NVIDIA RTX PRO 6000 Blackwell
Now live on JarvisLabs — billed by the minute
NVIDIA's professional Blackwell GPU is live on JarvisLabs. 96GB GDDR7 with ECC memory, 5th-gen Tensor Cores with FP4 support. Launch a PyTorch instance in under 60 seconds.
Powering teams that push boundaries
Trusted by companies including: Tesla, Hugging Face, Kaggle, Zoho, Weights & Biases, upGrad, Saama
Professional Blackwell, available today
The RTX PRO 6000 Blackwell brings next-gen Blackwell architecture and 96GB ECC memory to JarvisLabs — launchable now with per-minute billing.
96GB GDDR7 with ECC
Professional-grade memory with error correction. Double the RTX 6000 Ada's 48GB. Run large models reliably.
5th-Gen Tensor Cores
FP4 native support for efficient inference. Significant upgrade from the RTX 6000 Ada's 4th-gen cores.
Professional Drivers & Support
ISV-certified drivers, long-term stability, enterprise support. Critical for production deployment.
Blackwell Architecture
Next-generation performance improvements across training and inference. Built for professional AI workloads.
Key specs at a glance
Other professional GPUs on JarvisLabs
Need a different memory tier or price point? These professional GPUs are also live, with per-minute billing.
What the RTX PRO 6000 Blackwell is built for
From professional AI development to large-scale model inference.
Professional AI Development
ISV-certified for enterprise AI frameworks. Stable drivers for production workloads. ECC memory prevents silent data corruption.
Large Model Inference
96GB fits 70B models in FP8 comfortably. Professional reliability for customer-facing inference.
AI Workstation
Desktop form factor with professional cooling. Runs 24/7 without datacenter infrastructure.
Model Training & Fine-Tuning
96GB enables LoRA/QLoRA for up to 70B models. Full fine-tuning for 13B-34B models.
Technical specifications
Hardware specifications for the NVIDIA RTX PRO 6000 Blackwell. Some specs are estimated based on NVIDIA's Blackwell professional lineup announcements.
| Specification | Value | Notes |
|---|---|---|
| Architecture | NVIDIA Blackwell | Next-gen professional GPU |
| CUDA Cores | ~18,000+ (estimated) | Blackwell professional tier |
| Tensor Cores | 5th gen | FP4 / FP8 / FP16 / BF16 / INT8 |
| RT Cores | 4th gen | Real-time ray tracing |
| VRAM | 96 GB GDDR7 (ECC) | Professional-grade with error correction |
| Memory Bandwidth | ~1,800-2,000 GB/s (estimated) | GDDR7 high-bandwidth |
| FP4 Support | Yes (native) | Efficient quantized inference |
| NVLink | NVLink C2C | 2-GPU configurations |
| PCIe | Gen5 x16 (expected) | Host data transfer |
| TDP | ~350-400W (estimated) | Professional workstation cooling |
| Form Factor | Dual-slot PCIe | Desktop workstation compatible |
| Manufacturing | TSMC 4NP | Advanced process node |
| ECC Memory | Yes | Data integrity for production workloads |
| Driver Support | ISV-certified professional | Enterprise stability & support |
RTX PRO 6000 Blackwell is live on JarvisLabs. Launch a PyTorch instance with 96GB GDDR7 ECC memory, per-minute billing, and 1–8 GPU configurations. Sign up and spin one up in under a minute.
Frequently asked questions
Everything you need to know about the NVIDIA RTX PRO 6000 Blackwell on JarvisLabs.
Sign in, pick the PyTorch template, and select the RTX PRO 6000 GPU — instances are typically ready in under 60 seconds.
Launch RTX PRO 6000 Blackwell on JarvisLabs
96GB GDDR7 ECC. 5th-gen Tensor Cores. Blackwell architecture. Per-minute billing.