NVIDIA has introduced two new graphics cards to its Ampere workstation lineup: the RTX A4500 20 GB and the RTX A2000 12 GB. Both cards are focused on distinct workloads and have the most recent Ampere architecture for speedier content production, AI, and scientific use cases.
The NVIDIA RTX Ampere Workstation graphics card range currently includes six configurations from which consumers may choose. The two new entrants compete in the high-end and entry-level sectors, respectively.
NVIDIA RTX A4500 20 GB Specifications
Starting with the NVIDIA RTX A4500, the graphics card features 7168 CUDA Cores in 56 SMs enabled on the GA102 GPU. Due to the sheer core count, the Ampere GPU offered in the RTX A4500 is the most compact workstation variant of the GA102 GPU (with just 66 percent active cores). There are also 224 Tensor units, 56 RT cores, and a core speed of up to 1.63 GHz. The card includes 20 GB of GDDR6 memory that runs at pin speeds of up to 16 Gbps through a 320-bit bus interface, providing the GPU with 640 GB/s of bandwidth. The card’s TDP is set at 200W.
In terms of performance, the NVIDIA RTX A4500 can achieve up to 23.7 TFLOPs FP32, 46.2 TFLOPs RT, and 182.2 TFLOPs Tensor. The card has four DisplayPort 1.4 outputs and is available in a dual-slot form with an active cooling solution. Power is supplied by a single 8-pin connection. On the A4500, NVIDIA has an NVLINK connector that allows up to two cards to function in parallel with a total RAM of 40 GB.
NVIDIA RTX A2000 12 GB Specifications
NVIDIA is also providing an enhanced model of their RTX A2000 graphics card with double the memory capacity, in addition to the higher-end alternative. The remainder of the specifications is the same as in the original model.
The Ampere GPU architecture is used for the NVIDIA RTX A2000 and the card is powered by the GA106 GPU, which includes 3328 CUDA cores, 104 Tensor cores, and 26 RT cores, all of which provide a significant performance boost over prior generation products. The GPU has a performance rating of 8 TFLOPs FP32, 15.6 TFLOPs RT, and 63.9 TFLOPs Tensor. The memory is a 192-bit bus architecture, with the chips working at 12 Gbps pin speeds for a total bandwidth of 288 GB/s.
The RTX A2000 itself has a tiny form factor, with a half-height and half-length board. On the shroud of the card, there’s even a little blower fan. There are no power ports on this 70W TDP card. It’s a simple plug-and-play card with high efficiency in a small package. There are four Mini DisplayPorts (1.4) at the rear panel IO shroud, which also has a tiny vent to vent hot air.
NVIDIA Ampere Workstation Graphics Cards:
| Graphics Card | NVIDIA RTX A2000 | NVIDIA RTX A4000 | NVIDIA RTX A4500 | NVIDIA RTX A5000 | NVIDIA RTX A5500 | NVIDIA RTX A6000 |
| GPU | Ampere GA106 GPU | Ampere GA104 GPU | Ampere GPU (TBD) | Ampere GA102 GPU | Ampere GA102 GPU | Ampere GA102 GPU |
| GPU Process | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm | Samsung 8nm |
| Die Size | 276 mm² | 392.5 mm² | TBD | 628mm² | TBD | 628mm² |
| GPU Cores | 3328 | 6144 | 7168 | 8192 | TBD | 10752 |
| Tensor Cores | 104 | 192 | 224 | 256 | TBD | 336 |
| Boost Clock | ~1200 MHz? | 1536 MHz | TBD | 1697 MHz | TBD | 1455 MHz |
| Single Precision | 8.0 TFLOPs | 19.2 TFLOPs | TBD | 27.8 TFLOPs | TBD | 31.2 TFLOPs |
| VRAM | 6 GB GDDR6 | 16 GB GDDR6 | 20 GB GDDR6 | 24 GB GDDR6 | 32 GB GDDR6? | 48 GB GDDR6 |
| NVLINK VRAM | N/A | N/A | N/A | 48 GB GDDR6 | 64 GB GDDR6? | 96 GB GDDR6 |
| Memory Bus | 192-bit | 256-bit | 320-bit | 384-bit | 384-bit | 384-bit |
| Memory Bandwidth | 228 GB/s | 448 GB/s | TBD | 768 GB/s | TBD | 768 GB/s |
| TDP | 70W | 140W | ~200W | 230W | ~250W | 300W |
| Launch Price | TBC | TBC | TBC | TBC | TBC | $4650 US |
| Launch Date | October. 2021 | April, 2021 | Q4 2021? | April, 2021 | Q4 2021? | December, 2020 |
