← All stories
● Covered by 1 source · 1 reportMedium impact1 neutral

AMD Highlights Native FP64 on Instinct MI430X, Contrasting Nvidia's Emulation

🔄 Updated 4d ago
New to BrevFeed? We gather this story from every outlet covering it into one summary — ranked by real-world impact, not just the latest headline — so you never miss what matters. What is BrevFeed? →

Key points

  • AMD promotes native FP64 support on Instinct MI430X at SC26.
  • Nvidia uses emulation (e.g., Ozaki schemes) for FP64 on its GPUs.
  • AMD claims emulation is only effective for dense GEMM operations.
  • Most HPC workloads require native FP64 vector performance.

AMD's Focus on Native FP64 at SC26

During Supercomputing 2026 (SC26), AMD highlighted the native FP64 (double-precision floating-point) capabilities of its Instinct MI430X accelerator. This emphasis targets customers in high-performance scientific computing and simulation, such as national laboratories and supercomputing centers, where FP64 is a critical data type.

Nvidia's Approach to FP64

Nvidia has reallocated die area on its data center GPUs to prioritize lower-precision matrix math, which benefits AI inference workloads. To address FP64 requirements, Nvidia employs emulation techniques, such as Ozaki schemes, rather than native hardware support. This strategic shift reflects the growth in the AI compute market compared to the relatively stable demand for double-precision math in supercomputing.

Limitations of Emulation for HPC Workloads

AMD contends that Nvidia's FP64 emulation is not a suitable replacement for native double-precision processing across the majority of HPC workloads. The company states that emulation methods like Ozaki schemes primarily assume dense general matrix multiply (GEMM) operations. However, AMD's profiling indicates that less than 10% of HPC workloads involve dense GEMM operations, making emulation less effective for the broader spectrum of scientific computations.

Workloads Requiring Native FP64 Vector Performance

For many critical HPC applications, including sparse iterative linear algebra, stencils, hydrodynamics, high-order finite element analysis, Monte Carlo simulations, particle transport, and classical molecular dynamics, AMD asserts that strong native FP64 vector performance is essential for optimal results. These workloads do not map efficiently to GEMM-centric emulation.

Additionally, even for workloads that align with GEMM operations, AMD notes that emulation performance can be sensitive to matrix sizes and operand scaling. The MI430X's native FP64 vector math capabilities are not affected by these operand sizing issues, offering consistent performance.

✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →

The daily brief

One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.

One email a day. Unsubscribe in one click, any time.

Today's brief

Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.

~4 min · 3 stories · Nov 21

▶ Play today's brief Listen on Spotify

New every morning, and the back catalogue is archived by date.

Reporting from

At SC26, AMD emphasized the native FP64 double-precision support in its Instinct MI430X accelerator, positioning it against Nvidia's FP64 emulation methods. AMD argues that emulation is insufficient for most high-performance computing (HPC) workloads, which often require strong FP64 vector performance beyond dense matrix multiplies. This matters for scientific computing and simulation fields that rely on precise double-precision math.