NVIDIA
GPUDatacenterReleased Mar 16, 2026

NVIDIA RTX PRO 4500 Blackwell Server Edition

256-bit · 10,496 cores

The NVIDIA RTX PRO 4500 Blackwell Server Edition is a single-slot, passively cooled data center accelerator announced at GTC 2026 as the power-efficient member of the RTX PRO Blackwell server line. It combines fifth-generation Tensor cores with FP4 support and fourth-generation RT cores, so it can serve graphics, virtual workstation and AI inference workloads from the same board. Ninth-generation NVENC and sixth-generation NVDEC engines with 4:2:2 support handle video pipelines, and Universal MIG can partition the board into two isolated instances. The full-height, half-length form factor and modest power draw suit it to mainstream enterprise servers and edge platforms.

General

Manufacturer
NVIDIA
Model
RTX PRO 4500 Blackwell Server Edition
Segment
Datacenter
GPU chip
GB203
Architecture
Blackwell
Release date
Mar 16, 2026

Memory

Memory size
32 GB
Memory type
GDDR7
Bus width
256-bit
Bandwidth
800 GB/s
Memory clock
3,125 MHz

Clocks & silicon

GPU clock
—
Boost clock
—
Shader units
10,496 CUDA Cores
Tensor / AI cores
328
RT cores
82
Transistors
45.6 billion
Technology (nm)
4 nm
TSMC 4N
TDP / TBP
165 W

Interface & APIs

Bus interface
PCIe 5.0 x16
SLI / multi-GPU
None (no SLI / NVLink)
DirectX
12 Ultimate (12_2)
OpenGL
4.6
Vulkan
1.4
Other APIs
CUDA 12.0, OpenCL 3.0

AI support

Precisions
FP8FP16BF16INT8INT4TF32FP64
AI accelerator
5th Gen Tensor Cores + 4th Gen RT Cores (82)
Peak AI TOPS
1,600 TOPS
Notes
NVIDIA-published Tensor Core throughput: FP4 1.6 PFLOPS, FP8 811 TFLOPS, FP16/BF16 406 TFLOPS, TF32 203 TFLOPS, FP32 51 TFLOPS, peak RT Core 154 TFLOPS. NVIDIA does not publish clock speeds for this board. Universal MIG allows up to two isolated 16 GB instances.
All NVIDIA GPUs →

Datacenter gpus from NVIDIA announced around the same time as the NVIDIA RTX PRO 4500 Blackwell Server Edition. Add any of them to your compare list to see the specs side by side.