NVIDIA a40 GPU enables an evolutionary leap in data center performance and multi workload capabilities. It combines excellent professional graphics performance with powerful computing and AI acceleration capabilities to meet today’s design, creative and scientific challenges. NVIDIA a40 can drive a new generation of virtual workstations and server based workloads, and provide professionals with advanced functions in the fields of ray tracing rendering, simulation, virtual production and so on anytime, anywhere.
NVIDIA ampere architecture CUDA ® core
Double speed single precision floating point (fp32) arithmetic processing and improved energy efficiency can significantly improve the performance of graphics and simulation workflows, such as complex 3D computer aided design (CAD) and Computer Aided Engineering (CAE).
Second generation RT core
The throughput of the second generation RT core is twice that of the previous generation, and can run ray tracing and shading or noise reduction functions at the same time, which can significantly speed up the running speed of workload, such as realistic rendering of film content, architectural design evaluation and virtual prototyping of product design. This technology can also accelerate the rendering of dynamic blurred images with ray tracing effect, so as to obtain visual presentation faster and more accurately.
The third generation tensor core
The new tensor float 32 (tf32) precision provides five times the training throughput of the previous generation, and can accelerate the training of AI and data science models without changing the code. In terms of hardware, the sparsity of structure is supported to double the reasoning throughput. Tensor core also introduces AI into graphics processing through DLSS, AI noise reduction and other functions, and enhances the editing function of specific applications.
48 GB GPU video memory
Ultra high speed gddr6 video memory can be expanded to up to 96 GB through nvlink, providing data scientists, engineers and creative professionals with the high-capacity video memory they need to deal with large data sets and workloads such as data science and simulation.
Third generation NVIDIA nvlink ®
Up to two a40 GPUs can be connected, expanding GPU video memory from 48 GB to 96 GB. The higher GPU to GPU interconnection bandwidth provides one-piece scalable video memory, which can accelerate graphics and computing workload and handle larger data sets. The new and more compact nvlink connector enables interconnection in more types of servers.
The new generation of improvements brought by NVIDIA virtual GPU (vgpu) software can provide larger and more powerful virtual workstation instances for remote users, so as to support high-end remote design, AI and computing workloads.
PCI Express 4th generation
PCI Express generation 4 doubles the bandwidth of PCIe generation 3, thus improving the speed of data transmission from CPU memory and better supporting data intensive tasks such as AI, data science and 3D design. Faster PCIe performance can also accelerate GPU direct memory access (DMA) transmission, which supports gpudirect in GPU and ® For video devices provide faster video data I / O communication speed, which brings a powerful live broadcast solution. A40 is backward compatible with PCI Express generation 3, which provides deployment flexibility.
Data center efficiency and security
NVIDIA a40 adopts a dual slot, energy-efficient design, which improves the energy efficiency by 2 times compared with the previous generation, which has been verified by various NVIDIA certification systems produced by OEMs around the world. NVIDIA a40 also provides secure and measurable boot through the hardware trusted root function to ensure that the firmware is not tampered with or damaged.