GitHub delivery evidence
Build and verify the change with GitHub context alongside it.
AWS · GCPHourly + monthly + spotGPU specs in the row
Every price for that instance, already on the row.
Hourly, monthly and spot for the same instance, side by side, in one table. Not a pricing page, then a calculator, then a datasheet, then a spot dashboard. One row, and you have the answer.
m7i.2xlargeaws8 vCPU · 32 GB · x86_64 · us-east-1
Hourly
$0.4032
on-demand
Monthly
$294.34
× 730 hrs
Spot
$0.1391
same row
Compute pricing is planning input for the engineering stage.
One connected loop, held on the stage this capability serves. The other stages stay as context so you can see what feeds in and what comes next.
BD
Discovery
Feedback
Product
Engineering
Release
Reliability
Evidence
Engineering
Build and verify the change.
Available now
VCS insights
Inspect delivery workflow activity in the connected release workspace.
Test intelligence
Investigate test outcomes and flakiness evidence across runs.
API Explorer
Keep API contracts inspectable next to the release that changed them.
Compute pricing
Compare public cloud compute pricing as planning evidence.
What this is, and what it isn't
This is a comparison catalog, not an infrastructure calculator. You don't assemble a hypothetical architecture and get a total. You look up instances that exist, compare them on equal terms, and get to a shortlist. Every row carries its price-effective and indexed date. Confirm the final number with the provider before you buy.
AWS and GCP, normalized into one table.
Two clouds, one schema. Every row carries hourly, spot and monthly together, with the GPU spec beside them and a magnitude bar so you rank by size before you read a digit. The instance column stays put while the rest scrolls — 17 columns, one row per instance. A reserved-only instance shows a dash, never a misleading $0.00.
| Instance | vCPU | Memory | Hourly | Spot | Monthly | GPU | GPU mem | GPU cores |
|---|---|---|---|---|---|---|---|---|
| c7g.2xlargeaws | 8 | 16 GB | $0.2900 | $0.1015 | $211.70× 730 | None | — | — |
| n2-standard-8gcp | 8 | 32 GB | $0.3880 | $0.0970 | $283.24× 730 | None | — | — |
| g5.xlargeaws | 4 | 16 GB | $1.0060 | $0.3520 | $734.38× 730 | 1× A10G | 24 GB | 9,216288 tensor |
| a2-highgpu-1ggcp | 12 | 85 GB | $3.6730 | $1.1020 | $2,681.29× 730 | 1× A100 | 40 GB | 6,912432 tensor |
| p5.48xlargeaws | 192 | 2048 GB | — | — | — | 8× H100 | 640 GB | 16,896528 tensor |
Eight of seventeen columns shown. Toggle CPU arch, processor, generation, region, purchase, spot monthly and GPU gen back on any time.
Monthly is right there.
Providers quote hourly and let you do the arithmetic. We do it on the row, at 730 hours by default, or 720 or 744 if that's how your month is counted. Set a quantity and the row scales with it. You never leave the table to find out what it costs.
- Hourly, as quoted$0.4032/hr
- Monthly · 730 hrs$294.34/mo
- Monthly · × 6 instances$1,766.02/mo
Cheapest and best-value are different questions.
The lowest sticker price rarely wins once you divide by what you're buying. Every ratio here is a real sort key: rank by price per vCPU, per GB of memory, or per GB of GPU memory and the ordering answers the question you actually asked.
a2-highgpu-1ggcp$3.6730/hr on-demand
- Price per vCPU$0.3061$3.6730 ÷ 12 vCPU
- Price per GB memory$0.0432$3.6730 ÷ 85 GB
- Price per GB GPU memory$0.0918$3.6730 ÷ 40 GB
Spot, specs, and value ratios.
The spot price, on the on-demand row.
Spot pricing normally lives in a different console entirely, so the comparison never actually happens. Here it's joined onto the row you're already reading. The discount is a glance, not a project.
same rowper regionno console switch
- on-demand$0.4032/hr
- spot$0.1391/hr
- difference−65%
Read the card without leaving the row.
