Ready to use
The operating system, drivers, and frameworks are ready. Register, choose a GPU type, and start running.
GPU Cloud
Sign up and start using GPUs billed by the minute
Glows.ai is KONST's GPU cloud platform partner, using compute from data centers built or managed by KONST. Register online, choose a GPU type, and launch an instance with the operating system, drivers, and frameworks already configured. No data center or environment setup is required. Instances are billed by the minute, with no monthly fee or minimum usage commitment, and billing stops at shutdown.
FEATURE｜ Service details and specifications
Start training as soon as your instance is ready
No data center or environment setup is required. Choose a GPU type and run your workload. Instances are billed by the minute, and billing stops when they are shut down.
Per-minute billing
No monthly fee or minimum usage commitment. Billing stops when the instance is shut down.
Change GPU types without rebuilding
Data is mounted when the instance starts and retained after shutdown. Snapshots can be restored to a new instance.
SOLUTION｜ The KONST approach
Configure GPUs for development, team use, or online inference
Launch single-node or multi-GPU instances on demand for development and short training runs. Virtualized clusters combine nodes into a shared resource pool, while shared and dedicated inference services support latency-sensitive online workloads.
Interactive development and short training runs
For experimental environments that need to start and stop as needed, launch a single-node or multi-GPU instance and choose a GPU type.
Shared long-running training
When several people share the same GPUs, combine multiple nodes into a schedulable resource pool that the team can allocate as needed.
Online inference
For latency-sensitive traffic that cannot be interrupted, shared instances are billed by the minute while dedicated instances reserve fixed resources.
FEATURE｜ Service details and specifications
Keep data between sessions and change GPU types without rebuilding
Data and environments persist beyond an instance's lifecycle. Resume from the previous state after changing GPU types, stopping, or restarting, while public models and private data remain separate.
Keep data after shutdown
Data is mounted when the instance starts and retained after shutdown. There is no need to upload it again each time.
Save and restore environment snapshots
A snapshot preserves the entire environment state. Resume directly after changing GPU types or interrupting a run.
Public access · Private storage
Access public models and datasets directly. Private data stays in the user's account space, with high-performance storage available for larger capacities.
FEATURE｜ Service details and specifications
Self-service provisioning billed by actual usage
Register online, choose a GPU type, and launch an instance without signing a contract. Usage is prepaid and billed by the minute, with no monthly fee or minimum commitment. Current spending is visible at any time.
1minute
Minimum billing unit; billing stops after shutdown
0monthly fees
No monthly fee or minimum usage commitment; billing stops at shutdown
BENEFIT｜ Benefits
Upgrade the GPU and reduce training time from 8.5 hours to 70 minutes
A higher hourly price can still lower total per-minute costs when the job finishes much sooner, freeing that time for the next experiment.
| Training task | GPU class | Runtime |
|---|---|---|
| Image classification | Entry-level workstation GPU | 8.5 hours |
| Same task | Higher-end accelerator from the same generation | 70 minutes |
| Object detection | Top-tier data center accelerator | 25 minutes |
PRICE｜ Pricing and fees
Fund one team account and allocate quotas by role
Manage multiple members under one account. Add funds centrally, assign quotas, and avoid separate payments. Administrators can also limit the GPU types and images available to each member.
| Role | Permissions | Maximum members |
|---|---|---|
| Creator | Creates and owns the team, with full permissions | 1 |
| Administrator | Create accounts, allocate quotas, manage resource access, and release instances | Up to 2 |
| User | Launch and use instances within assigned quotas and resource permissions | Up to 100 |
FAQ
How does Glows.ai price its GPU Cloud service?
Pricing is transparent, with support for H100, H200, and RTX5090 GPUs, per-minute billing, and automatic scaling. Contact our sales team for details.
Are plans available for schools or research organizations?
Yes. Education plan quotas, GPU types, and teaching support are tailored to the organization's size. Contact our sales team to discuss the options.
Do I need to rebuild the environment when changing GPU types?
No. Data is mounted when the instance starts and retained after shutdown, so you do not need to upload it again. Snapshots preserve the entire environment and can be restored to a new instance, letting you resume work after changing GPU types or interrupting a run.
Which GPUs should I use for small and large models?
It depends on the size of the model and dataset. Workstation GPUs are sufficient for experimental training with smaller models. Large models and multi-GPU parallel training require the memory capacity and interconnect bandwidth of higher-end servers.
Ready to launch your first GPU instance?
Tell us about your team size and use case, and we will recommend a plan and quota configuration.