Free tools Windows power users keep installed
One-click scans. No signup required.
Cloud GPU cost is the sum of the GPU accelerator, the virtual machine that hosts it, and every supporting service your workload uses. On-demand capacity costs the most per hour but has no commitment; one- or three-year reservations and commitments lower the effective rate when utilization is predictable; Spot capacity can be dramatically cheaper but may be evicted. There is no universally cheapest cloud for an H100, A100, or any other model until you fix the region, host shape, operating system, term, utilization and tolerance for interruption.
The prices below are dated snapshots from provider pages accessed in 2026. Use the provider calculator or API for a live quote before purchase.
As an Amazon Associate I earn from qualifying purchases.
What a cloud GPU bill actually contains
A GPU-hour is only one line on most invoices. Google explains that each GPU adds to the cost of the instance in addition to the machine type. The host charge covers vCPUs, system memory and sometimes local disks; the accelerator meter covers the GPU itself. Storage, network traffic, image licensing, orchestration and checkpoint recovery can be separate line items.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
A practical estimate is:
Total cost = GPU hours × GPU rate + VM hours × host rate + storage + network + software licensing + orchestration + recovery overhead.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
- GPU rate: the accelerator model, such as a T4, V100, A100 or H100, in a specific region and billing currency.
- Host rate: vCPU and RAM for the machine type attached to the accelerator. A large host can cost more than the GPU for light workloads.
- Storage: boot disks, datasets, container layers, snapshots and retained checkpoints.
- Networking: inter-zone traffic and data egress, especially when data or inference clients cross regions.
- Software and platform fees: operating-system images, commercial drivers, managed orchestration or marketplace images when applicable.
- Operational cost: engineering time to checkpoint, restart and monitor interruptible jobs.
GPU availability and price vary by region and zone. Compare identical GPU models, memory sizes, host shapes, regions, operating systems and currencies; otherwise a lower number may simply describe a different product.
On-demand, reserved and Spot: the three billing choices
| Model | Commitment | Interruption risk | Best fit | Main trade-off |
|---|---|---|---|---|
| On-demand | None | Provider capacity can still be unavailable, but there is no Spot-style eviction policy | Experiments, irregular demand and services that need flexibility | Highest effective unit price |
| Reserved or committed | Usually one or three years, depending on provider and product | Billing discount does not by itself make a workload fault-tolerant | Steady usage with a known configuration | You pay for the commitment even when the GPU is idle or mismatched |
| Spot | None; uses spare capacity | Can be preempted or evicted | Checkpointed training, batch inference, rendering and restartable CI | Capacity and runtime are unpredictable; no availability guarantee on Azure Spot |
How on-demand GPU pricing works
On-demand is usage-based capacity with no long-term contract. AWS describes On-Demand Instances as paying by the hour or second, with a 60-second minimum. The exact metering unit can differ by service, so confirm whether a product bills per second, minute or hour and whether a minimum applies.
Choose on-demand when you are still measuring utilization, testing a new model, serving traffic with uncertain growth or moving between GPU types. It lets you stop a machine as soon as a job ends instead of carrying an unused commitment. The price you see for a GPU is not the complete machine price: add the host VM and supporting services before comparing it with another provider.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
On-demand is also a useful baseline. Record the same region, machine shape and workload duration, then compare committed and Spot alternatives against that baseline rather than comparing headline GPU rates from different configurations.
When a reservation or commitment is cheaper
Reserved pricing exchanges a one- or three-year commitment for a lower effective rate or an automatic discount. AWS Reserved Instances cover a consistent instance configuration, while AWS Savings Plans discount eligible committed spend rather than tying the discount to one exact machine. AWS also offers Capacity Blocks for reserving GPU clusters; a capacity reservation and a billing discount are separate decisions.
