LEADTEK NVIDIA RTX PRO 6000 Blackwell Server Edition 96GB GDDR7 ECC GPU

Your inference queue is backing up because 24 GB just isn't enough anymore. You've outgrown the mid-range card, but you don't want to wait three weeks for a national chain's back-order or ship a rack unit overseas. This is the card that sits on your shelf and handles the workload today.

The LEADTEK RTX PRO 6000 Blackwell Server Edition gives you 96 GB of GDDR7 with error-correcting code on a 512-bit bus. That's 1,597 GB/s of memory bandwidth behind 24,064 CUDA cores and 752 fifth-generation Tensor Cores. You get 4 PFLOPS of FP4 AI throughput for inference workloads that used to need a cluster.

Split it into up to four Multi-Instance GPU partitions at 24 GB each if you're serving multiple tenants. Confidential Compute is supported, so sensitive model weights stay encrypted in transit and at rest. Secure Boot with a hardware Root of Trust closes the supply-chain gap.

It slots into a PCIe 5.0 x16 lane, runs dual-slot (4.4 in H × 10.5 in L), and draws up to 600 W — configurable down if your PSU headroom is tighter. Four DisplayPort 2.1 connectors are on board for local debug or direct attach.

If your workload is AI inference, large-model serving, or GPU-accelerated rendering in a server chassis, this is the spec you've been circling. Call us and we'll confirm fit with your chassis and power budget before it ships

• 96 GB GDDR7 with ECC on a 512-bit interface delivering 1,597 GB/s bandwidth
• 24,064 CUDA cores, 752 5th-gen Tensor Cores (4 PFLOPS FP4), 188 4th-gen RT Cores (355 TFLOPS)
• Up to 4 Multi-Instance GPU partitions at 24 GB each for multi-tenant serving
• PCIe 5.0 x16, 4× DisplayPort 2.1, dual-slot form factor (4.4 in H × 10.5 in L)
• Confidential Compute support and Secure Boot with hardware Root of Trust
• Configurable power draw up to 600 W; 4× NVENC, 4× NVDEC, 4× JPEG engines