As predicted before the festival, in the recent GTC 2020 (GPU Technology Conference) keynote speech, NVIDIA officially unveiled the 8nm process Ampere (ampere) architecture of rtX A6000, RTX A40 two professional graphics cards and computing cards, respectively, for workstations and server products.
NVIDIA says the performance of the new RTX Professional Card is excellent, with a twice as much performance as ray-traced scenes, providing faster, more realistic scenes and movie creation for filmmakers. Even vision tool manufacturers, such as Luxion, have tripled the performance of the RTX AA6000 from 34.7 FPS (Quadro RTX A6000) to 88.9 FPS in their KeyShot visualization tools.
In terms of positioning, the RTX A6000, RTX A40 can be divided into successors to the Quadro RTX 8000 and Tesla T4. It’s worth noting that this time NVIDIA didn’t use the Quadro brand to name it.
NVIDIA says that because RTX can represent everything in future graphical computing, it has abandoned the usual Quadro and Tesla product lines.
Specifications, the RTX A6000, RTX A40 and GTX 3080 (8704 CUDA), 3090 (10496 CUDA) the same GA102 GPU, upgraded to a “full blood” version, with 84 sets of SM, containing 10752 CUDA cores, more than RTX 3090 more than 256.
The RTX A6000 is a workstation graphics card, while the A40 is for server use, so technically the A40 can be considered a successor to the Tesla T4.
The difference is that the A40 will support the display output, while the Tesla server card was not. Both the RTX A6000 and A40 are equipped with four DisplayPort 1.4 interfaces.
Both new cards are equipped with 48GB GDDR6 memory, but the specifications vary, with the A6000’s memory frequency and bandwidth of 16Gbps, 768 GB/s, while the A40 is relatively conservative, at 14.5Gbps and 696GB/s.
And both new cards support the third generation NV Link, can be connected to two cards synchronization operations, memory capacity can be expanded to 96GB for dual GPU to share.
Third generation NVLINK connector.
In terms of power consumption, the TDP of the RTX A6000 and RTX A40 is 300W, using the new EPS 12V 8-pin power interface instead of the traditional PCIe 8-pin or RTX 30 Series 12-pin.
In addition, although both are dual-slot radiators, the RTX A6000 uses a single active turbofan to dissipate heat, while the RTX A40 uses passive cooling.
NVIDIA said the RTX A6000 would go on sale in mid-December, while the A40 would go on sale early next year at an unannoled price.