Transformer training shootout, part 2: AWS Trainium vs. NVIDIA V100 @juliensimonfr
Transformer training shootout, part 2: AWS Trainium vs. NVIDIA V100  @juliensimonfr
Uploaded May 2023 | Updated September 2026, 2 weeks ago
In this video, I compare the cost/performance of AWS Trainium with the NVIDIA V100 GPU.

I first launch a trn1.32xlarge instance (16 Trainium chips) and a p3dn.24xlarge (8 V100s). Then, I run 3 benchmarks: language pretraining with GPT2, token classification with BERT Large, and image classification with the Vision Transformer

The results? Trainium is 2 to 5x faster, and 3 to 8x cheaper!

⭐️⭐️⭐️ Don't forget to subscribe to be notified of future videos ⭐️⭐️⭐️

- Amazon EC2 trn1: aws.amazon.com/ec2/instance-types/trn1
- Amazon EC2 p3: aws.amazon.com/ec2/instance-types/p3
- Training commands: gist.github.com/juliensimon/da64fc6d6a2fe39bd8c5af12389a227e
- Trainium with Optimum Neuron: youtu.be/FmjTWags__Q
- Trn1 vs G5 benchmark: youtu.be/2SquGhkld7k
Transformer training shootout, part 2: AWS Trainium vs. NVIDIA V100The Truth About MCP: Pros, Cons & Real-World Use CasesRun SLMs locally: Llama.cpp vs. MLX with 10B and 32B Arcee modelsEnterprise AI with the Hugging Face Enterprise HubRouting function calling queries to the best SLM/LLM with Arcee ConductorUnlocking Hidden Treasures: Your Data is More Valuable Than You Think! 💎Is AI the answer to every problem? 🤔 Think again!Retail AI at the edge - Cisco Live 2025 interviewArcee Orchestra - Build an Agentic Code Review WorkflowHow Witty Works leverages Hugging Face to scale inclusive languageSLM Inference on a Windows laptop 🤯 Intel Lunar Lake CPU/GPU/NPU + OpenVINOFine-tune Stable Diffusion with LoRA for as low as $1
Julien Simon |

Transformer training shootout, part 2: AWS Trainium vs. NVIDIA V100

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER