Please Wait...
Please Wait...
¹Preliminary specifications: May be subject to change
²With sparsity
Fast, global access to NVIDIA Blackwell GPUs on Vultr
Empower your business with top-tier NVIDIA GPU infrastructure designed for high-performance computing (HPC), AI training, and inference. Vultr handles infrastructure management so you can focus on building, training, and deploying AI models at scale.
Accelerate AI capabilities with the NVIDIA HGX™ B200
The NVIDIA HGX™ B200 propels a new era of accelerated computing and generative AI.
Specifications
NVIDIA HGX™ B200 GPU1
Blackwell GPUs
8
Fast Memory
Up to 1.5 TB
Aggregate Memory Bandwidth
Up to 64 TB/s
Aggregate NVLink Bandwidth
14.4 TB/s
FP4 Tensor Core
144 petaFLOPs²
FP8 Tensor Core
72 petaFLOPs²
INT8 Core
72 petaFLOPs²
GPU Memory
Up to 192 GB HBM3e per GPU
Decoders/GPU
2x7 NVDEC | 2x7 NVJPEG
Interconnect
5th Generation NVIDIA NVLink: 1.8 TB/s
Get ready to
build, test, and deploy on
The Everywhere Cloud.
Vultr Cloud GPU
The Future of AI is Here: NVIDIA HGX™ B200
The NVIDIA HGX™ B200 GPU delivers breakthrough performance for AI training, inference, and high-performance computing.
15x faster real-time inference performance
3x faster AI model training
3x lower cost
12x improved energy efficiency
Built for next-generation AI workloads
Purpose-built for the most demanding AI, data processing, and HPC workloads, the NVIDIA HGX™ B200 delivers up to 15x faster real-time inference performance.
Deploy in under 20 seconds
Pre-configured with essential drivers and software
Optimized for rapid scalability and performance
New class of AI superchip
Powered by NVIDIA Blackwell architecture
2nd-Generation Transformer Engine
Accelerating both inference and training
5th-Generation NVLink interconnect
Enables faster and wider communication between GPUs
Secure and confidential computing
Built for secure efficient AI operations
Decompression Engine
Detect and address potential issues early to reduce downtime
RAS Engine
Enhances reliability, diagnostics, and query performance
Accelerated large language model inference
With the second-generation Transformer Engine and FP4 precision, the HGX™ B200 achieves up to 15x faster inference for massive models like GPT-MoE-1.8T compared to the Hopper generation.
Enhanced AI training
The second-generation Transformer Engine with FP8 precision enables 3x faster training for large-scale models, improving the efficiency of AI development.
Advanced data analytics
Blackwell’s dedicated Decompression Engine accelerates data processing with support for compression formats such as LZ4, Snappy, and Deflate. This results in performance that is 6x faster than CPUs and 2x faster than NVIDIA H100 Tensor Core GPUs.