Ampere is introduced in a small form factor desktop discrete graphics card solution with the NVIDIA RTX A2000
Because the NVIDIA RTX A2000 for notebooks uses the GA106 GPU, the desktop card is expected to have a similar configuration but faster clock speeds and a greater memory capacity. The A4000 notebook and desktop variants have radically different specifications, with the notebook GPU having 8GB of GDDR6 memory and the desktop GPU having 16GB of GDDR6 memory.
If the NVIDIA RTX A2000 follows suit, the GPU will be constructed around a GA106 SKU with up to 12 GB of GDDR6 memory running on a 192-bit bus interface, with a more conservative alternative giving a scaled-down 128-bit bus interface with 8 GB GDDR6 VRAM.
The NVIDIA RTX A2000 features a full deck in a low-profile (half-length) dual-slot form factor, which is the most intriguing part of the graphics card. To startup, the card has a small fan on the cover and appears to have a single 6-pin connector. Near the rear panel I/O cover, there are four Mini DisplayPorts (1.4) and a small vent to blow out hot air. The card also looks to have a backplate, though comprehensive photographs are necessary to corroborate this.
The RTX A2000 will battle mostly in the entry-level workstation market, with many compact form factor graphics cards. AMD provides several low-profile solutions. However, they are only available on Polaris machines. Higher-end RDNA variants have full-length, single-slot configurations and are not sold alongside the RTX A2000.
- Architecture – Ampere, 8 nm
- Pipelines – 2560
- Memory Clock Speed – 12000MHz
- Bus Width – 128 Bit
- Memory Type – GDDR6
- Max Allowed Memory – 4096 MB
- DirectX Version – DirectX 12
Raw performance should be comparable to the GeForce RTX 3050 Ti for laptops at the same TGP level. For good performance, both GPUs rely substantially on good cooling and a high TGP. The RTX A2000 should be noticeably faster than the old Quadro T2000 and T2000 Max-Q at a similar power consumption level.
The GA107 chip contains 2.560 FP32 ALUs, half of which can also execute INT32 instructions (i.e., 1,280 INT32 ALUs). Turing permitted all shaders to continue to use FP32 or INT32 instructions. Nvidia claims that the chip’s raytracing and tensor cores have also been improved. The Ampere processors additionally have an updated 5th generation video encoder (NVENC for H.264 and H.265) and a 7th generation decoder (for various formats now, including AV1).
Samsung manufactures the GA107 chip at an 8nm (8N) node, which is not quite as fast as TSMC’s 7nm node (e.g., used by AMD) and the more advanced GA100 Ampere chip). The A2000 can also be used in thin and light laptops, depending on the TGP (with 35 Watt TGP, e.g.).