Compare
Nexpod vs AWS for GPU work.
8× B300 · €48.00 /hrone line per serverinvoiced in EUR
Teams running GPU work on Nexpod
Placeholder- Customer
- CUSTOMER
- customer
- Customer Co
- customer.ai
- CUSTOMER
GPU teams on Nexpod
Placeholder- Customer
- CUSTOMER
- customer
- Customer Co
- customer.ai
- CUSTOMER
Before you choose
Plan for more than the GPU.
On a broad cloud, the GPU is one line in a larger setup. Fill in this worksheet for your own account before you move a run.On a broad cloud, the GPU is one line inside a larger setup. Before you move a training run or an endpoint, fill in this worksheet for your own account and region.
- GPU quota requestfamily8× ____region____status awaiting review
Capacity and quotas
Confirm the GPU model, the region and the quota you need before you set a launch date.
- Lines on one month's billGPUstoragetransfersupport
Every line on the bill
Add compute, storage, data transfer and support together. The GPU rate is only one part.
- Before the first job2 / 5
- network
- image
- roles
- scripts
- first job
Setup before the first job
Count the networking, images, roles and scripts a job needs before it can start.
- Invoice currencyEURUSDGBPagreed with finance: ?
Procurement and currency
Check the contract path and the currency your finance team will be invoiced in.
Before you choose
Plan for more than the GPU.
On a broad cloud, the GPU is one line in a larger setup. Fill in this worksheet for your own account before you move a run.On a broad cloud, the GPU is one line inside a larger setup. Before you move a training run or an endpoint, fill in this worksheet for your own account and region.
- 01GPU family offered in your region
- 02Quota approved for 8 GPUs or more
- 03Storage and data transfer estimated
- 04Networking, images and roles scoped
- 05Invoice currency agreed with finance
- 06Reserved term and how to exit it
Capacity and quotas
Confirm the GPU model, the region and the quota you need before you set a launch date.
Every line on the bill
Add compute, storage, data transfer and support together. The GPU rate is only one part.
- network
- image
- roles
- scripts
- first job
Setup before the first job
Count the networking, images, roles and scripts a job needs before it can start.
Procurement and currency
Check the contract path and the currency your finance team will be invoiced in.
Why Nexpod
Eight GPUs, three ways to run them.
Pricing
Rates you can read in one line.
One price per GPU-hour for each family, metered per second, in euros. A full server costs eight times that.One price per GPU-hour for each family, metered per second, in euros. A full server costs eight times the GPU rate. That is the whole formula.
Rates differ by region, instance size and term, so there is no single number to put here.
→ check AWS's own price list for your region
AWS rates differ by region, instance size and term. Check AWS's own price list for your region.
Placeholder quote. A customer who moved GPU work from AWS describes, in two plain sentences, what changed on the invoice and in the day-to-day.
Comparison
Nexpod and AWS, side by side.
Our answer on the left. On the right, the question to put to your own AWS account.Our answer on the left. On the right, the question to put to your own AWS account, because the answer depends on region, instance type and agreement.
- What you rent
- Whole 8-GPU server
- ?Which instance types carry eight GPUs?
- GPU families
- B300 · B200 · H200 · H100
- ?Which GPU families run in your region?
- Billing granularity
- Per second
- ?Per second or per hour, for your OS and service?
- Invoice currency
- EUR
- ?Which currency is your account billed in?
- Egress from network storage
- None
- ?What does data transfer out of storage cost you?
- Serverless GPU endpoints
- ?Which service hosts autoscaling GPU endpoints?
- Multi-node clusters
- ?Which networking joins GPU instances into one job?
- Root over SSH on the node
- ?A full host, or a virtual machine on one?
- Rate shown before you deploy
- ?Where is the full hourly rate shown before launch?
- What you rent
- Nexpod: Whole 8-GPU serverWhat to check on AWS: ?Which instance types carry eight GPUs?
- GPU families
- Nexpod: B300 · B200 · H200 · H100What to check on AWS: ?Which GPU families run in your region?
- Billing granularity
- Nexpod: Per secondWhat to check on AWS: ?Per second or per hour, for your OS and service?
- Invoice currency
- Nexpod: EURWhat to check on AWS: ?Which currency is your account billed in?
- Egress from network storage
- Nexpod: NoneWhat to check on AWS: ?What does data transfer out of storage cost you?
- Serverless GPU endpoints
- Nexpod:What to check on AWS: ?Which service hosts autoscaling GPU endpoints?
- Multi-node clusters
- Nexpod:What to check on AWS: ?Which networking joins GPU instances into one job?
- Root over SSH on the node
- Nexpod:What to check on AWS: ?A full host, or a virtual machine on one?
- Rate shown before you deploy
- Nexpod:What to check on AWS: ?Where is the full hourly rate shown before launch?
Workload fit
Pick by workload, not by habit.
Most teams end up using more than one cloud. Use this to decide where each workload belongs.
Nexpod fits when
- Your work is GPU-bound: training, fine-tuning, inference or rendering.
- You want a whole 8-GPU server that nobody else shares.
- You want per-second billing and invoices in euros.
- Your data can live in EU-AMS-1, EU-FRA-1 or US-NYC-1.
AWS may fit better when
- Most of your data, apps and permissions already live in one AWS account.
- You need managed services right next to the GPU, not only the GPU.
- Your procurement already runs through an existing AWS agreement.
Talk to an engineer
15 minutes with a GPU engineer.
Bring your AWS setup and a monthly budget. We map it to Nexpod servers and a quote in euros, or tell you to stay put.Bring your current AWS setup and a rough monthly budget. We map it to Nexpod servers and a quote in euros, or tell you plainly to stay where you are.
Try it on one workload
Move one workload. Compare the invoice.
Move one training run or endpoint, keep the rest where it is, and judge on your own numbers.Start with a single training run or endpoint on Nexpod, keep everything else where it is, and judge on your own numbers.
- B300· 288 GB HBM3e€6.00/GPU·hr
- B200· 180 GB HBM3e€5.00/GPU·hr
- H200· 141 GB HBM3e€4.00/GPU·hr
- H100· 80 GB HBM3€3.00/GPU·hr