Frequently Asked Questions

Refunds are handled case by case. For monthly plans, you can shut down resources anytime and you only pay for what you’ve used so far that month. For 6 and 12-month plans, contact support. Partial refunds are possible if the plan hasn’t been heavily used. Free credits aren’t refundable but don’t expire until the 30-day window ends.

Stopped instances still incur storage charges for their attached volumes and any reserved public IPs, but you don’t pay for the compute (vCPU/RAM/GPU) while they’re stopped. To stop charges completely, terminate the instance and delete the attached volumes. The console shows the cost breakdown so you can see exactly what each idle resource is costing.

The 6-month plan is roughly 5% cheaper than monthly, and the 12-month plan is roughly 10% cheaper. The discount is built into the listed price, not applied at checkout. You can mix plans across resources. For example, run production GPUs on a 12-month plan for the discount, and run dev environments monthly so you can shut them down without commitment. There’s no penalty for canceling early on monthly plans.

Yes. New customers receive ₹20,000 in free credits (around $200 for international accounts), valid for 30 days. The credits work across all AceCloud products: GPU instances, compute, storage, Kubernetes, databases. So you can benchmark your actual workload before committing. There’s no credit card requirement to claim the credits.

Yes, but it’s off by default. Databases sit on a private network and are only reachable from your AceCloud VMs and Kubernetes clusters out of the box. If you need a BI tool, a developer laptop, or a third-party SaaS to connect from outside, you enable a public endpoint and restrict it with IP allowlisting and TLS. Most production setups keep databases private and connect through VPN or a bastion host instead.

For PostgreSQL and MySQL, logical replication is the standard path. You set up replication from your current database to the AceCloud instance, let it catch up, then cut over by switching the connection string. For smaller datasets, pg_dump and mysqldump work fine. The same approach moves you in from AWS RDS, Azure Database, GCP Cloud SQL, or a self-managed setup. For complex migrations, the migration team can plan the cutover to minimize downtime to seconds.

Yes. HA is an opt-in deployment option for PostgreSQL, MySQL, and MariaDB. It runs a primary plus a synchronous standby replica. If the primary fails, the standby is promoted automatically and your application reconnects through the same endpoint. HA roughly doubles the resource cost, which is why it’s optional. Most production workloads need it, most dev environments don’t.

Daily automated backups run by default and retention is configurable. Point-in-time recovery is available on PostgreSQL and MySQL, so you can restore to any moment inside the retention window. This is the feature that saves you when a bad migration runs in production or an accidental DELETE hits the wrong table. You can also trigger manual snapshots before risky changes.

Yes. New users get ₹20,000 (around $200) in free credits, valid for 30 days, no credit card required.

You can bring your existing tools. AceCloud uses standard APIs and protocols: S3-compatible storage, Kubernetes-native interfaces, standard PostgreSQL/MySQL endpoints, OpenStack-based compute. Terraform, Ansible, kubectl, Helm, the AWS CLI (for S3-compatible storage), boto3, and most other infrastructure tools work directly. There’s a console for click-driven setup, but nothing forces you to use it.

Yes. New users get ₹20,000 in free credits (about $200), valid for 30 days, with no credit card required to claim them. The credits work across every product: GPU instances, compute, storage, Kubernetes, databases, networking. The point is to let you actually run your workload and benchmark it before committing, not to gate you behind a sales call.

Support is available 24/7 via live chat in the console, email at [email protected], and phone (+91-789-789-0682 for India, +1-205-784-3930 for US). Unlike hyperscalers, you don’t need to buy a paid support tier to talk to a real engineer. Human support is included on all accounts. Response times depend on severity, with critical production issues handled in minutes.

Under 5 minutes from launch to SSH access. Pre-built images for PyTorch, TensorFlow, CUDA, and Triton come with the instance.

No. Monthly billing is the default and you can shut down whenever the project ends. The 6 and 12-month plans exist because they’re cheaper for teams with steady usage, not because they’re required. If your workload is genuinely short, contact sales before launching so they can suggest the cheapest fit.

GPU Cloud is one VM with one or more GPUs attached. GPU Clusters are multiple GPU nodes connected with high-speed networking for distributed training. Use Cloud for development, fine-tuning, and inference. Move to Clusters when a single node stops being enough for your training job.

Dedicated. Each GPU is passed through directly to your VM with full memory and full bandwidth. No time-slicing, no shared SMs, no noisy neighbours. If you specifically want to partition a single card across services, MIG is supported on the GPUs that have it (A100, H100, H200, RTX Pro 6000).

Yes, with a free HA control plane and 99.99% SLA. You only pay for worker nodes no per-cluster control plane fee.

No, bandwidth is unmetered on Kubernetes workloads. This is the line item where hyperscaler bills get noisy, especially for media APIs, image-serving backends, and public webhooks. If your traffic pattern is closer to CDN-scale, the team will talk it through during onboarding, but standard application traffic isn’t separately billed.

Flannel, Calico, and Canal are supported for networking. NGINX is offered as the default ingress controller during cluster creation. You can install any other ingress controller (Traefik, HAProxy, Istio Gateway) on top using Helm, the same as on any Kubernetes cluster.

