NVIDIA Unveils Six Ada Lovelace-based Professional GPUs At GTC 2023

NVIDIA’s latest (and still virtual) GTC has just kicked-off, and without wasting a single second, the company’s unveiled lots to talk about, with large focuses on proviz, healthcare, automotive, robotics, and lithography. For this post, we’re going to focus on what gets us (and much of our audience) the most excited: professional visualization.

Since NVIDIA released its first Ada Lovelace-based GPUs last fall, there haven’t been too many models released overall. On the gaming side, there have been three – RTX 4070 Ti, RTX 4080, and RTX 4090 – while for professional, there’s only been the NVIDIA RTX 6000 “Ada Generation”. That changes with this latest GTC, as a new Ada pro card hits the desktop, and five land in laptops.

With NVIDIA having only launched the high-end RTX 6000 Ada option for the desktop so far, we would have guessed that this GTC would bring about the RTX 5000 series – but not so. Instead, we get a 4000-class card that caters to the SFF workstation crowd. Its name? RTX 4000 SFF. At least it can’t be confused with the four-year-old Quadro RTX 4000!

NVIDIA specs its RTX 4000 SFF (6,144 cores) at 19.2 TFLOPS for its single-precision operation, which is almost identical to the Ampere-based (and full-size) A4000. If that seems underwhelming, you should know about a couple of tricks the RTX 4000 SFF has up its sleeves. For starters, it gains 4GB over the last-gen A4000, settling in at 20GB (yes, it’s ECC-capable). Second, this RTX 4000 SFF requires half the power of the A4000. Its 70W TDP means the card doesn’t even have a power connector. For its stature and power requirement, this GPU seems potent.

Every small form-factor GPU targets a specific niche, but with this RTX 4000 SFF, NVIDIA has talked lots about its benefits for working with video, either for playback (with Quadro Sync II capabilities) or editing. While the rendering performance of the RTX 4000 SFF will be modest in comparison to the other released Ada Lovelace GPUs, video work will rely on the built-in encoder a lot more. Ultimately, an SFF GPU with 20GB of ECC memory and robust encoders (including AV1) sounds tasty. Those who want the same GPU but in a regular form-factor will be able to make the change with an included bracket.

On the mobile side of the fence, NVIDIA’s five new Ada Lovelace GPUs cater to most price-points. Those five include the RTX 2000 (14.5 TFLOPS), RTX 3000 (19.9), RTX 3500 (23.0), RTX 4000 (33.6), and RTX 5000 (42.6). The RTX 5000 offers similar FP32 performance as the desktop RTX 4070 Ti, which is an impressive amount of performance in a notebook.

ECC seems to be creeping up in more places lately, which is not just great to see, but some would say overdue. The top four of these Ada workstation GPUs include ECC support, with only the RTX 2000 forgoing it. While not everyone needs or wants ECC, it’s nice to have it more readily available to more customers who do want to take advantage of it.

If you want something even lower-end than the RTX 2000 mobile GPU, NVIDIA’s filling out the bottom with last-gen Ampere models: A1000 6GB and A500 4GB. All but the A500 support VR and variable-rate shading. Every GPU is ready to take advantage of NVIDIA’s numerous tools, including Omniverse, Canvas, and Broadcast.

While announced last month, NVIDIA wanted to re-share information of its latest top-end workstations that include its RTX 6000 GPUs. Soon, the likes of BOXX, Dell, HP, and Lenovo will offer new workstations with up to 4x RTX 6000 GPUs and 2x 56-core Intel Xeon W CPUs:

We’ve covered every generational GPU launch NVIDIA’s had for nearly two decades, and without fail, each new one makes us stand back and appreciate the kind of performance that can be had in a single workstation. A solution like this one won’t be cheap, but considering that a single RTX 6000 is much faster than the previous generation top-end, having four of them in a single rig is downright jaw-dropping. These workstations are such beasts, that 25GbE networking appears to be standard!