AMD Instinct™ MI350 Series GPUs

Setting New Standards

Since their inception, AMD Instinct™ accelerators have provided data center customers and those looking to lean into the possibilities of AI with performance, efficiency, and scalability. Each generation has set new standards, delivered industry-leading specifications, and helped to optimize performance and reduce TCO.¹

Now, with the introduction of AMD Instinct™ MI350 Series GPUs to market as part of the recent  Advancing AI event, AMD is raising the expectation once again.

AI Driven, HPC Optimized, Leadership Performance

Now’s the time to introduce your customers to the new AMD Instinct™ MI350X and AMD Instinct™ MI355X GPUs and their respective platforms, each built on the cutting-edge 4th Gen AMD CDNA™ architecture and boasting up to 288GB HBM3E memory capacity and 8TB/s of bandwidth. Designed for everything from massive AI model training and high-speed inference to complex HPC workloads, AMD Instinct MI350X GPUs deliver up to 2.05X the FP6 performance of Nvidia’s B200 platform,² while AMD Instinct MI355X GPUs boast a 2X FP6 advantage over the GB200,³ setting a new bar for density, efficiency, and throughput at scale.

As customer and infrastructure demands rise, AMD has designed these accelerators to keep pace. These new air-cooled GPUs integrate seamlessly with previous-generation AMD Instinct™ MI300X and MI325X infrastructure, making them a pain-free, cost-effective upgrade for virtually any scenario where higher density computing is a necessity.

Both AMD Instinct GPUs offer expanded support for FP6 and FP4 datatypes in addition to enhanced support for FP16 and FP8 processing, delivering uncompromising computational throughput and memory bandwidth utilization while maximizing energy efficiency. AMD Instinct MI350 Series GPUs also offer up to 4X better performance running FP4 compared to AMD Instinct MI300X GPUs running FP8,⁴ positioning them to deliver incredible performance when it comes to advanced generative AI models and pushing the boundaries of the space further than ever.

Integrated With Next-Generation AMD ROCm™ Software

Building on the AMD commitment to open-source innovation, AMD Instinct MI350 Series GPUs are integrated with the next-generation AMD ROCm™ software stack, the industry’s premier open alternative for AI and HPC workloads.

Paired with the launch of these new accelerators, the latest AMD ROCm software enhancements take AI workloads to the next level, further optimizing AI inferencing, training, and framework compatibility, delivering high throughput, low latency results for demanding workloads such as natural language processing (NLP), computer vision, and more.

ROCm software delivers Day-0 support for AI platforms and models provided by leaders such as OpenAI, Meta, PyTorch, Hugging Face, xAI, DeepSeek, and more, thanks to strategic and deep collaborations with key partners. All of this adds up to ensuring AMD Instinct GPUs are optimized to run the latest AI models and frameworks as they launch, enabling both developers and businesses to accelerate how they integrate AI into their workflows.

There’s a reason industry titans like Microsoft and Meta trust AMD Instinct GPUs to power large-scale AI deployments for models like Llama 405B and GPT. Speak to your AMD representative or visit amd.com to learn more and help enable your customers with the power of AMD Instinct accelerators.