Picking a GPU instance usually means a vendor datasheet in another tab to remember how much memory an A100 has. The spec is curated per model and readable in place, so the instance and the silicon are one decision.
memory + typecuda + tensor coresfp16 tflops
NVIDIA A100nvidia
- Memory
- 80 GB
- Type
- HBM2e
- CUDA cores
- 6,912
- Tensor cores
- 432
- FP16
- 312 TFLOPs
- Launched
- 2020
Sort by what you're actually buying.
Cheapest instance and best value are different questions. Sort by price per vCPU, price per GB of memory, or price per GB of GPU memory, and the ranking answers the second one.
price per vCPUprice per GBprice per GPU GB
- c7g.2xlarge · arm64$0.0362
- m7i.2xlarge · x86_64$0.0504
- r7i.2xlarge · x86_64$0.0661
Paste the table, not a screenshot.
Every filter, sort, quantity, monthly-hours choice and column toggle is serialized into the URL. Drop it in a thread and your colleague opens the exact table you were looking at — same rows, same order, same columns — not an empty catalog they have to rebuild.
/compute-pricing?provider=gcp&purchaseOption=spot&vcpuMin=8&gpuClass=a100&sortBy=pricePerVcpu&sortDirection=asc&monthlyHours=730&quantity=4
- provider=gcp
- purchaseOption=spot
- vcpuMin=8
- gpuClass=a100
- sortBy=pricePerVcpu
- sortDirection=asc
- monthlyHours=730
- quantity=4
Seventeen columns. You pick, it remembers.
Nobody needs all seventeen at once. Turn off the ones you never read — processor, spot monthly, GPU gen — and they stay off next time, per person, saved to your account and the URL alike.
- Instance
- Provider
- Purchase
- Region
- vCPU
- Memory
- CPU arch
- CPU Gen
- Processor
- Hourly
- Spot
- Monthly
- Spot monthly
- GPU
- GPU mem
- GPU cores
- GPU gen
The small things, done.
A comparison tool lives or dies on whether it gets out of your way. These are the parts nobody markets and everybody feels.
- Quantity
- Set how many you're running and every price on the row scales with it. Six instances is a number you read, not one you compute.scales the row
- Hours per month
- 730 by default, or 720 or 744. Whichever convention your finance team uses, the monthly matches.720 · 730 · 744
- Unavailable, honestly
- High-end GPU instances that only sell through reserved capacity report a $0.00 on-demand rate. We show a dash and a tooltip instead, so 'not actually on-demand' never reads as 'free'.dash + tooltipreserved-only
- Say what you mean
- Hover any architecture, generation, processor or purchase option and it explains itself. Nobody should have to already know what committed-use means to use a price table.inline explanations
- Dated, always
- Every row carries when the price took effect and when we indexed it. A stale number that admits it is stale is worth more than a fresh-looking one that isn't.price-effectiveindexed at
Faceted, not flat.
You rarely start from 'show me everything'. You start from 'arm64, 8 vCPU or more, with an A100, in Frankfurt'. Every one of those is a facet, and each facet counts its matches before you click.
Providers
Normalized together.
awsgcp
Purchase options
How you buy.
on_demandspot
Size
Ranges, not guesses.
vcpu min/maxmemory min/maxfamilygeneration
Silicon
Down to the model.
gpu classgpu modelprocessorx86_64arm64
Where + how
Region and OS.
regionlinuxwindowsrhelslesubuntu_pro
Sort by
Fifteen keys.
hourlyPricemonthlyPricepricePerVcpupricePerMemoryGbgpuMemoryGbgpuCudaCores
2
clouds normalized into one row: AWS and GCP
3
prices on every row: hourly, monthly, and spot
15
sorts, including price per vCPU and per GB of GPU memory
17
columns you can turn on or off, and it remembers
More on the way
Reserved and committed-use pricing, more providers, and saved comparisons are what we're building next. The rule stays the same: if a number matters to the decision, it belongs on the row, not one tab away.
See what else the engineering stage carries.
Get to the shortlist in one tab.
Start free
Free to browse. Confirm the final number with the provider.