
NVIDIA RTX PRO 5000 Blackwell 72GB
Paid in Bitcoin, Monero, USDT or another coin you choose.
Free insured shipping. US delivery in about 5 to 11 business days. Shipping details
Overview
The NVIDIA RTX PRO 5000 Blackwell 72GB is the 72GB version of NVIDIA's RTX PRO 5000 Blackwell, which adds 50% more memory than the 48GB card for larger local models. It has 14,080 CUDA cores and 72GB of GDDR7 with ECC on a 384-bit bus, for 1,344 GB/s of bandwidth. It is built for AI inference and fine-tuning, simulation and rendering, with 65 TFLOPS FP32 and up to 2,064 AI TOPS.
It is a standard two-slot, full-height blower card at 267 mm long, so several fit side by side in a workstation. Board power is 300W (NVIDIA), it takes 1x PCIe CEM5 16-pin, and it drives 4x DisplayPort 2.1b. Other features: Multi-Instance GPU (up to 2 x 36GB), 3x NVENC and 3x NVDEC, ECC memory, PCIe 5.0 x16.
For local AI, 72GB holds 70B models at 4-bit to 6-bit and gpt-oss-120b on a single card, with 1,344 GB/s of bandwidth and ECC memory for long training and inference runs. CUDA support is broad, so PyTorch, llama.cpp, Ollama, LM Studio and ComfyUI run on it directly.
Specifications
| GPU | RTX PRO 5000 Blackwell (72GB), Blackwell |
|---|---|
| CUDA cores | 14,080 |
| Memory | 72GB GDDR7 with ECC, 384-bit |
| Memory bandwidth | 1,344 GB/s |
| PCIe | PCIe 5.0 x16 |
| Board power | 300W (NVIDIA) |
|---|---|
| Power connectors | 1x PCIe CEM5 16-pin |
| Size | 267 x 112 mm, full height, 2 slots |
| Display outputs | 4x DisplayPort 2.1b |
|---|---|
| Cooling | Active blower that exhausts out of the rear |
| Card features | Multi-Instance GPU (up to 2 x 36GB), 3x NVENC and 3x NVDEC, ECC memory, PCIe 5.0 x16 |
| Warranty | 3-year limited warranty (PNY) |
| Model | PNY VCNRTXPRO5000B72-PB |
Questions and answers
Still unsure? Visit the help center or message us.
What power supply does the NVIDIA RTX PRO 5000 Blackwell need?
The card draws up to 300W through one 16-pin CEM5 connector. Size the power supply for the whole workstation: allow about 300W per card on top of the CPU and the rest of the system, plus 20% headroom.
Will the NVIDIA RTX PRO 5000 Blackwell fit my case?
It is a standard two-slot, full-height blower card at 267 mm long, so several fit side by side in a workstation.
Is 72GB enough for local AI?
For local AI, 72GB holds 70B models at 4-bit to 6-bit and gpt-oss-120b on a single card, with 1,344 GB/s of bandwidth and ECC memory for long training and inference runs.
Does it support NVIDIA's professional drivers?
Yes. It runs NVIDIA's RTX Enterprise drivers, certified for major CAD, DCC and simulation apps, as well as the standard CUDA toolkit for AI frameworks.
Pairs well with
Picked to go with this product. Check sizes and connectors for your build before you order.







