Nvidia unveiled new GPUs, but no, it’s not the RTX 4000 series. Instead the company is introducing a number of workstation GPUs – one for the desktop and several for professional laptops. The Nvidia RTX A5500 is based on the GA102 GPU with 10,240 CUDA cores. This is the same configuration as an RTX 3080 Ti. However, this GPU uses ECC memory (Error Correction Code), which can detect and correct errors. This doesn’t matter so much for games, but it is crucial for professional tasks – simulations, 3D design, machine learning and so on - where errors can lead to incorrect results.
The A5500 has a whopping 24 GB GDDR6 (with
ECC) and that can be expanded to 48 GB using Nvidia NVLink to hook up a second
GPU. The A5500 offers double the performance for ray tracing compared to the
previous generation and can be up to 12x as fast for machine learning tasks. A
single card can be used by multiple remote users thanks to virtualization.
Nvidia also unveiled a laptop version of
the RTX A5500. This one is based on the GA103 chip (same as the laptop RTX 3080
Ti) and has 7,424 CUDA cores, nearly 30% less than its desktop counterpart.
Correspondingly, single-precision (FP32) performance falls from 34.1 to 24.7
TFLOPS. Also, this GPU comes with 16 GB of GDDR6 VRAM of the non-ECC variety.
There’s no word on pricing yet, but the
desktop RTX A5500 is available today from select Nvidia partners and will be
featured in pre-built systems in Q2. Workstation GPUs aren’t cheap even in
better times, however, so if your budget can’t stretch to an A5500, you still
have some options. The Nvidia RTX A500 may be the smallest Ampere GPU for
laptops yet with just 2,048 CUDA cores and 4 GB of VRAM (for comparison, a 3050
has 2,560 cores). There are also the A1000, A2000 and A3000 with 4 GB, 8 GB and
12 GB of VRAM respectively.
The mid-range is the RTX 4500, which is
based on a GA104 with 5,888 CUDA cores (this parallels the RTX 3070). This one
is equipped with 16 GB of RAM, the same as the pricier A5500.
