GB each)
CUDA
custom GPU parallelization

The model had:

60 million parameters
650,000 neurons
5 convolution layers
3 fully connected layers

The clever trick was splitting the network across two GPUs.

Why was GTX 580 enough?

Because AlexNet was small by today’s standards.

A simplified comparison:

Model	Parameters	GPU memory needed
AlexNet (2012)	60M	~3GB
GPT-2 124M	124M	~GBs
Llama 7B	7B	~14GB+ FP16
GPT-4 class	unknown	huge

The breakthrough was not raw GPU power. It was realizing:

CNN + ReLU + dropout + GPU CUDA
        ↓
train much larger networks
        ↓
beat traditional computer vision

Interesting historical connection

Before AlexNet, GPUs were mostly for graphics.

AlexNet showed:

GPU
 |
 +-- graphics
 |
 +-- scientific computing
 |
 +-- neural network training

This directly created the modern NVIDIA AI business.

Jensen Huang’s famous quote:

“The more GPUs you sell, the more AI gets created.”

came from this transition.

The RTX generation came much later:

GTX 580 → Fermi (2010)
GTX 680 → Kepler (2012)
GTX 980 → Maxwell (2014)
GTX 1080 → Pascal (2016)
RTX 2080 → Turing (2018)
RTX 4090 → Ada Lovelace (2022)

Your RTX 4070 today has roughly 20× the compute and 8× the VRAM of the two GTX 580s that powered the AlexNet revolution.

Back Donate