Google Cloud lists one-year and three-year commitment prices for eligible GPUs and requires an attached reservation for resource-based GPU commitments. The commitment can reduce the accelerator rate, but the machine type, storage and network charges remain.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
Azure Reserved VM Instance discounts apply automatically to matching VM usage and are offered for one- and three-year terms. Azure says reservations are use-it-or-lose-it by hour: unused entitlement in one hour cannot be carried forward to cover a later hour. Azure does not offer reservations for Spot VMs.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Calculate the utilization you need
Use the following break-even test before signing a term:
Committed cost ÷ on-demand hourly cost = break-even hours.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
Then compare those hours with the number of hours you can realistically run the matching GPU during the commitment. Include planned maintenance, weekends, scaling down and the chance that a newer GPU will make the committed SKU unattractive. A discount is not a saving if the unused portion exceeds the price difference.
Discount versus capacity assurance
A reserved discount may lower the bill without guaranteeing that a scarce GPU is physically available at the moment you request it. Products marketed as reservations or capacity blocks can improve access, but they do not remove storage, networking, software or data-transfer charges. Confirm separately whether you are buying a price commitment, a capacity reservation, or both.
How Spot GPU pricing works
Spot instances consume spare provider capacity. AWS positions Spot as cost-effective for applications that can be flexible about when they run and can tolerate interruption. Google describes Spot VMs as suitable for batch and fault-tolerant work that can withstand preemption.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
Google Cloud’s pricing page accessed in 2026 shows Spot discounts of 60–91% off corresponding on-demand prices for most machine types and GPUs. The same page says Spot discounts can change as often as every 30 days. Azure guidance accessed in 2026 advertises savings of up to 90% compared with regular pay-as-you-go VMs, but actual prices vary by region and SKU.
Azure Spot VMs can be evicted when capacity is needed or when the price exceeds the maximum you configured. They have no availability guarantee or SLA. A low maximum price can therefore make a VM easier to evict, while a higher maximum does not create guaranteed capacity.
Designing for an interruption
- Checkpoint frequently: write model weights, optimizer state and data-loader position to durable storage at a cadence that limits lost work.
- Make startup repeatable: keep the image, container digest, dependencies and launch parameters in version control so a replacement VM can start without manual repair.
- Separate state from compute: store datasets, checkpoints and job metadata outside the ephemeral GPU boot disk.
- Automate replacement: use a queue or scheduler that notices eviction, requests another compatible GPU and resumes from the newest checkpoint.
- Set an escape path: fall back to on-demand capacity, another region or another GPU class when a deadline matters.
- Test recovery: deliberately terminate a worker and measure restart time, lost progress and any duplicate outputs before relying on Spot for production work.
Spot is a poor fit for a stateful, latency-sensitive service without failover. A short interruption can be more expensive than the hourly discount if it causes missed requests, corrupted state or a human recovery effort.
Published GPU-rate examples
The following figures are GPU-only examples from Google Cloud’s pricing page accessed in 2026. They are not complete VM prices, are region-, SKU-, currency- and date-dependent, and should be recalculated before purchase.
| GPU | On-demand per GPU-hour | One-year commitment per GPU-hour | Three-year commitment per GPU-hour |
|---|---|---|---|
| NVIDIA T4 | $0.35 | $0.22 | $0.16 |
| NVIDIA V100 | $2.48 | $1.562 | $1.116 |
What those numbers mean in practice
At exactly 100 GPU-hours, the T4 accelerator portion would be $35 on demand, $22 under the one-year figure or $16 under the three-year figure. The V100 accelerator portion would be $248, $156.20 or $111.60 respectively. These calculations exclude the host machine and every other charge, and they assume the stated commitment rate applies to all 100 hours. A commitment can still cost more overall if you pay for unused hours or cannot keep the matching configuration.
