Rent NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs

Rent an RTX PRO 6000 Server Edition on Akash

As of October 20, 2026, on-demand RTX PRO 6000 Server Edition GPUs on Akash start at $2.02 per GPU-hour.

Built for data center inference

• Passively cooled for 24/7 operation in standard servers, with air- and liquid-cooled versions.
• Faster than L40S: NVIDIA cites up to 5× higher LLM inference throughput for agentic AI and 3.3× faster text-to-video generation versus the L40S.
• FP4 Tensor Cores: 4 PFLOPS of FP4 and 2 PFLOPS of FP8.
• Confidential computing: NVIDIA describes it as the first universal GPU to enable secure AI with NVIDIA Confidential Computing.
• Multi-tenant: MIG splits one card into up to 4 isolated GPUs of 24 GB, and NVIDIA vGPU profiles range from 8 GB to 96 GB.

Is the RTX PRO 6000 Server Edition the right GPU for you?

Choose the Server Edition when you need a large-memory inference GPU that also handles graphics and video.

• Choose the RTX PRO 6000 Server Edition

  if you serve models up to 96 GB, run video or rendering pipelines, or need isolated multi-tenant GPUs.

• Choose the H100 or H200

  if you train across many GPUs and need NVLink and HBM bandwidth. See H100 pricing →

• Choose the Workstation Edition

  if you want the highest single-card bandwidth (1,792 GB/s). See Workstation Edition →

What is the RTX PRO 6000 Server Edition used for?

The NVIDIA RTX PRO 6000 Blackwell Server Edition is used for AI inference, video generation, rendering and virtual workstations in the data center. NVIDIA benchmarks against the L40S show:

• Agentic AI and LLM inference

  Up to 5× higher LLM inference throughput than L40S, with an optimized NIM profile for gpt-oss-120b.

• Multimodal and video generation

  3.3× faster text-to-video generation than L40S.

• Genomics and drug discovery

  Nearly 7× faster genomics sequencing and up to 4.8× faster OpenFold2 inference than L40S.

• Physical AI and digital twins

  Robotics simulation and industrial digital twins on Omniverse; supported hardware for Cosmos world-model microservices.

• Rendering and recommender systems

  Over 2× faster rendering and nearly 2× faster recommender inference than L40S.

• Virtual workstations and secure multi-tenant AI

  VGPU with up to 48 concurrent VMs per GPU, MIG isolation and confidential computing for finance, healthcare and government.

Not ideal for: large multi-node training. Choose H100, H200 or B200 for NVLink-connected training clusters.

Deploy an RTX PRO 6000 Server Edition in three steps

1. 1

   Add your app. Pick a template or paste a Docker image.

2. 2

   Choose RTX PRO 6000 Server Edition. Turn on GPU, select it and check current provider capacity.

3. 3

   Deploy. Request quotes, accept one and start your workload.

NVIDIA RTX PRO 6000 Blackwell Server Edition specifications

NVIDIA does not state whether the Server Edition tensor figures are sparse or dense; they are listed as peak.

NVIDIA RTX PRO 6000 Server Edition FAQ