NVIDIA has unveiled two new Blackwell-architecture graphics playing cards, the NVIDIA RTX PRO 4000 Blackwell SFF Version and RTX PRO 2000 Blackwell — each concentrating on the acceleration of machine studying and synthetic intelligence (ML and AI) workloads on smaller kind issue techniques — and in a way more affordable 70W energy envelope.
“Functions have gotten more and more AI accelerated, and extra customers want AI efficiency, irrespective of the dimensions or form of their workstation,” claims NVIDIA’s Stacy Ozorio in assist of the corporate’s newest launch. “The RTX PRO 4000 SFF and RTX PRO 2000 function fourth-generation RT [Ray Tracing] Cores and fifth-generation Tensor Cores with decrease energy in half the dimensions of a standard GPU. The brand new GPUs are designed to convey next-generation efficiency to a variety {of professional} workflows, offering unbelievable speedups for engineering, design, content material creation, AI and 3D visualization.”
NVIDIA has introduced two new small kind issue GPUs, the RTX PRO 4000 Blackwell SFF (high) and RTX PRO 2000 Blackwell (above). (📷: NVIDIA)
The low-profile playing cards, designed to fit into small kind issue techniques the place full-size add-in boards are too tall, relies on NVIDIA’s Blackwell structure — the launch of which gave the corporate no small quantity of hassle owing to a since-rectified design flaw that lowered the variety of working chips manufacturing associate Taiwan Semiconductor (TSMC) might get from every silicon wafer.
Each playing cards should not solely smaller than earlier Blackwell boards, however significantly much less energy hungry with NVIDIA claiming a 70W most energy consumption even beneath load — but, the corporate says, the RTX PRO 4000 Blackwell SFF delivers “as much as” 2.5 instances the AI efficiency, 1.7 instances the ray-tracing efficiency, and 1.5 instances the bandwidth of its predecessor. The RTX PRO 2000 Blackwell, in the meantime, is claimed to supply large beneficial properties for on-device generative AI workloads with 1.4 instances the efficiency for picture era and a couple of.3 instances for textual content era in comparison with its last-generation equal.
The playing cards are each primarily based on NVIDIA’s newest Blackwell structure, which has had a bumpy street to market. (📷: NVIDIA)
In precise specification phrases, that boils right down to: 8.960 CUDA cores, fifth-generation Tensor and fourth-generation RT cores, two ninth-gen NVENC encoders and two sixth-gen NVDEC decoders, 24GB of GDDR7 RAM with error correcting code (ECC) on a 192-bit interface with 432GB/s bandwidth, and a PCI Specific Gen. 5 eight-lane connection to the host for the RTX PRO 4000 Blackwell SFF; and 4,352 CUDA cores with fifth-generator Tensor and fourth-generation RT cores, one ninth-gen NVENC encoder and one sixth-gen NVDEC decoder, 16GB of GDDR7 RAM with ECC on a 128-bit interface with 288GB/s bandwidth, and the identical PCIe Gen. 5 eight-lane connection for the RTX PRO 2000 Blackwell. NVIDIA claims an “efficient FP4 AI TOPS” of 545 tera-operations per second (TOPS) for sparse work on the RTX PRO 2000 Blackwell, however has not launched an equal determine for the RTX PRO 4000 Blackwell SFF.
Extra info on the brand new add-in boards is offered on the NVIDIA web site on the RTX PRO 4000 Blackwell SFF and RTX PRO 2000 Blackwell product pages; the corporate has confirmed that each will launch “later this 12 months” from companions together with PNY, TD SYNNEX, BOXX, Dell, HP, and Lenovo, although had not but publicly disclosed pricing on the time of writing.