AMD's Instinct MI430X GPU will be the fastest FP64 chip, delivering up to 6x the performance of Rubin in classic HPC workloads.
AMD Delivers The Biggest Leap In FP64 Compute With MI430X, 6x Rubin Performance & Deployment In The Discovery System at ORNL By 2028
The AI segment continues to push up compute FLOPS in the exascale range through low-precision formats such as FP4, FP6, and FP8. These formats lead the AI spectrum, and while they are important for neural networks, higher-precision formats such as FP64 still hold a lot of value in High-Performance Computing workloads.
AMD has always been a leader in terms of FP64 FLOPs, and their next-generation MI430X has now been previewed, offering up to 200 TFLOPs of raw FP64 performance, the highest ever for any HPC GPU. This phenomenal improvement comes from AMD's advanced CDNA architecture, which is made using high-tech process and packaging technologies and couples loads of HBM4 memory together.
The AMD Instinct MI430X GPUs will be part of the MI400 series, which spans multiple products. The line also includes the MI450X, which is AMD's primary AI accelerator.
AMD also previewed the performance of its MI430X GPU, offering up to 6x the compute capability of its upcoming rival, NVIDIA Rubin. NVIDIA Rubin GPUs offer a peak of 33 TFLOPs of FP64 vector compute and up to 200 TFLOPs using Tensor-Core based emulation algorithms. Natively, that's 200 TFLOPs for AMD's MI430X and 33 TFLOPs for NVIDIA's Rubin GPUs.
The following is the breakdown for the GPUs and their respective FP32/FP64 capabilities:
| Feature | Hopper GPU | Blackwell GPU | Rubin GPU | MI430X GPU |
| FP32 vector (TFLOPS) | 67 | 80 | 130 | TBD |
| FP32 matrix (TFLOPS) | 67 | 227* | 400* | TBD |
| FP64 vector (TFLOPS) | 34 | 40 | 33 | 200 |
| FP64 matrix (TFLOPS) | 67 | 150* | 200* | TBD |
But it doesn't stop at just FP64; AMD also confirms that the MI430X accelerator offers leadership low-precision AI capabilities alongside its superb HPC performance, all in a single package. This shows that HPC is still relevant, and for that, AMD is announcing two major projects that will deploy the MI430X.
The first one is the Discovery Supercomputer at Oak Ridge National Laboratory in the USA. The system is planned for deployment in 2028 and will expand America's leadership in AI and HPC. The system will also be used by the DOE as its flagship product, driving breakthroughs in the fields of energy, biology, national security, advanced materials, and manufacturing innovation. The system will be made using many AMD Instinct MI430X and AMD EPYC CPUs.
In Europe, the AMD MI430X accelerators will be deployed alongside next-generation EPYC CPUs in the Alice Recoque system, which aims to become Europe's leading exascale-class supercomputer.
About the author: A Software Engineer by training and a PC enthusiast by passion, Hassan Mujtaba serves as Wccftech's Senior Editor for hardware section. With years of experience in the industry, he specializes in deep-dive technical analysis of next-generation CPU and GPU architectures, motherboards, and cooling solutions. His work involves not only breaking news on upcoming technologies but also extensive hands-on reviews and benchmarking.
Follow Wccftech on Google to get more of our news coverage in your feeds.
