NVIDIA has a new GPU for small form factor workstations, the new NVIDIA RTX 2000 Ada. This generation adds newer Ada Lovelace generation workstation graphics to a $625 MSRP point. This is fairly exciting for those who want a low profile GPU for compact workstations.
New NVIDIA RTX 2000 Ada 16GB SFF GPU Launched
In the new dual slot, low profile card, we get a number of improvements, one of the big ones is a bigger memory footprint at 16GB of GDDR6 memory, with ECC. The memory interface on this one is only 128-bit with 224GB/s as the rated memory bandwidth.
The new card has 2816 CUDA cores, 88 Tensor cores, and 22 RT cores. Here are NVIDIA’s compute performance claims:
- Single-Precision Performance1 12.0 TFLOPS
- RT Core Performance1 27.7 TFLOPS
- Tensor Performance1 191.9 TFLOPS
The big features are really some of the more physical ones. We get four mDP 1.4a display outputs, important for small workstations that need to support multi-monitors. The 2.7″ tall and 6.6″ long dual slot GPU is rated at only 70W. It also uses only a PCIe Gen4 x8 connection instead of a higher-end x16 connection making it more useful for adding into systems with limited I/O and space.
On the encode/ decode side, the card features a single NVENC / NVDEC, and in this generation supports AV1 encode & decode. That is a feature many will want to see in newer cards.
Final Words
If you are doing high-end design work or AI, then the NVIDIA RTX 2000 Ada is probably not the GPU for you. On the other hand, with Ada Lovelace generation cores, and 16GB of memory, along with a 70W TDP and a small footprint, this is the type of GPU that can bring AI inference to many locations that higher-end GPUs simply cannot. The primary use case, of course, for this card is as a four display output workstation GPU, but many workstation applications have been adding AI inferencing to speed workflows, so that is becoming more important.
Wonder if we’ll get:
NVIDIA RTX 2000 Ada Ti
NVIDIA RTX 2000 Ada OC
NVIDIA RTX 2000 Ada Super
NVIDIA RTX 2000 Ada Super OC
NVIDIA RTX 2000 Ada Super Ti
NVIDIA RTX 2000 Ada Super Ti OC
NVIDIA RTX 2000 Ada Super Ti OC Extreme
NVIDIA RTX 2000 Ada Super Ti OC Extreme RGB
Nvidia does not do silly suffix versions for professional card, but they will introduce number in the middle version if they figured out how to cut a GPU die in a different way. Like RTX 2500 2600 2700 2880 etc.
I’m wondering why this 70W (total board)GPU has to take dual slots? A single-slot half-height cooler should be enough for 70W in my opinion.
Dual slot because either or both: A) the chip is hot B) you can’t fit it in a rack chassis
I wonder when they’ll just move to USB-C for these type of cards and ditch mDP. Bonus if this hypothetical card can do power delivery when an optional PCIe aux or SATA power connector is plugged into the card.