New B200 spot capacity is live in US East from $1.69 per GPU-hour. See availability
Comparison

H100 NVL vs H100 PCIe

Both run on the same nodes, the same templates and the same tiers here — so the choice comes down to memory, bandwidth and price. Every figure below is from the catalogue or the vendor datasheet.

H100 NVL
$0.49/h94 GB · Hopper · 32 available
H100 PCIe
$0.45/h80 GB · Hopper · 96 available
Price gap
8%On spot, per GPU-hour
Over 100 h
$4.00Difference for the same GPU-hours
In short

What actually separates them.

Four differences that change a decision, computed from the numbers rather than asserted.

Bigger models fit on the H100 NVL

94 GB against 80 GB. At FP16 that is about 37B parameters on one GPU, against 32B.

The H100 PCIe costs 8% less per hour

$0.45 against $0.49 on spot — $4.00 of difference over 100 GPU-hours.

Token generation is faster on the H100 NVL

3,938 GB/s against 2,000 GB/s. Inference reads the whole model from memory for every token, so bandwidth, not FLOPS, usually sets the ceiling.

More memory per dollar on the H100 NVL

192 GB per dollar per hour against 178 GB. The usual tie-breaker when both cards fit the job.

Side by side

Every figure we hold on both cards.

A row is marked only when both values exist and one is genuinely better. H100 NVL leads on 5 rows, H100 PCIe on 4.

  H100 NVL H100 PCIe
GPU memoryDecides which models fit at all 94 GB HBM3 80 GB HBM2e
Memory bandwidthSets token throughput on inference 3,938 GB/s 2,000 GB/s
8-bit float (FP8)Halves the memory a model needs, where the stack supports it Supported Supported
Architecture Hopper (2023) Hopper (2022)
Board powerLower is cheaper to run at scale — for us, not for your bill 400 W 350 W
GPU-to-GPU linkOnly matters for multi-GPU training NVLink PCIe only
Node sizes 1×, 2×, 4×, 8× 1×, 2×, 4×, 8×
Spot price $0.49/h $0.45/h
On-demand price $0.79/h $0.69/h
Memory per dollarGPU memory per $1 of spot time, per hour 192 GB 178 GB
30-day reclaim rateShare of spot instances reclaimed <5% <5%
Available now 32 GPUs 96 GPUs

Specifications come from each vendor's datasheet — linked on the H100 NVL page and the H100 PCIe page. Prices are ours, per GPU-hour, billed per minute.

What fits

The same models, on one GPU of each.

Weights only, with 20% of the memory kept free for activations and the KV cache. A cross means one GPU is not enough — the model page has the multi-GPU arithmetic.

Model Precision H100 NVL 94 GB H100 PCIe 80 GB
Llama 3.1 8B8B parameters FP16 · 16 GB Fits Fits
FP8 · 8 GB Fits Fits
INT4 · 4 GB Fits Fits
Qwen2.5 32B32B parameters FP16 · 64 GB Fits Fits
FP8 · 32 GB Fits Fits
INT4 · 16 GB Fits Fits
Llama 3.3 70B70B parameters FP16 · 140 GB Needs 2× node Needs 4× node
FP8 · 70 GB Fits Needs 2× node
INT4 · 35 GB Fits Fits
Mixtral 8x22B141B parameters FP16 · 282 GB Needs 4× node Needs 8× node
FP8 · 141 GB Needs 2× node Needs 4× node
INT4 · 71 GB Fits Needs 2× node
Cost

The same 100 GPU-hours on each.

Compute only, billed per minute. Storage is $0.08 per GB-month on either card.

 H100 NVLH100 PCIe
One hour, spot$0.49$0.45
One day, spot$11.76$10.80
100 GPU-hours, spot$49.00$45.00
100 GPU-hours, on-demand$79.00$69.00
A month, reserved$430.70$357.70

Take the H100 NVL when…

  • Training and full fine-tuning — Enough memory for optimiser states and activations on models a smaller card can only run in inference.
  • Serving large models — A 70B model at FP8 across two GPUs, with room for a long context window.
  • Multi-GPU jobs over NVLink — GPU-to-GPU traffic stays off the PCIe bus, which is what makes tensor and pipeline parallelism worth it.

Full H100 NVL page

Take the H100 PCIe when…

  • Training and full fine-tuning — Enough memory for optimiser states and activations on models a smaller card can only run in inference.
  • Serving large models — A 70B model at FP8 across two GPUs, with room for a long context window.
  • Partitioned serving (MIG) — The GPU can be split into isolated instances, each with its own memory and compute slice.

Full H100 PCIe page

FAQ

H100 NVL or H100 PCIe?

Which is cheaper, the H100 NVL or the H100 PCIe?

The H100 PCIe, at $0.45 per GPU-hour on spot against $0.49 — 8% less. On-demand: $0.69 against $0.79.

Which one should I take for inference?

Take the one that holds the model with room for the KV cache, then the one with more memory bandwidth. Here that is the H100 NVL on capacity (94 GB) and the H100 NVL on bandwidth.

Can I run both on the same job?

Not inside one instance — a node holds GPUs of a single model. You can run two instances in parallel, on the same persistent volume in turn, or split the work: a big card for the model that needs capacity, a cheap one for the parts that do not.

Other comparisons: the full list · all 47 models

Get started

Try both for less than the price of deciding.

An hour on each costs $0.94 on spot. Nothing renews, and your disk moves between them.

Billed per minute from the moment the instance is reachable. Minimum credit $40, no subscription.