packetaia hosted·ai project
Start building
GPUs/NVIDIA H100

H100
coming soon.

80 GB of HBM3 at 3.35 TB/s and the Transformer Engine. The most widely deployed data-centre GPU for training and serving LLMs.

Get notified Compare GPUs
HopperComing soon
80GB HBM3
Not on sale yetComing soonTell us what you would run on itNotify me →

Not available to rent yet

Memory80GB
Bandwidth3,350GB/s
ArchitectureHopper
Form factorSXM5
Good for

What the H100
is built for.

Tensor-core precisions: FP8, BF16, FP16, TF32 and FP64.

Best at

  • Serving 7B–70B models
  • Training and fine-tuning
  • RLHF and alignment runs (PPO, DPO, GRPO)
  • FP8 training and inference

Coming soon

The H100 is not on sale yet. Tell us what you would run on it and we will let you know when it launches.

Get notified
Sizing

Will my model fit
on 80 GB?

Rules of thumb for weights plus a modest KV cache or optimizer state. Longer contexts and bigger batches need more.

WorkloadNeedsOn an H100
Llama 3.1 8B · FP16 inference~18 GBFits
Qwen 32B · FP8 inference~36 GBFits
Llama 70B · 4-bit inference~42 GBFits
Llama 70B · FP8 inference~78 GBTight
Llama 8B · LoRA fine-tune (BF16)~24 GBFits
32B · QLoRA fine-tune~40 GBFits
Flux.1 [dev] · BF16~33 GBFits
Llama 70B · FP16 inference~140 GBToo big
Other GPUs

Not the right fit?
Compare the fleet.

These are on sale now, billed hourly with no contract.

Blackwell

B200

180 GB HBM3e · from $4.25/hr

Blackwell

RTX PRO 6000

96 GB GDDR7 · from $0.66/hr

Ampere

A100 80GB

80 GB HBM2e · from $0.89/hr

Ada Lovelace

RTX 4090

24 GB GDDR6X · from $0.39/hr

The H100
is coming soon.

Tell us what you would run on it and we'll let you know when it launches.

Get notified
No contractTalk to a human any time

Black Friday special - Top up $100, get $50 free with code BLACKFRIDAYSPCL

AVAIL NOW
Start Building →
packet.ai/GPUs/NVIDIA H100
Coming soon, join waitlist
NVIDIA Hopper80 GB HBM3SXM5 / PCIe

NVIDIA H100 GPU

The workhorse of modern AI.

The NVIDIA H100 is the most widely deployed data-center GPU for production AI, 80 GB of HBM3 memory at 3.35 TB/s, 4th-gen Tensor Cores, and the first Transformer Engine purpose-built for large language models. Available on packet.ai from $2.50/GPU-hour.

from $2.50/GPU-hr· Launching soon
≈ 56% below market median
See all pricing
No contractsHourly billingSSH in <5 min
NVIDIA H100 GPU
80GB
HBM3 memory
3.35TB/s
Memory bandwidth
990TFLOPS
FP16 Tensor compute
900GB/s
4th-gen NVLink
Specifications

NVIDIA H100 specifications.

SpecificationValueGreat for
GPU architecture
NVIDIA HopperGH100 · 80B transistors
The most proven architecture in large-scale LLM training and inference.
GPU memory
80 GBHBM3 (SXM) / HBM2e (PCIe)
Fits 7B–70B models natively.
Memory bandwidth
3.35 TB/sSXM5
2× A100 bandwidth.
FP8 Tensor compute
~1.98 PFLOPS
High-throughput low-precision training.
FP16 Tensor compute
~990 TFLOPS
Standard mixed-precision training.
Transformer Engine
4th generation
Up to 4× LLM training speedup vs FP16 on Ampere.
NVLink
900 GB/s4th generation
Near-linear multi-GPU scaling.
MIG support
Up to 7 instances
Partition one H100 into isolated GPU slices.
Availability
On-demandUS & EU regions
SSH-ready in under 5 minutes.
Architecture

The first GPU built for transformers.

Hopper introduced the Transformer Engine, hardware specifically designed to accelerate the attention and feed-forward layers that dominate LLM compute.

4th-gen Tensor Cores + FP8

FP8 Tensor Cores deliver 3× the throughput of A100 on transformer workloads. The current frontier training standard.

80 GB HBM3 at 3.35 TB/s

HBM3 delivers 68% more bandwidth than A100. Critical for large batch sizes and long context lengths.

NVLink 4.0 (900 GB/s)

900 GB/s NVLink for tight multi-GPU coupling. Essential for tensor-parallel training across 8+ GPUs.

Transformer Engine

Hardware-accelerated FP8 mixed precision with dynamic scaling. Native support in PyTorch and JAX.

Use Cases

What the H100 is built for.

LLM inference at scale

80 GB HBM3 and the Transformer Engine handle 7B–70B models in production with sub-100ms latency.

  • Sub-100ms p99 latency
  • 7B–70B models natively
  • FP8 Transformer Engine

Training & fine-tuning

The most common GPU for training frontier models, mature tooling, widespread framework support.

  • PyTorch / JAX / TF native
  • Multi-node NVLink 4.0
  • 4th-gen Transformer Engine

RLHF & alignment

Run PPO, DPO, and GRPO workflows on bursty hourly capacity.

  • Hourly billing
  • Full 80 GB VRAM
  • Fast iteration cycles
Pricing

H100 pricing.

H100 is launching soon on packet.ai. Join the waitlist to be notified.

FAQ

NVIDIA H100, answered.

What is the NVIDIA H100?

The H100 is NVIDIA’s flagship Hopper GPU, 80 GB HBM3 at 3.35 TB/s, the first Transformer Engine, and the most widely deployed GPU for LLM training.

How much does H100 cost?

H100 starts at $2.50/GPU-hour dynamic. See pricing when available.

H100 vs H200?

Same compute die. H200 upgrades to HBM3e: 141 GB vs 80 GB and 4.8 TB/s vs 3.35 TB/s.

H100 vs A100?

H100 is roughly 3× faster for LLM workloads: 990 vs 312 TFLOPS FP16, plus Transformer Engine.

Does H100 support MIG?

Yes, up to 7 isolated MIG instances per H100.

How fast can I get an H100?

On Dynamic, SSH-ready in under 5 minutes.

Coming soon

H100 SXM: Launching soon. Get notified.

Join the notify list and be first to know when H100 SXM capacity opens on packet.ai.

Talk to a human

No commitment · we’ll notify you by email

NVIDIA H100from $2.50/GPU-hr