# Fireworks is powered by AMD

> Fireworks is a proud partner of AMD. Our inference and training products are often powered by AMD Instinct GPUs in addition to the NVIDIA Blackwell and Hopper hardware. Fireworks is collaborating with AMD to provide streamlined access to production-ready LLM inference, powered by AMD Instinct™ GPUs. Through the AMD AI Developer Program, developers can now access Fireworks credits, enabling immediate experimentation with state-of-the-art models, without the need to manage infrastructure.

## Frontier Inference Powered by AMD GPUs

AI developers looking to get started quickly often face challenges like access to the hardware, inference compute, or dev ops expertise. These can slow down experimentation and delay time to value.

Fireworks is collaborating with AMD to provide streamlined access to production-ready LLM inference, powered by **AMD Instinct™ GPUs**. Through the **AMD AI Developer Program,** developers can now access **$50 of Fireworks credits**, enabling immediate experimentation with state-of-the-art models, without the need to manage infrastructure.

## $50 in Fireworks Credits for AMD Developers

AMD + Fireworks

As part of the AMD AI Developer Program, eligible developers can receive:

   - **$50 in Fireworks credits**
   - **90-day validity from issuance**

These credits provide immediate access to the Fireworks inference platform, allowing developers to evaluate performance and build applications without upfront cost.

[Join the AMD AI Developer Program](https://www.amd.com/en/developer/ai-dev-program.html)

## How to Get Started

The onboarding process is designed to be simple and easy:  

1. Join the [AMD AI Developer Program](https://www.amd.com/en/developer/ai-dev-program.html).

2. Fill out the cloud credit form and select Fireworks credit.

3. Approval: AMD will review the credit submission within 2 business days to validate eligibility.

4\. Receive Access Details. Approved developers will receive:

   - A credit code via email (valid for 90 days)
   - A link to create a Fireworks account

5\. Activate Credits and Build.

## Frontier Inference + Frontier Hardware

AMD Instinct™ MI350 Series GPUs set a new standard for Generative AI and high performance computing (HPC). Built on AMD's new cutting-edge 4th Gen CDNA™ architecture, these GPUs deliver exceptional efficiency and performance for training large >300B models, and powering the production inference workloads that AI natives and enterprise AI leaders require.

- **AMD Instinct™ MI350X**: The MI350X is AMD's workhorse data center GPU for serious AI and high performance computing (HPC) workloads. Built on 4th Gen CDNA™ architecture, it delivers 288GB HBM3E memory and 8TB/s bandwidth alongside expanded MXFP6 and MXFP4 datatype support. Built explicitly for efficient inference at scale, it compares against NVIDIA's B200 platform with up to 2.05x FP6 performance, making it a strong option for teams looking to diversify beyond CUDA. Paired with the open-source ROCm stack, it's built for developers who want performance without ecosystem lock-in. · [Learn more about AMD Instinct MI350X GPUs](https://www.amd.com/en/products/accelerators/instinct/mi350/mi350x.html)
- **AMD Instinct™ MI355X**: The MI355X is AMD's flagship accelerated computing chip. It delivers up to 4x peak theoretical performance over the MI300X and up to 2.8x faster training on MLPerf 5.1 workloads. Against NVIDIA's GB200, the MI355X delivers up to 2x FP6 performance advantage, with inference throughput gains that make it particularly compelling for large-model serving. For teams pushing frontier model training or running high-concurrency inference on large models like Kimi or Qwen, the MI355X is AMD's answer to the top of the stack. · [Learn more about AMD Instinct MI355X GPUs](https://www.amd.com/en/products/accelerators/instinct/mi350/mi355x.html)

## What Developers Can Build

With Fireworks running on AMD Instinct GPUs, developers can quickly build and deploy:

   - Conversational AI and chat applications
   - Intelligent document processing and summarization
   - Code generation and developer tools
   - Retrieval-augmented generation (RAG) systems
   - Real-time inference services