Cluster Autoscaler is on by default and adds or removes worker nodes based on pending pod pressure, up to 1000 nodes per cluster. Inside the cluster, Horizontal Pod Autoscaler and Vertical Pod Autoscaler work the same way they do in any Kubernetes. You set min and max bounds per node group, so dev clusters can scale to zero overnight and production keeps a baseline.

Yes, and the migration service is included at no extra cost. Most migrations cut over with minutes of downtime, not hours.

The migration team stays available during the post-cutover period, typically 2–4 weeks, to handle any issues. The standard practice is to keep the source environment running for a buffer period after cutover so you can roll back if something unexpected appears. Most issues come down to configuration drift (a hardcoded IP, a forgotten cron job, an external integration’s allowlist), and the team works through them with you.

Smaller datasets (under a few TB) move over the internet using standard transfer tools. Larger datasets use direct replication from your existing storage to AceCloud’s storage. For very large datasets where bandwidth would take too long, the team can arrange physical drive shipment, similar to AWS Snowball. The choice depends on data size, your available bandwidth, and how soon you need the migration complete.

Yes. AceCloud has a dedicated VMware-to-AceCloud migration path that moves VMs without requiring you to rebuild them. The team handles conversion, network mapping, and validation. This is the most common migration request from enterprises moving off on-prem VMware or VMware Cloud on AWS, especially after the Broadcom licensing changes.

Yes. AceCloud CDN caches your static content at edge locations to reduce origin load and improve page load times for users far from your primary region. It’s a separate product from the core network, billed by traffic served, and integrates with object storage so you can serve files from a bucket directly through the CDN without re-architecting.

Yes, through Floating IPs. A Floating IP is a public IPv4 address that you can attach to any VM and reassign to a different VM whenever you want. This is useful for failover (swap the IP to a backup VM during an outage), zero-downtime migrations, and DNS records that need a stable address. The IP belongs to your account, not the VM.

DDoS protection is included on all AceCloud network traffic at no additional cost. The platform absorbs volumetric attacks at the network edge before they reach your VMs. For applications that need extra application-layer protection (Layer 7 attacks targeting login pages or APIs), the Firewall-as-a-Service product adds WAF-style filtering. Volumetric protection is automatic and always on.

AceCloud offers managed Layer 4 (TCP/UDP) and Layer 7 (HTTP/HTTPS) load balancers. Layer 4 is for raw TCP traffic: databases, custom protocols, gaming backends. Layer 7 handles HTTP-aware routing, path-based rules, host-based rules, and SSL termination for web applications. Health checks, sticky sessions, and certificate management are built in.

AceCloud has an incident response process that includes detection, containment, customer notification, and remediation. If an incident affects your data or resources, you’ll be notified through the contact channels on your account, with details of what happened and what action is needed. Post-incident reports are shared with affected enterprise customers. For prevention, the platform runs continuous monitoring, vulnerability scanning, and patch management as part of the ISO 27001 program.

By default, no AceCloud employee can read your data. Storage volumes and object buckets are accessible only through your credentials. Operations staff can see metadata (resource names, sizes, usage metrics) for billing and support, but not the contents of your files or databases. If you open a support ticket and explicitly share access, that’s logged and revocable. There’s no backdoor access path used for support.

Your data lives in the region you select when you provision the resource. Indian regions store data in India, US region stores it in the US. Data doesn’t move between regions unless you explicitly set up cross-region replication or backup. This matters for data residency rules like India’s DPDP Act, GDPR for European customers, and HIPAA for US healthcare data.

Data is encrypted at rest on storage volumes and in transit using TLS. Block Storage and Object Storage both support encryption by default, and Object Storage uses server-side encryption with AceCloud-managed keys. For customers who need to control their own keys, customer-managed encryption (BYOK) is available on request. Database backups are encrypted before they leave the database host.

Volume Snapshots are point-in-time copies of a Block Storage volume, used for quick rollback before risky changes. Volume Backups are longer-term, stored separately from the original, and used for disaster recovery. Both can be scheduled or triggered manually. Restoring a snapshot is faster than restoring a backup, but a backup survives the loss of the original volume.

Block Storage for anything a VM mounts directly: databases, OS disks, application files. Object Storage for files your app reads and writes by URL: media, backups, ML datasets, logs. Glacier for data you need to keep but rarely touch: compliance archives, old backups, raw footage. The rule of thumb: how often will you read it? Daily means Block, occasionally means Object, almost never means Glacier.

Block Storage on AceCloud runs on NVMe with PCIe Gen5 attachment, delivering up to 1.2 million IOPS on the highest tiers. That’s enough for production databases, high-throughput logging, and GPU training datasets that need to keep up with the GPU. Lower-tier volumes are still SSD-backed, not spinning disk.

No. There are zero egress fees on AceCloud storage. You can read your data out as often as needed without bandwidth charges piling up. This is the line item where AWS S3 bills get expensive: at $0.09 per GB egress on AWS, pulling 10 TB out costs around $900. On AceCloud, the same pull costs nothing extra.