Nvidia presented the RTX A6000 and A40 , new professional video cards for workstations and servers based on the Ampere architecture. Both solutions are based on a GA102 GPU with 48 GB GDDR6 . These are professional solutions designed for engineers, designers and scientists, proposals that until the last generation Nvidia would have been called Quadro, but Nvidia removed this branding.
Technical specifications
The new Nvidia RTX A6000 will be available from partners PNY, Leadtek, Ingram Micro, Ryoyo and on nvidia.com around mid-December, then will arrive on workstations from BOXX, Dell, HP and Lenovo. As for the technical specifications, the A6000 replaces the Turing Quadro RTX 8000/6000 solutions as it combines a GA102 GPU (the same as the GeForce RTX 3090 and 3080 graphics cards) with 10752 CUDA cores, 336 Tensor cores and 84 cores. RT. It is accompanied by 48 GB of GDDR6 memory at 16 Gbps on a 384-bit bus .
Nvidia RTX A6000 Features
According to Nvidia, the new RTX A6000 offers approximately double the performance of the Quadro RTX 8000 in certain situations, particularly on operations leveraging more CUDA and RT cores as well as higher processing throughput. For the moment, the American company has not disclosed the frequencies and power in “flops” of the newcomer, but we are talking about a TGP of 300W (50W less than the RTX 3090). The GPU should therefore operate at lower frequencies.
The new RTX A6000 obviously benefits from all the new features of the Ampere architecture, such as a greater number of data types supported by the Tensor cores (to name only the BFloat16), but also the new decoder for the AV1 codec and the PCI Express 4.0 interface . There is also an NVLink connector for connecting two A6000s in parallel, as well as partial ECC support, via software. The RTX A6000 uses 3D Vision Pro and is equipped with four DisplayPort 1.4.
Nvidia A40 Features
As for the Nvidia A40 , it is a very similar solution to the RTX A6000, except that it is passive. The target customer is different. It is intended for high-density servers. The other difference compared to the A6000 concerns the memory, set at 14.5 Gbps , while the TGP remains equal to 300W. It presents video outputs: a new feature which allows the card to control a screen.
Nvidia justified the presence of video ports on these cards generally absent on previous versions following requests from users in the multimedia and broadcast industry. The Nvidia A40 will be available starting next year on servers from Cisco, Dell, Fujitsu, Hewlett Packard Enterprise and Lenovo. These cards are connected to an 8-pin EPS connector, not a regular PCIe.
