NVIDIA

Cloud GPU · verified 2026-09-05

NVIDIA RTX PRO 6000

Blackwell96 GB

96 GB of GDDR7 with ECC on Blackwell, the workstation that clouds now rent like a datacenter card. Not HBM. Not NVLink 900. Often the honest hour when 80 GB Hopper is overkill and 32 GB 5090 is a toy.

VRAM96 GB
From /GPU/hr$0.66
ArchitectureBlackwell
VendorNVIDIA

In brief

What to know before you rent

Catalog listings

RTX PRO 6000 Pricing and Availability

Published on-demand rates first. Spot and waitlists are labeled. Confirm the live rate before you provision.

ProvidersRegionBilling typeInterconnect/GPU/hr
Packet.aiIn stockUS-WestOn-demandPCIe$0.66Open provider
Vast.aiIn stockUS-EastOn-demandPCIe$0.72Open provider
RunPodIn stockUS-EastOn-demandPCIe$0.79Open provider
KoyebIn stockEU-WestOn-demandPCIe$0.89Open provider
LambdaLimitedUS-WestOn-demandPCIe$0.99Open provider

Our reading

Who this GPU is for

I rent RTX PRO 6000 when I want 96 GB without Hopper tax and I can live without NVLink fabric. Inference, LoRA, creative + AI mixed workloads. I do not rent it to train a 70B with AllReduce across 8 GPUs as if it were HGX.

Rent it if Single-GPU 70B-class inference at moderate context, fine-tunes that want 96 GB, shops that needed ECC and got tired of 5090s dying mid-job.

Skip it if Multi-GPU NVLink training. Jobs that are bandwidth-bound on HBM. Anyone who thought this was RTX 6000 Ada 48 GB.

How we compare hours

Datasheet

Technical specifications

NVIDIA RTX PRO 6000 Blackwell · Per GPU · Workstation / Server

Compute

FP32
125 TFLOPSWorkstation edition; server SKUs can be lower
AI TOPS
Up to 4,000NVIDIA slide number, precision-dependent
CUDA cores
24,064
Tensor cores
7525th generation
RT cores
1884th generation
Precision support
FP4, FP8, FP16, BF16, TF32, FP32, INT8

Memory

Capacity
96 GB GDDR7ECC
Bandwidth
1,792 GB/sServer edition closer to 1,597 GB/s
Bus width
512-bit
ECC
Yes

Silicon

Architecture
Blackwell
Die
GB202 class

Fabric and host

GPU interconnect
None at NVLink 4/5 classPCIe peer-to-peer only
Host interface
PCIe 5.0 x16

Power

Board power
600 WMax-Q 300 W; server configurable

Platform

Partitioning
MIG up to 4×24 GB or 2×48 GB

Workstation, Max-Q, and Server editions share the 96 GB story and not the TDP. Clouds usually rent the server board. If the hour is suspiciously cheap, check whether you got a 300 W Max-Q.

Trade-offs

Pros and cons of RTX PRO 6000

What this GPU does well, and when the hour is the wrong buy.

Strengths

  • 96 GB ECC is the rare workstation number that actually changes which models fit
  • Blackwell FP4 on a PCIe board you can actually find
  • MIG slices for sharing a card without a second hour
  • Usually cheaper than H100 when you do not need HBM or NVLink

Limits

  • GDDR7 is not HBM3. Bandwidth is ~1.8 TB/s, not 3.35+
  • No NVLink 900 GB/s. Multi-GPU is PCIe plus hope
  • 600 W workstation vs 300 W Max-Q vs server boards, same name, different hour
  • Easy to confuse with Ada RTX 6000 48 GB. That is a different decade of VRAM

Not a cut-price H100

96 GB vs 80 GB looks like a win in a spreadsheet. Memory type is the plot. HBM wins bandwidth-bound training. GDDR7 wins “I need the bytes and a PCIe slot.” I pick this card for fit and ECC, not for AllReduce.

If your trainer is NCCL across 8 GPUs, you wanted HGX. If your trainer is one model, one card, 96 GB, welcome.

Not RTX 6000 Ada

Ada RTX 6000 is 48 GB. This page is 96 GB Blackwell. Catalogs still mix the strings. If VRAM says 48, you are on the wrong generation and the wrong review.

How I actually rent it

  1. 01

    Read the edition

    600 W vs 300 W vs server. Power is performance on this silicon.

  2. 02

    Decide if you needed NVLink

    If yes, leave. If no, 96 GB ECC is the pitch.

  3. 03

    Compare to H100 and 5090

    H100 if bandwidth/fabric. 5090 if 32 GB is enough and you accept consumer risk.

FAQ

RTX PRO 6000 questions

Who has the cheapest RTX PRO 6000?+

As of September 5, 2026, the cheapest verified in-stock on-demand RTX PRO 6000 tracked by GPUBeacon is $0.66 per GPU per hour from Packet.ai.

How much does the RTX PRO 6000 cost per month?+

As of September 5, 2026, at 720 hours per month, one RTX PRO 6000 costs an estimated $569 at the median on-demand price in this sample. Cheapest verified in stock: $475 per month on-demand.

Where can I rent an RTX PRO 6000?+

GPUBeacon currently tracks RTX PRO 6000 configs from 5 providers. The cheapest verified in-stock on-demand configs come from Packet.ai, Vast.ai, RunPod and Koyeb. See the full price comparison above for every provider and config.

Are RTX PRO 6000 prices going up or down?+

GPUBeacon publishes a catalog sample, not a 90-day price index. As of September 5, 2026, the median on-demand RTX PRO 6000 in this sample is $0.79 per GPU per hour. The cheapest verified in-stock hour is $0.66.

Why choose the RTX PRO 6000?+

Single-GPU 70B-class inference at moderate context, fine-tunes that want 96 GB, shops that needed ECC and got tired of 5090s dying mid-job.

When is the RTX PRO 6000 not a good fit?+

Multi-GPU NVLink training. Jobs that are bandwidth-bound on HBM. Anyone who thought this was RTX 6000 Ada 48 GB.

Can RTX PRO 6000 replace an H100?+

For single-GPU inference and many fine-tunes, sometimes. For NVLink training and HBM-bound steps, no. Compare the job, not the VRAM integer.

Is this the same as RTX 5090 with more memory?+

Same architecture family, different product: ECC, MIG, workstation/server boards, 96 GB. 5090 is 32 GB consumer. Do not mix reliability stories.

Keep comparing

Alternatives to NVIDIA RTX PRO 6000

Same job, different envelope. Compare VRAM, bandwidth and the hour before you lock a badge.

NVIDIA H100

80 GB HBM3 · 87% more bandwidth than the RTX PRO 6000

From$1.73/ GPU / hr

Compare

NVIDIA RTX 5090

32 GB GDDR7 · similar bandwidth to the RTX PRO 6000

From$0.34/ GPU / hr

Compare

NVIDIA L40S

48 GB GDDR6 · 52% less bandwidth than the RTX PRO 6000

From$0.79/ GPU / hr

Read the verdict

NVIDIA A100

80 GB HBM2e · 14% more bandwidth than the RTX PRO 6000

From$1.19/ GPU / hr

Compare

Browse all GPUs

Sources

Where these numbers come from

TFLOPS and TDP are from the vendor datasheet. Hourly prices are from the GPUBeacon catalog sample. Open the originals, then confirm the live SKU with the provider before you rent.