NVIDIA RTX PRO™ 6000 Blackwell Server Edition GPU

NVIDIA RTX PRO™ 6000 Blackwell Server Edition

Powered by the innovative NVIDIA Blackwell architecture, the NVIDIA RTX PRO™ 6000 Blackwell Server Edition integrates cutting-edge AI processing power with exceptional visual computing performance to dramatically accelerate complex workloads in enterprise data centers. Featuring 96GB of ultra-fast GDDR7 memory, the NVIDIA RTX PRO 6000 Blackwell delivers unparalleled performance and flexibility for a wide range of use cases, from agent AI, physical AI, and scientific computing to rendering, 3D graphics, and video processing.

Product Specifications

	NVIDIA RTX PRO™ 6000 Blackwell Server Edition	NVIDIA L40S
CUDA Parallel Processing cores	24,064	18,176
NVIDIA Tensor Cores	752	568
NVIDIA RT Cores	188	142
Frame buffer memory	96 GB GDDR7 with ECC	48GB GDDR6 with ECC
Memory Bandwidth	1597 GB/s	864 GB/s
Max Power Consumption	600W	350W
Graphics Bus	PCI Express 5.0 x16	PCI Express 4.0 x16
Display Connectors	DisplayPort 2.1 x4	DisplayPort 1.4a x4
Form factor	4.4” H x 10.5” L, FHFL Dual Slot	4.4” H x 10.5” L,dual slot
External power supply	1x PCIe CEM5 16-pin	1x PCIe CEM5 16-pin

You can expect performance improvements not only in graphics performance but also in learning and inference.

Servers equipped with the NVIDIA RTX PRO 6000 Blackwell Server Edition can be offered through OEM vendors Supermicro and DELL Technologies.

We can also guide you to a server that has been verified, so if you are interested, please contact us from the following.

Comparison of RTX PRO 6000 and the previous generation L40S

Compared to the previous generation L40S, it can deliver several times the performance, including a large-scale language model.

Integration with NVIDIA NIM

By using NVIDIA NIM, the Llama model and other models offer up to approximately 5 times the throughput of the previous generation, and with fine tuning, it can be up to approximately 2 times faster.

RTX PRO 6000 MIG function

The RTX PRO 6000 series, with its MIG (Multi-Input/Multi-GPU) functionality, can be physically divided into up to four sections (1/4 MIG: 24GB). Furthermore, by utilizing the Time Slice vGPU function, while normally one GPU is occupied by one container, it is possible to schedule 12 containers per 1/4 MIG, enabling scheduling of up to 48 containers.