H100 SXM vs. H200 SXM: Which GPU Should You Choose?

We checked provider pricing and NVIDIA specifications to settle whether H100 SXM or H200 SXM is the better fit for your workload and budget.

Updated 2026-08-23 · refreshed daily

Given the 0.44 on-demand price ratio, choose H100 SXM when the workload fits one GPU, choose H200 SXM when its larger memory capacity reduces the GPU count, with 7B full fine-tuning the key exception.

Before committing, check the live table to confirm that the on-demand minimums remain $1.74 for H100 SXM and $3.99 for H200 SXM.

How much GPU memory does your workload actually require? Choose H100 SXM for FP8, BF16, or FP32 workloads within 80GB VRAM; choose H200 SXM when the workload exceeds that limit but still meets FP8, BF16, or FP32 workloads within 141GB VRAM, and use a multi-GPU configuration beyond it.
Is finishing sooner worth the difference in cost per job? Choose H200 SXM only if its measured runtime is necessary to meet the deadline; otherwise choose H100 SXM, with an effective cost-per-job ratio of 0.44 versus H200 SXM.
Can your workload tolerate interruptions and restart safely? Run one representative job on a short interruptible rental to verify that it resumes cleanly after preemption.

Today's prices

USD/hr · 1× GPU · lowest on-demand price per provider

ProviderH100 SXMH200 SXM
Vast.ai $1.74 Not offered
Hyperstack $3.20 · Not reported $3.99 · Not reported
DataCrunch $3.25 $4.00
RunPod $3.29 $4.59
Crusoe $3.90 · Not reported $4.29 · Not reported
Together AI $3.99 · Not reported $5.99 · Not reported
Paperspace $5.95 · Not reported Not offered

see all H100 prices →

Prices refresh daily. Marlin always books the cheapest available. Try Marlin →
0.44×
On-demand pricing favors H100 SXM when one GPU fits the workload
0.44×
H100 SXM offers the lower effective on-demand cost per compute-equivalent job
$1270/mo
Monthly cost of running one H100 SXM continuously at its on-demand minimum

What your workload needs

Find your parameter count on the model's Hugging Face card. Serving maps to the inference rows, training to the fine-tuning rows.

WorkloadVRAM neededH100 SXMH200 SXMCheapest today
70B Q4 inference 42 GB 1× $1.74/hr 1× $3.99/hr H100 SXM ×1 → $1.74/hr
70B FP16 inference 168 GB 3× $1.74/hr 2× $3.99/hr H100 SXM ×3 → $5.22/hr (est.)
70B QLoRA 46 GB 1× $1.74/hr 1× $3.99/hr H100 SXM ×1 → $1.74/hr
7B full fine-tune 112 GB 2× $1.74/hr 1× $3.99/hr H100 SXM ×2 → $3.48/hr (est.)
70B FP8 training 154 GB 2× $1.74/hr 2× $3.99/hr H100 SXM ×2 → $3.48/hr (est.)
Downshift recommendation 42 GB 1× $1.74/hr 1× $3.99/hr H100 SXM ×1 → $1.74/hr

What the numbers say

H100 SXM carries an on-demand price ratio of 0.44 relative to H200 SXM, so it is the cost-first choice when either GPU can run the job on the same device count. Choose H200 SXM when its larger memory capacity reduces the number of GPUs required.

H200 SXM matches H100 SXM at the datasheet-based performance ratio of 1, which is a peak specification rather than measured application throughput. On that indirect basis, H100 SXM has an effective on-demand cost-per-job ratio of 0.44 versus H200 SXM, though actual results depend on the workload.

H100 SXM covers FP8, BF16, or FP32 workloads within 80GB VRAM, while H200 SXM supports the same precisions with a larger VRAM ceiling. The exception is a workload whose model state, activations, and runtime memory exceed the first ceiling but fit on one H200 SXM; choose H200 SXM when that headroom avoids a multi-GPU configuration, and choose H100 SXM otherwise.

Before you rent

Data egress fees can raise the total bill; verify the provider's current per-GB transfer terms for your deployment region before renting.
Crossing a single GPU's VRAM ceiling can force a multi-GPU configuration, multiplying device-hours while communication overhead extends billed runtime.
Lower-priced spot rows are interruptible, so workloads without reliable checkpoint recovery risk repeated compute and missed completion windows.
Billing granularity can make short experiments cost more than runtime alone suggests; each provider's pricing page lists the applicable minimum charge.

Marlin matches your workload requirements to the lowest-priced suitable GPU across major cloud providers.

Source coverage is limited to the listed providers and vendor specifications, while prices and availability can change without notice. Workload costs and effective-cost comparisons are modeled from listed rates, VRAM formulas, and datasheet peak performance rather than measured end-to-end benchmarks. https://vast.ai/pricing · https://www.coreweave.com/pricing · https://www.hyperstack.cloud/gpu-pricing · https://www.together.ai/pricing · https://datacrunch.io/pricing · https://resources.nvidia.com/en-us-tensor-core/nvidia-tensor-core-gpu-datasheet · https://www.nvidia.com/en-us/data-center/h200/ · https://www.runpod.io/pricing · https://lambda.ai/service/gpu-cloud · https://crusoe.ai/cloud/pricing

Stop comparing. Start running.

Marlin matches your workload to the lowest-priced suitable GPU across supported providers.

Get started with Marlin