← Retour au blog
tech 14 July 2026

Benchmarking 15 "E-Waste" GPUs with Modern Workloads

Explore the efficiency of older Tesla GPUs in modern contexts. Discover how to maximize the use of these often underestimated resources.

Article inspired by the original source
Benchmarking 15 "E-Waste" GPUs with Modern Workloads ↗ esologic.com

Introduction

In the ever-evolving world of technology, decommissioned NVIDIA enterprise GPUs, often deemed as "e-waste," can still hold significant value. With cards like the K80 boasting 24GB of GDDR5 selling for just $60, it's worth exploring if these GPUs can still be effectively utilized with modern workloads. This article delves into the results of a nearly year-long benchmarking project aimed at assessing the efficiency of 15 such GPUs in a contemporary environment.

Capabilities of EOL GPUs

Tesla series GPUs, such as the K80, P100, and V100, were initially designed for intensive workloads in data centers. However, with their software updates discontinued, they are often sidelined from new applications. Despite this, they still offer impressive performance for certain uses. For instance, the Tesla K80, though older, can still provide decent performance in massive data processing due to its large VRAM capacity.

Hardware Under Test

The tests involved various models such as the Tesla K80, P100, and V100, each having distinct technical specifications and performance metrics. The K80, for example, features dual Kepler GPUs with a total of 4,992 CUDA cores and 24GB of memory capacity. The P100, on the other hand, uses the Pascal architecture and offers up to 16GB of memory. Finally, the V100, based on the Volta architecture, provides up to 16GB of HBM2 memory and 5,120 CUDA cores.

Performance in Real-World Scenarios

The tests were conducted by simulating modern workloads including tasks like machine learning, graphics rendering, and video processing. Results revealed that while these GPUs cannot compete with the latest models in terms of raw performance, they can still be effectively used for specific tasks.

  • Machine Learning: Models like the P100 and V100 proved effective for machine learning, particularly for frameworks like TensorFlow and PyTorch, thanks to their parallel computing capabilities.
  • Graphics Rendering: The V100 particularly stood out in rendering applications due to its fast HBM2 memory, making it suitable for rendering tasks that require fast and abundant memory.
  • Video Processing: The GPUs also demonstrated their utility in video processing, enabling effective and fast transcodings compared to a CPU alone.

Economic and Ecological Considerations

Utilizing these older GPUs offers not only an economic opportunity but also a more ecological approach by reusing existing hardware. Purchasing these cards at low prices can significantly cut costs for small labs or personal projects requiring GPU compute power.

Conclusion

Ultimately, while Tesla "e-waste" GPUs are not the top performers in today's market, their affordable cost and specific capabilities make them valuable in certain contexts. For those looking to maximize their tech resources without breaking the bank, these older GPUs can represent an excellent investment.

Let's discuss your project in 15 minutes.

e-waste GPUs NVIDIA Tesla benchmarking GPU performance modern workloads
Deepthix newsletter · 100% AI · every Monday 8am

An AI agent reads tech for you.

Our AI agent scans ~200 sources per week and ships the best articles to your inbox Monday 8am. Free. One click to unsubscribe.

Visit the newsletter page →

Want to automate your operations?

Let's talk about your project in 15 minutes.

Book a call