GGPU Sizing Studio
Try the calculator
Live pricing across 6 providers

Know exactly which GPU your model needs.

Size the VRAM, KV cache, and cloud instance for any open LLM in seconds. Then turn it into a report your client will actually read, with your name on it, not ours.

Try the free calculatorBrowse LLM GPU requirements

No signup to run it. Free account for exports. Teams work in a shared workspace. Sign in.

0cloud & GPU providers
0steps to a client-ready plan
$0to run a sizing
Real sizing result from GPU Sizing Studio: Llama 4 Scout memory breakdown at 280.9 GB total VRAM, ranked against RunPod, Azure, CoreWeave, AWS, and GCP instances.

Live output from GPU Sizing Studio, sizing Llama 4 Scout in real time.

How it works

From a model name to a client-ready answer

Three steps, no spreadsheet. Built for the way consultants and platform teams actually size hardware.

STEP 01

Pick a model, get the memory

Choose from the catalog or paste a Hugging Face config. Weights and KV cache broken out for dense or mixture-of-experts models, from FP32 down to INT4.

STEP 02

Compare it across the clouds

Ranked GPU instances across hyperscalers and neoclouds, with multi-GPU topology, live pricing, and throughput and latency estimates for the traffic you expect.

STEP 03

Hand over a report

Export a branded PDF, or audit a running fleet and flag what's over- or under-provisioned.

Why teams switch

What this actually replaces

The real cost isn't the calculator work. It's what happens after a wrong guess ships.

Sizing GPUs off a spreadsheet and a hunch
Exact VRAM and KV cache math for any open model
Picking a cloud on sticker price alone
Real instance fit ranked across hyperscalers and neoclouds
A plan nobody outside your head can check
A branded report a client or lead can actually read
Fleets that quietly over- or under-provision
Usage auditing that flags waste before it costs you

Coverage

Sized against the clouds you'll actually deploy on

Real instance shapes, live pricing, and topology, not list prices from a spec page.

AWS
Hyperscaler
EC2 P4, P5, and G5: A100, H100, A10G.
Azure
Hyperscaler
ND-series VMs with A100 and H100, InfiniBand networking.
Google Cloud
Hyperscaler
A2 and A3: A100 and H100 across regions.
RunPod
Neocloud
Secure Cloud and community pods, priced by the GPU-hour.
CoreWeave
Neocloud
Multi-node H100/A100 clusters built for large training and inference.
Lambda
Neocloud
On-demand and reserved instances built for ML workloads.

Stay in the loop

New models, providers, and pricing, straight to your inbox.

Pricing

Start free. Upgrade when it earns its keep.

The calculator is free for anyone. A free account adds exports and saved runs.

FREE
$0
For a quick answer
  • Full calculator, no account
  • Shareable result link
  • Every model in the directory
Open the calculator
MOST POPULAR
PRO
$19/mo
For serious individual use
  • PDF and JSON export
  • Saved run history
  • Quantization comparison
  • Team-sharing links
Create an account
TEAMS
Talk to us
For consultants & platform teams
  • Multi-client projects
  • Branded client reports
  • Multi-node sizing
  • Live fleet & waste alerts
See the team workspace

Early feedback

What early users are telling us

Real feedback from teams already using GPU Sizing Studio to plan deployments.

Cut our GPU sizing overhead by around 70%. What used to eat a whole afternoon in spreadsheets now takes minutes.

ML
ML infrastructure leadEarly user

Everything we needed was under one roof: models, providers, pricing. No more juggling half a dozen docs and sheets.

LD
LLM deployment engineerEarly user

Reliable, and a genuine time saver. It just works, every time we've reached for it.

AI
AI platform teamEarly user

FAQ

Frequently asked questions

Does this account for quantization?

Yes. Switch precision on any sizing run (FP32 down to INT4) and weights, KV cache, and instance fit all recalculate.

Do I need an account to use the calculator?

No. The full calculator runs with no signup. A free account just adds exports and saved run history.

Which providers are included?

AWS, Azure, and Google Cloud, plus GPU-focused neoclouds like RunPod, CoreWeave, and Lambda, all ranked side by side on every run.

Can I audit an already-running fleet, not just plan a new one?

Yes, on the Teams plan. Connect real usage data and GPU Sizing Studio flags instances that are over- or under-provisioned.

Ready to size your fleet?

No credit card, no signup. Enter a model and get a plan in under a minute.

Try the free calculator