Hopper

NVIDIA H100 SXM

Name: NVIDIA H100 SXM
Brand: NVIDIA
Availability: InStock

The gold standard for LLM training and large-scale AI

VRAM

80 GB

Bandwidth

3.35 TB/s

FP16

989.4 TFLOPS

TDP

700W

Get H100 SXM Pricing View Comparisons

Technical Specifications

VRAM	80 GB HBM3
Memory Bandwidth	3.35 TB/s
FP16 Performance	989.4 TFLOPS
BF16 Performance	989.4 TFLOPS
FP32 Performance	66.9 TFLOPS
INT8 Performance	1,979 TOPS
TDP	700W
Form Factor	SXM5
Interconnect	NVLink 4.0 (900 GB/s)
PCIe Interface	PCIe Gen5 x16
Max GPUs per Server	8 (HGX H100)

Prices vary with supply and import costs. Contact for current India pricing.

Best For

LLM training (GPT, LLaMA, Mistral, Gemma)

Multi-node distributed training with NVLink + InfiniBand

High-performance computing and scientific simulation

Large-scale inference serving at high throughput

Not Ideal For

Budget-constrained inference (consider L4 or L40S for better cost per token)
Desktop workstation use (SXM requires purpose-built server chassis)

Overview

The NVIDIA H100 SXM is the most sought-after data center GPU for AI training. Built on the Hopper architecture, it delivers 989 FP16 TFLOPS and features fourth-generation NVLink providing 900 GB/s GPU-to-GPU interconnect within an 8-GPU HGX baseboard.

For LLM training, the H100 SXM delivers roughly 3x the training throughput of an A100 on transformer-based models. Combined with InfiniBand NDR networking, 8-GPU nodes can scale linearly across hundreds of nodes for training frontier-class models.

The SXM form factor requires a compatible baseboard (HGX H100). We supply complete server solutions from Supermicro, Dell, and HPE that include the chassis, baseboard, CPUs, RAM, and networking pre-configured for your cluster size. All units come with GST invoicing and pan-India delivery.