NVIDIA launched enhanced double-precision mathematical algorithms within the cuBLAS library under CUDA Toolkit 13.4.2 to accelerate scientific workloads. The system leverages the Ozaki-II algorithm and the 2M method to minimize hardware operations while executing exact computations via GPU integer Tensor Cores. Memory-parsimonious tiling is incorporated into the pipeline to actively curb workspace memory overhead by discarding residue pairs after computation. Testing demonstrates significant end-to-end acceleration across modern hardware configurations and production codes like BerkeleyGW.
Key points
Existing scientific software stacks can access these acceleration features automatically…: Existing scientific software stacks can access these acceleration features automatically without requiring any application-level code modifications.
Innovative workspace management techniques such as memory-parsimonious tiling prevent alg…: Innovative workspace management techniques such as memory-parsimonious tiling prevent algorithmic memory requirements from hindering computation.
Practical research applications benefit immediately, with the BerkeleyGW materials scienc…: Practical research applications benefit immediately, with the BerkeleyGW materials science code observing speedups reaching up to 6.6 times.
Sources checked by our analysis pipeline: 1
Many scientific simulations, from molecular chemistry to materials science, spend the majority of their execution time computing high-precision dense general matrix multiplications (GEMM). These operations are called DGEMM for real 64-bit floating-point numbers and ZGEMM for complex 64-bit floating-point numbers. cuBLAS is widely used by HPC and AI applications to accelerate GEMM and other Basic Linear Algebra Subprograms (BLAS) on NVIDIA GPUs. Rapid advances in lower-precision hardware have motivated exploration of a new class of algorithms that deliver full double-precision results on the faster integer Tensor Cores.
With CUDA Toolkit 13.4.2, cuBLAS no longer has to choose between accuracy and speed. cuBLAS breaks each double-precision number into integer pieces, multiplies them on the GPU’s Tensor Cores, and reassembles the exact result. The answers keep full double-precision accuracy, but the work runs on faster hardware.
Two new algorithms make this more efficient: Ozaki-II needs fewer integer multiplications than the earlier Ozaki-I approach, and the 2M method cuts a complex matrix multiplication from four real multiplications to two. This builds on the Automatic Dynamic Precision (ADP) framework introduced in Unlocking Tensor Core…
Federal regulators halted departures as a safety precaution after overloaded legacy telecommunications equipment disrupted communications between air traffic controllers and airborne pilots. The widespread network collapse left around 1.5 million travelers stranded across the country. Officials are actively investigating the cascade failure but have not yet provided a concrete schedule for full restoration.
Swiss researchers are trialing predictive artificial intelligence tools to give vulnerable communities advance notice ahead of deadly landslides. Meanwhile, scientific teams are mobilizing ocean-monitoring hardware, from ice-measuring submersibles to global buoy arrays, alongside ongoing assessments of the entertainment industry. These cross-sector developments highlight how specialized automation and data collection continue to reshape environmental protection and commercial media.
Technology users worldwide experienced significant disruptions following a major technical issue that impacted core services. The company addressed the incident by confirming that external security threats were not the cause of the widespread service interruption. Leadership has formally acknowledged the problem while attempting to reassure affected clients about the nature of the breakdown.