NVIDIA “clarifies” U.S. export controls, GeForce RTX 4090 D will not be subject to this limitation
In a recent update to high-tech export control requirements by the U.S. Department of Commerce, new restrictions were added to ECCN 4A0030.C, prohibiting the export of products exceeding 70 TFLOPS to China. However, NVIDIA has clarified that their GeForce RTX 4090 D and H20 GPUs will not be affected by these new restrictions.
According to a report by Reuters, the ECCN 4A0030.C provisions include limitations on “adjusted peak performance” and “weighted teraflops”, which only apply to systems with memory consistency rather than individual GPUs. NVIDIA explained that the chip needs to be able to access any memory location in the system in hardware through cache or memory addressing to be subject to the provision.
Furthermore, the “adjusted peak performance” provision specifically applies to the weighted FP64 throughput, which is scaled based on whether it is done through vector units or scalar processors. By calculating the total FP64 throughput of a system and applying a multiplier, the determination can be made if the product exceeds the 70 weighted teraflops limit.
In the case of the GeForce RTX 4090 D, which has a FP64 throughput of 1.15 TFLOPS, it does not surpass the limit and therefore is not subject to the export restrictions. This clarification from NVIDIA ensures that their latest GPU will not be impacted by the recent changes in U.S. export controls.
This clarification from NVIDIA provides assurance to consumers and partners that their products will continue to be available for export without restrictions.