RTX PRO 6000

01 / 07

Contact

NVIDIA RTX PRO 6000 Blackwell Workstation Edition

NVIDIA RTX PRO 6000 Blackwell

92 billion transistors under a black anodized shroud.. RTX 6000 PRO is the largest memory NVIDIA has ever put on a workstation board.

Check on Amazon

As an Amazon Associate, this site earns from qualifying purchases.

SCROLL↓
01 — The die

Ninety-two
billion.

GB202 die... 750 mm² of TSMC 4N silicon. This card enables 188 of its 192 streaming multiprocessors.

GPU
GB202
Architecture
NVIDIA Blackwell
Process
TSMC 4N
Transistors
92.2 billion
Die size
750 mm²
Streaming multiprocessors
188
L2 cache
128 MB
Boost clock
2,617 MHz
02 — Memory

96GB

GDDR7 with ECC across a 512-bit interface, 1,792 GB/s of bandwidth. Enough capacity to hold a model that wouldn’t fit under a single desk.

Capacity
96 GB GDDR7
Error correction
ECC
Interface
512-bit
Data rate
28 Gbps
Bandwidth
1,792 GB/s
Partitioning
Up to 4 instances(MIG)
03 — Compute

24,064
cores.

752 fifth-generation Tensor Cores, and 188 fourth-generation RT Cores, clocking up to 2,617 MHz.

CUDA cores
24,064
Tensor cores
752 · 5th gen
RT cores
188 · 4th gen
FP32
126 TFLOPS
RT core
382 TFLOPS
AI
4,000 TOPS(FP4 with sparsity)
04 — Generation over generation
RTX 6000 AdaPRO 6000
Memory48 GB GDDR62×96 GB GDDR7
Bandwidth960 GB/s1.87×1,792 GB/s
CUDA cores18,1761.32×24,064
FP3291.1 TFLOPS1.38×126 TFLOPS
RT core210.6 TFLOPS1.81×382 TFLOPS
L2 cache96 MB1.33×128 MB

On FP8, the gain is 1.38×.

05 — Cooling

Double flow-through: 2 axial fans draw air out through the shroud.. rather than dumping it back into the chassis.

06 — Thermals and I/O

600W

Max-Q cuts power to 300 W without sacrificing memory or bandwidth.

Host interface
PCIe 5.0 ×16
Display outputs
4 × DisplayPort 2.1b
Max displays
2 × 8K at 60 Hz(or 4 × 4K at 120)
Encode
4 × NVENC, 9th gen
Decode
4 × NVDEC, 6th gen
Form factor
Dual slot · 12.0 in × 5.4 in

01

Workstation

600 W

The full card. Double flow-through cooling, four DisplayPort 2.1b outputs, 1,792 GB/s.

  • 600 W TGP
  • 1,792 GB/s
  • 126 TFLOPS FP32
  • Dual slot

02

Max-Q

300 W

Half the power for desks that can’t handle 600 W, keeping all 96 GB and the same memory bandwidth.

  • 300 W TGP
  • 1,792 GB/s
  • 110 TFLOPS FP32
  • Full height, 10.5 in

03

Server

up to 600 W

Passive or liquid cooled, built for dense systems.

  • Configurable TGP
  • 1,597 GB/s
  • 120 TFLOPS FP32
  • Passive / liquid
Announced at GTC, March 2025

The whole
workstation.

24,064

CUDA cores

188 SMs across 12 GPCs

96

GB GDDR7

512-bit, with ECC

1.8

TB/s bandwidth

1,792 GB/s at 28 Gbps

600

W total board power

single 16-pin 12V-2x6

As an Amazon Associate, this site earns from qualifying purchases.

Peak rates use GPU boost clocks. AI TOPS are FP4 with sparsity.