NVIDIA “clarifies” U.S. export controls
The U.S. Department of Commerce recently updated its high-tech export control requirements, imposing new restrictions on products exceeding 70 TFLOPS from being exported to China under ECCN 4A0030.C. However, NVIDIA has clarified that the GeForce RTX 4090 D and H20 GPUs will not be affected by these limitations.
The ECCN 4A0030.C provisions, which include “adjusted peak performance” (APP) and “weighted teraflops” (WT), only apply to systems with memory consistency and not individual GPUs. NVIDIA explained that the chip must be able to access any memory location in hardware through cache or memory addressing to comply with the requirements, excluding GPUs from this provision.
Furthermore, the “adjusted peak performance” restriction only applies to “weighted” FP64 throughput, which is scaled based on whether it is a vector or scalar processor. This calculation is used to determine the total FP64 throughput of the system and multiply by 0.9 for vector processors or 0.3 for non-vector processors.
The GeForce RTX 4090 D, with minimal FP64 throughput of 1.15 TFLOPS, is not subject to the restrictions due to its low performance in this aspect. NVIDIA reassured that their products comply with the regulations and are not impacted by the new export controls.
It is important for companies to understand the intricacies of export controls to ensure compliance and avoid any unintended limitations on their products. NVIDIA’s clarification brings relief to consumers and investors, ensuring continued access to their high-performance GPUs in the global market.
Source: [Editorial Department/News Center]