NVIDIA RTX 4000 “Small Form Factor” ADA GPU tested
The latest workstation GPU based on Ada Lovelace architecture has now been tested by Japanese outlet Jisakuhibi.
The RTX 4000 SFF is a mid-range workstation GPU designed as a small form factor video card. By far the most important feature of this card is no requirements for any auxiliary power connector, as this card is designed to operate within the 70W TDP range, which exclusively provided through the PCIe slot.
NVIDIA RTX 4000 SFF comes with two I/O brackets, either low profile or a standard full height bracket that fits any standard motherboard. The card is equipped with four Mini DisplayPort 1.4 connectors, and it is shipped with at least one adapter for full DP cable. This tiny Ada GPU weighs only 322g, which is almost half of what the dual-fan RTX 3060 Ti and RTX 4070 GPUs used for the same test weighed.
The card is equipped with a AD104 GPU featuring 6144 CUDA cores, which is a better configuration than RTX 4070 (5888 cores), but not as good as RTX 4070 Ti (7680).
More importantly, we now have a confirmation on the RTX 4000 clocks, which were not shared by NVIDIA yet. The card officially ships with a 720 MHz base clock and 1560 MHz boost clock, which is definitely not a high GPU clock. Despite its size, this card features 20GB of GDDR6 memory which runs at 14 Gbps. This gives a bandwidth of 280 GB/s, on par with many mid-range gaming solutions.
The size of the card and the fact that it features 8th Gen NVIDIA NVEnc encoding, means that the card can encode in AV1 format, an upgrade from the Ampere series which were only capable of AV1 hardware accelerated decoding. Furthermore, the media outlet confirms that just like all RTX workstation GPUs, the card is not limited to 5 concurrent encoding sessions like gaming series either.
Chances are this GPU will be used as a companion GPU for certain workloads, or to extend the capabilities of existing workstations. Some users may even be considering using this card for their home theater systems, as this is the smallest Ada GPU yet.
The reviewer posed the following summary:
- GPU compatible with full HD/high frame rate and WQHD PC gaming
- An average of 20% faster than the RTX 3060
- An average of 70% faster than the RTX 3050 (≒RTX A2000)
- Supports AI double speed interpolation DLSS 3, the latest feature of GeForce RTX 40
- Supports hardware encoding of next-generation codec AV1 with high compression and high image quality
- Unlimited number of simultaneous encoding sessions
- Supports dedicated features for professional GPUs such as NVIDIA Mosaic
- Compact size with total length of 169mm and occupied 2 slots
- Compatible with low profile PCIE bracket
- No need for PCIE auxiliary power with TGP70W
- GPU cooler that can sufficiently cool the TGP70W with a noise level of 35dB or less
- Video output is Mini DisplayPort only (usually one DP conversion adapter is included)
- Very expensive at 220,000 yen including tax (as of May 2023)
The reviewer determined that the RTX 4000 SFF ADA is around 20% faster than RTX 3060 gaming GPU, while doing does not require any external power connectors while it remains within the TGP envelope. It has strong encoding capabilities and unlike the GeForce RTX 40 series, it is not limited to concurrent sessions. By far the biggest issue is the price, as the card has an MSRP of $1,250.
Source: jisakhuhibi via Wccftech