Provider differences that affect a fair comparison
| Provider | On-demand | Committed or reserved | Spot behavior | Important qualification |
|---|---|---|---|---|
| AWS | No long-term commitment; billed by the service’s stated time unit with a 60-second minimum for On-Demand Instances | Reserved Instances for consistent configurations; Savings Plans for committed spend; Capacity Blocks for GPU-cluster reservations | Interruptible spare capacity | Do not treat a billing commitment as automatic physical capacity |
| Google Cloud | GPU charge is separate from machine-type charge | Eligible GPUs have one- and three-year committed-use prices; resource-based GPU commitments require an attached reservation | 60–91% off corresponding on-demand prices for most machine types and GPUs on the page accessed in 2026; discounts may change as often as every 30 days | Region and zone availability can dominate the nominal rate |
| Azure | Regular pay-as-you-go VM pricing | One- and three-year Reserved VM Instances discount matching usage automatically | Variable by region and SKU; maximum-price setting; eviction when capacity is needed or the price exceeds that maximum | Spot has no availability guarantee or SLA; reservations are not available for Spot |
A step-by-step way to compare clouds
- Specify the workload: record model type, batch size, precision, expected GPU count, training or inference duration, latency target and checkpoint frequency.
- Lock the hardware definition: compare the same GPU model and memory, not merely a similar product name. Include whether the host offers one GPU or a multi-GPU topology.
- Choose a region and zone: use the location where data, users and required capacity actually exist. Check quotas and current availability before assuming a price is usable.
- Normalize the host: match vCPUs, RAM, local storage, operating system and virtualization features. Subtracting only GPU rates produces a misleading result.
- Price three scenarios: calculate on-demand, a commitment at realistic utilization and Spot with a recovery allowance. Include boot time and idle time, not just active kernel time.
- Add non-compute charges: include persistent disks, snapshots, object-storage requests, inter-zone traffic, internet egress, licenses, orchestration and monitoring.
- Value interruption: estimate lost work and engineering time from an eviction. If a job loses six hours of training, a 70% Spot discount may not beat a stable machine.
- Validate with a live quote: provider calculators and APIs reflect current currency, regional supply and contract terms more reliably than a static example.
Common mistakes and how to correct them
- Comparing GPU-only prices: add the host VM and storage. A cheap accelerator attached to an oversized host can be expensive.
- Assuming a three-year rate is always best: model utilization and hardware obsolescence. Select a term only when the workload can consume it.
- Using Spot without checkpoints: move state to durable storage and automate resume before switching production jobs.
- Ignoring regional scarcity: verify quota, zonal stock and fallback regions. A nominally cheap GPU that cannot be allocated has no practical value.
- Confusing reservation with discount: read whether the product guarantees capacity, lowers price, or both.
- Forgetting time granularity: check minimum billable duration and stop idle workers promptly. AWS On-Demand Instances, for example, have a 60-second minimum.
- Leaving data-transfer costs out: keep data and compute close when possible, and estimate egress for every output path.
- Using one provider’s currency or tax treatment against another’s: normalize currency, taxes, credits and contract-specific discounts before declaring a winner.
Which model fits common workloads?
- Interactive experimentation: start on-demand while you measure runtime and utilization. Move repeatable, long-running jobs to a commitment only after the pattern is stable.
- Scheduled batch inference: use Spot when jobs can wait, checkpoint and retry; retain on-demand fallback capacity for deadlines.
- Large model training: combine committed or reserved baseline capacity with Spot workers if the training system supports distributed recovery and frequent checkpoints.
- Rendering and media transcode: Spot is attractive for a queue of independent, restartable tasks; keep completed outputs in durable storage.
- Latency-sensitive inference: prefer on-demand or committed capacity with redundancy. Spot alone is unsuitable when an eviction would violate a response-time objective.
- Proofs of concept: use the least-commitment option until the GPU type, region and host requirements are known.
Bottom line
On-demand is the flexible baseline, commitments are cheaper only when you reliably use the contracted configuration, and Spot offers the lowest potential rate in exchange for eviction risk. Compare the complete machine and workload cost in the same region, then choose the model whose operational behavior matches your deadline and recovery design. Because prices and capacity change, confirm the final combination with a provider calculator or API immediately before deployment.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Recommended Free Tools
