Apr 13, 2021 11:10:00

NVIDIA announces 8 types including new Ampere-based GPU 'RTX A5000'

At the annual conference event 'GTC 2021 ' held by NVIDIA from April 12, 2021, we announced eight GPUs with Ampere architecture for next-generation notebook PCs / desktops / servers.

Nvidia unveils 8 new RTX GPUs for workstations and laptops | VentureBeat
https://venturebeat.com/2021/04/12/nvidia-unveils-8-new-rtx-gpus-for-workstations/

Nvidia's A100 Gets Baby Brothers: The A10 and A30 | Tom's Hardware
https://www.tomshardware.com/news/nvidia-a10-a30-announced

Nvidia Expands RTX Studio GPUs With New Ampere Cards | Digital Trends
https://www.digitaltrends.com/computing/nvidia-rtx-studio-ampere-desktop-mobile-gtc-2021/

The newly announced GPUs are ' NVIDIA RTX A5000 ' and ' A4000 ' for desktops, 'RTX A5000 Laptop GPU', 'A4000', 'A3000' and 'A2000' for notebook PCs, and 'NVIDIA A16 GPU' and 'A2000' for data centers. All 8 types of 'NVIDIA A10 GPU'. All of these models are equipped with Ampere architecture, achieving up to twice the throughput of the previous generation, and the '2nd generation RT core' that can perform ray tracing, shading, and noise removal tasks at the same time, and compared to the previous generation. '3rd generation Tensor core' that realizes up to 2 times the throughput and up to 10 times the sparseness, and newly supports the TF32 / BFloat16 data format, FP32 throughput up to 2.5 times compared to the previous generation, graphics and computing It has a 'CUDA core' that greatly improves the ing workload.

As for the GPU for desktop, 'RTX A5000' is equipped with 24GB, 'A4000' is equipped with 16GB GDDR6 memory by default, and 'A5000' can also increase the memory to 48GB by connecting two with NVIDIA NVLink. .. The A5000 also supports NVIDIA RTX vWS software that can run multiple high-performance virtual workstation instances, allowing remote users to run high-end design, AI, and compute workloads. PCIe is Gen 4 with twice the bandwidth of its predecessor.

The specifications of 'RTX A5000' and 'A4000' for desktop are as follows.

	A5000	A4000
memory	ECC compatible 24GB GDDR6	ECC compatible 24GB GDDR6
Video output	DisplayPort 1.4 x 4	DisplayPort 1.4 x 4
Graphic path	PCI Express Gen 4x16	PCI Express Gen 4x16
Form factor	11.2cm (H) x 26.7cm (L) dual slot	11.2cm (H) x 24.1cm (L) single slot
Maximum power consumption	230W	140W
NVLink	2-way low profile (2-slot / 3-slot bridge)	×
Virtual GPU software support	NVIDIA Virtual PC, Virtual Applications, NVIDIA RTX vWS, NVIDIA Virtual Compute Server	×
VR compatible	○	○

For notebook PCs, it is equipped with 3rd generation Max-Q technology that has achieved further noise reduction and efficiency, and has a maximum GPU memory of 16GB. The 'RTX A5000 Laptop GPU' is equipped with 6144 CUDA cores, 192 Tensor cores, and 48 RT cores, and consumes 80 to 165W. The 'A4000' has 5120 CUDA cores, 160 Tensor cores, and 40 RT cores. The 'A3000' has 4096 CUDA cores, 128 Tensor cores, and 32 RT cores. The 'A2000' has 2560 CUDA cores, 80 Tensor cores, and 20 RT cores, and consumes 35 to 95 W.

In addition, for notebook PCs, the 'NVIDIA T1200 Laptop GPU' and 'NVIDIA T600' based on the previous generation Turing architecture are also newly introduced. The two are designed for professional workflows with multiple applications, with significant performance and functionality upgrades from integrated graphics.

For data centers, the specifications are specialized for designers and engineers, and the 'A16' has achieved up to twice the user density even though the total cost is lower than the previous generation. Achieve an enhanced virtual desktop infrastructure experience. The A10 offers up to 2.5 times more virtual workstation performance than its predecessor. The A16 and A10 combine with NVIDIA RTX virtual workstations and NVIDIA virtual PC software to further improve performance, memory, and speed, and streamline all workflows such as graphics, AI, and VDI. Users can spend a lot of time on efficient and productive work.