Finally, Nvidia has completed its series of graphics cards for professional use, with the announcement of the RTX A5000, A4000, A3000 and A2000 graphics cards.
The GTC 2021 left several headlines and among them, the announcement of the RTX A5000 and RTX A4000 graphics cards stands out, although they also took the opportunity to announce the specifications of the RTX A3000 and A2000 models.
The new RTX A5000 and A4000 graphics cards feature new RT cores, Tensor cores and CUDA cores that allow accelerated AI, graphics and rendering in real-time.
The RTX A5000 model uses the GA102 GPU with a custom 8nm node from Samsung. The chip has about 6144 CUDA cores and a power of calculation of 21.7 TFLOPs. The card has 16 GB of GDDR6 memory with ECC technology and a bandwidth of 448 GB / sec. The A4000, A3000 and A2000 models do not have the NVLink function.
RTX A5000 also has another advantage over the RTX A4000, and that is that the first has the support of virtualization through the Nvidia RTX vWS tool. With those two features alone, the A5000 looks like a much more compelling option than the A4000.
The RTX A4000 uses the GA104 GPU with 5120 CUDA cores and up to 17.8 TFLOPs of power in single precision. The graphics card has 16 GB of GDDR6 memory and up to 384 GB / s of bandwidth. The TDP of both is 230W for the RTX A5000 and 140W for the RTX A4000. Below we have the full specification table.
Specifications table
| RTX A2000 | RTX A3000 | RTX A4000 | RTX A5000 | |
|---|---|---|---|---|
| GPU | GA106 | GA104 | GA104 | GA102 |
| Transistors | 12 billion | 17.4 billion | 17.4 billion | 28 billion |
| Node | 8 nm | 8 nm | 8 nm | 8 nm |
| Chip Tam. | 276 mm² | 392.5 mm² | 392.5 mm² | 628.4 mm² |
| FP32 ALUs | 2,560 | 4,096 | 5,120 | 6,144 |
| INT32 ALUs | 1,280 | 2,048 | 2,560 | 3,072 |
| SMs | 20th | 32 | 40 | 48 |
| Tensor | 80 | 128 | 160 | 192 |
| RT | 20th | 32 | 40 | 48 |
| FP32 performance | 9.3 TFLOPS | 12.8 TFLOPS | 17.8 TFLOPS | 21.7 TFLOPS |
| RT performance | 18.2 TFLOPS | 25.0 TFLOPS | 34.8 TFLOPS | 75.6 TFLOPS |
| Tensor performance | 74.7 TFLOPS | 102.2 TFLOPS | 142.5 TFLOPS | 174.0 TFLOPS |
| Ability | 4GB | 6 GB | 8 GB | 16 GB |
| Storage type | GDDR6 | GDDR6 | GDDR6 | GDDR6 |
| Memory interface | 128 bit | 192 bits | 256 bit | 256 bit |
| Memory bandwidth | 192 GB / s | 264 GB / s | 384 GB / s | 448 GB / s |
| TDP | 35 - 95 W. | 60 - 130 W. | 80 - 140 W. | 80 - 165 W. |
0 Comments