NVIDIA has spear two new entries in its Ampere workstation line, the RTX A4500 20 GB and RTX A2000 12 GB graphics cards. Both cards are intended for a different set of workloads and feature the latest Ampere architecture for a faster content, AI, and scientific use cases.
NVIDIA RTX A4500 20GB and RTX A2000 12GB Ampere graphics cards launched
The NVIDIA RTX Ampere Workstation line of graphics cards now consists of six options for users to choose from. The two new entrants are positioned in the high-end and entry-level segments.
NVIDIA RTX A4500 20 GB Specifications
Starting with the NVIDIA RTX A4500, the graphics card scales 7168 CUDA cores in a total of 56 SMs that are enabled on the GA102 GPU. The number of cores makes the Ampere GPU featured on the RTX A4500 the smallest workstation configuration of the GA102 GPU (with only 66% active cores). Other specs include 224 Tensor units, 56 RT cores, and a base clock of up to 1.63 GHz. The card has 20 GB of GDDR6 memory that operates over a 320-bit bus interface at spindle speeds of up to 16 Gbps, providing the GPU with 640 Gbps of bandwidth. The card’s TDP is rated at 200W.
When it comes to performance, the NVIDIA RTX A4500 offers up to 23.7 TFLOP FP32, 46.2 TFLOP RT and 182.2 TFLOP of Tensor performance. The DisplayPort 1.4 4-output toggle card and comes in a dual-slot configuration with an active cooling solution, power is supplied through a single 8-pin connector. NVIDIA also has an NVLINK interconnect on the A4500 which allows up to two cards to run in parallel with 40 GB combined memory.
NVIDIA RTX A2000 12 GB Specifications
In addition to the high-end option, NVIDIA also offers an improved variant of its RTX A2000 graphics card which now offers twice the memory capacity. The rest of the specifications remain the same as the original variant.
The NVIDIA RTX A2000 features the Ampere GPU architecture. In terms of specs, the RTX A2000 rocks the GA106 GPU which is equipped with 3,328 CUDA cores, 104 Tensor cores, and 26 RT cores, all of which offer a nice performance improvement over previous generation offerings. GPU performance is rated at 8 TFLOP FP32, 15.6 TFLOP RT and 63.9 TFLOP Tensor. The memory is a 192-bit bus design with the chips operating at 12 Gbps spindle speeds for a total effective bandwidth of 288 Gbps.
The RTX A2000 itself is compact in design, with a half-height and half-length board. The board even has a small blower-type fan on the fairing. As this is a 70W TDP card, there are no power connectors. It is a simple plug-and-use board that offers high efficiency in a compact design. There are four Mini DisplayPorts (1.4) near the rear panel IO fairing which also comes with a small vent to exhaust hot air.
NVIDIA Ampere Workstation graphics cards:
|Graphic card||NVIDIA RTX A2000||NVIDIA RTX A4000||NVIDIA RTX A4500||NVIDIA RTX A5000||NVIDIA RTX A5500||NVIDIA RTX A6000|
|GPU||AMP GA106 GPU||AMP GA104 GPU||GPU ampere (TBD)||GPU Ampere GA102||GPU Ampere GA102||GPU Ampere GA102|
|GPU process||Samsung 8 nm||Samsung 8 nm||Samsung 8 nm||Samsung 8 nm||Samsung 8 nm||Samsung 8 nm|
|Die size||276 mm²||392.5 mm²||To be determined||628 mm²||To be determined||628 mm²|
|GPU cores||3328||6144||7168||8192||To be determined||10752|
|Tensor nuclei||104||192||224||256||To be determined||336|
|Boost the clock||~ 1200MHz?||1536 MHz||To be determined||1697 MHz||To be determined||1455 MHz|
|Simple precision||8.0 TFLOP||19.2 TFLOP||To be determined||27.8 TFLOP||To be determined||31.2 TFLOP|
|VRAM||6 GB of GDDR6||16 GB of GDDR6||20 GB of GDDR6||24 GB of GDDR6||32 GB of GDDR6?||48 GB of GDDR6|
|VRAM NVLINK||N / A||N / A||N / A||48 GB of GDDR6||64 GB of GDDR6?||96 GB of GDDR6|
|Memory bus||192 bits||256 bits||320 bits||384 bits||384 bits||384 bits|
|Memory bandwidth||228 GB / s||448 GB / s||To be determined||768 GB / s||To be determined||768 GB / s|
|TDP||70W||140W||~ 200W||230W||~ 250W||300W|
|Introductory price||To confirm||To confirm||To confirm||To confirm||To confirm||$ 4650 US|
|Release date||October. 2021||April 2021||Q4 2021?||April 2021||Q4 2021?||december 2020|