NVIDIA publishes this card's throughput without saying whether the figures are dense or with sparsity: the word sparsity does not appear on its product page. The four tensor figures shown here are recorded as sparsity figures, because NVIDIA's datasheet for the Workstation Edition version of this card labels the same 4,000 headline as an effective figure with sparsity, and because the ratio between NVIDIA's own FP16 and TF32 numbers only works if they are quoted on different bases. Dense equivalents are left empty rather than halved. NVIDIA publishes no INT8 or FP64 figure for this card, so none is shown.

NVIDIA RTX PRO 6000 Blackwell Server Edition

Blackwell PCIe 2025 4nm Spec confidence: Official
FP32
120
TFLOPS
VRAM
96 GB
GDDR7
Bandwidth
1.6 TB/s
memory
TDP
600 W
200.0 TFLOPS/kW

Overview

The NVIDIA RTX PRO 6000 Blackwell Server Edition is the rack-mount form of the same GB202 silicon that powers the Workstation Edition: 24,064 CUDA cores, 188 SMs, 752 fifth-generation Tensor Cores and 96 GB of GDDR7 with ECC on a 512-bit bus. What changes is everything around the die. Cooling is passive, so the card depends on chassis airflow and fits 1U and 2U servers; the thermal envelope costs it a little clock, giving 120 TFLOPS of FP32 against the Workstation Edition's 125; and memory bandwidth is 1,597 GB/s rather than 1,792 GB/s. In exchange it supports up to four MIG instances of 24 GB each, confidential computing and secure boot with root of trust, which is what makes it deployable as multi-tenant cloud inventory. This is the RTX PRO part that neoclouds actually rack, and it sits in an unusual place in the lineup: GDDR7 rather than HBM, so it gives up bandwidth against an H200 or a B200, but 96 GB of it at a fraction of the price, which suits inference, rendering and visual computing far better than large-scale training. Power is configurable up to 600 W.

Performance

Peak theoretical throughput by precision type

PrecisionDense2:4 Sparse
FP64
No verified data available Structured sparsity is a tensor-core feature; this vector precision has no sparse form
FP32
32-bit floating point
120TFLOPS Structured sparsity is a tensor-core feature; this vector precision has no sparse form
TF32
TensorFloat-32
234TFLOPS The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision
BF16
Brain Float 16
The vendor publishes only the 2:4 sparse figure for this precision 1,000TFLOPS
FP16
16-bit floating point
The vendor publishes only the 2:4 sparse figure for this precision 1,000TFLOPS
FP8
8-bit floating point
The vendor publishes only the 2:4 sparse figure for this precision 2,000TFLOPS
FP6
No verified data available The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision
FP4
4-bit floating point
The vendor publishes only the 2:4 sparse figure for this precision 4,000TFLOPS
INT8
No verified data available The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision

Specifications

Architecture

Blackwell

Form Factor

PCIe

Launch Year

2025

Process Node

4nm

Memory

96 GB GDDR7

Bandwidth

1,597 GB/s

TDP

600 W

Max Power

600 W

CUDA Cores

24,064

Spec Confidence

Official

Full Specifications

Compute Engine
CUDA Cores 24,064
Tensor Cores 752 (5th Gen)
Streaming Multiprocessors 188
Memory
VRAM 96 GB
Memory Type GDDR7
Bandwidth 1.6 TB/s
Interface Width 512-bit
Interconnect & I/O
PCIe 5.0 x16
Power & Thermal
TDP 600 W
Max Board Power 600 W
Power Connector 1x PCIe CEM5 16-pin
Cooling Passive
Enterprise Features
ECC Memory Yes
MIG Support Up to 4 MIGs @ 24GB
Sparsity Yes
Physical & Media
Width 2-slot
NVENC Engines 4
NVDEC Engines 4
General
Form Factor PCIe
Architecture Blackwell
Process Node 4nm
Launch Year 2025

Datasheet & Resources

Data Provenance

Every figure traced to a source

Primary Source

Publisher
NVIDIA
Published
2025-04-01

Data Quality

Spec confidence
Official
Clock basis
Boost
Core precisions with figures
6 of 9
Normalization
All values in TFLOPS

Rent this GPU

$1.89 / GPU-hr

Lowest of 4 live listings across 1 provider· median $1.89

See all rental prices →

Price history and every provider on the pricing page. Some links are affiliate links.

Similar GPUs

Frequently Asked Questions

How many TFLOPS does the NVIDIA RTX PRO 6000 Blackwell have?

The NVIDIA RTX PRO 6000 Blackwell delivers 120 TFLOPS FP32, 1,000 TFLOPS FP16 (2:4 sparse) and 2,000 TFLOPS FP8 (2:4 sparse) at peak.

What is the power consumption of the NVIDIA RTX PRO 6000 Blackwell?

The NVIDIA RTX PRO 6000 Blackwell has a TDP (Thermal Design Power) rating of 600 watts.

How much memory does the NVIDIA RTX PRO 6000 Blackwell have?

The NVIDIA RTX PRO 6000 Blackwell is equipped with 96 GB of VRAM with 1,597 GB/s of memory bandwidth.

What architecture is the NVIDIA RTX PRO 6000 Blackwell based on?

The NVIDIA RTX PRO 6000 Blackwell is based on the Blackwell architecture, launched in 2025.

Stay Updated on GPU Releases

Get notified when new GPUs are added or specifications are updated.

Loading verification...

No spam, unsubscribe anytime.

Back to GPUs