Blackwell Architecture · 8,960 CUDA Cores · 280 Tensor Cores (5th Gen) · 432 GB/s · PCIe 5.0 x8
RTX PRO 4000 Blackwell SFF Edition
NVIDIA Blackwell (GB203)
8,960
280 (5th Gen)
FP4 precision support
70 (4th Gen)
DLSS 4 Multi Frame Gen
24.05 TFLOPS
24 GB GDDR7 ECC
192-bit
432 GB/s
18 Gbps effective
48 MB
PCIe 5.0 x8
(Physical x16)
Low-Profile Dual-Slot
167×69×40 mm
70W
No aux power connector
4× Mini DisplayPort 2.1b
TSMC 4N (5nm)
45.6B transistors
9th Gen NVENC
6th Gen NVDEC
August 11, 2025
The NVIDIA RTX PRO 4000 Blackwell SFF Edition redefines what's possible in a compact, low-profile GPU. Packing 24 GB of ultra-fast GDDR7 ECC memory, 8,960 CUDA cores, and 280 fifth-generation Tensor Cores into a 70W dual-slot card that fits in Small Form Factor workstations, it delivers full-size Blackwell performance where space is at a premium. No auxiliary power connector is needed — all 70W is drawn from the PCIe slot.
Built on the GB203 GPU (the same silicon family as the GeForce RTX 5080) with TSMC's 4N process, the card achieves 24.05 TFLOPS of FP32 throughput and 432 GB/s of memory bandwidth on a 192-bit bus. The 5th-gen Tensor Cores add native FP4 precision support for efficient LLM inference, while 4th-gen RT Cores with DLSS 4 Multi Frame Generation accelerate photorealistic rendering. The 9th-gen NVENC encoder supports 4:2:2 H.264/HEVC for professional video workflows.
The RTX PRO 4000 SFF is the ideal GPU for AI developers and data scientists working in compact desktop workstations who need 24 GB of GPU memory for model prototyping, fine-tuning, and local inference. QS Compute supplies these cards with NVIDIA enterprise warranty across the APAC region.
Need NVIDIA RTX PRO 4000 Blackwell SFF?
Contact QS Compute for pricing and availability.
Request Quote