Siam AI Service Level Agreement (SLA)
Definitions
Quantity
The number of GPUs guaranteed by the service provider to always be available for service (e.g., if a user rents a 1K Cluster, the provider guarantees at least 1,000 GPU cards will be available).
SLA
Service Level Agreement is the service level agreed upon in the contract. If the service provider fails to meet the agreed levels, monthly availability is calculated using designated formulas and returned to the user as a monthly service fee discount or monthly service extension (Compensate Hours).
SLA Credit
A discount or monthly service extension credited when service levels fall below contractual terms (e.g., if the SLA drops below the agreed threshold, the difference is calculated and returned as a rental discount or monthly service extension (Compensate Hours).
Monthly Downtime
The total cumulative duration during which the user is unable to use the service throughout the month, resetting on the 1st day of every month.
Response Time
The elapsed time from when the system receives a failure alert or notification until the technical support team officially acknowledges and begins investigating the issue.
Incident
An unintended or unplanned event that disrupts service or degrades service quality.
Security Incident
An unintended cyber threat event caused by external malicious actors and/or internal personnel that impacts service operations and potentially escalates to system outages and organizational reputational damage.
SLA Calculation Principles
GPU Service Level Agreement
For GPU services, service level guarantees are based on total cumulative Monthly Downtime across all GPUs per month.
| SLA | SLA Credit | Monthly Downtime |
|---|---|---|
| 99.0% | Guarantee | 7h 18m 17s |
| 98.0%–99.0% | 5% | 14h 36m 35s |
| 95.0%–98.0% | 10% | 1d 12h 31m 27s |
| Less than 95.0% | 15% | > 1d 12h 31m 27s |
Demarcation Point
SLA covers strictly only at the Hardware, BIOS/Firmware, Host Driver, and GPU Health levels.
Exclusions
- User OS & Software Stack Layer Exclusion:
Excludes issues arising from user PyTorch/CUDA code, Model Out-Of-Memory (OOM), custom container crashes, or user configuration errors.
- Thermal Throttling Threshold Exclusion:
Performance throttling due to workloads exceeding standard hardware specs, user modifications to software/drivers/BIOS, or temporary cooling glitches is excluded if temperatures are restored within contractual limits without system crashes. Complete cooling failure causing system crashes/shutdowns remains covered
- Scheduled Maintenance Exclusion:
Power-offs for firmware updates (e.g., GPU BIOS, NVSwitch FW, InfiniBand Subnet Manager) notified at least 7 days in advance during off-peak hours are excluded from Monthly Downtime
Data Center Facility Infrastructure Availability
Facility Infrastructure Availability
Data center facility infrastructure availability guarantees follow the Data Center Operator’s Tier Level
| SLA | SLA Credit | Monthly Downtime |
|---|---|---|
| 99.99% | Guaranteed by DC | 4m 23s |
| 99.9%–99.99% | 5% | 43m 50s |
| 99.5%–99.9% | 10% | 7h 18m 17s |
| Less than 99.5% | 15% | > 7h 18m 17s |
- Power Availability: Guaranteed continuous power supply from primary and backup sources (UPS & Generator) to PDU feeds at rack level.
- Cooling Availability: Guaranteed continuous cooling and airflow according to DC operator standards.
Environmental Control Standards
Environmental control standards within the Server Hall shall strictly comply with the conditions and specifications established by the respective Data Center Operator.
- Temperature & Relative Humidity: Maintained within ASHRAE operating boundaries.
- Unit/Rack Cooling Supply: Maintained according to DC operator and OEM pressure/temperature specifications.
Facility Outage Definition
Facility Downtime begins upon any of the following events.
- Total power outage on all assigned rack feeds exceeding DC operator SLA Level.
- Total cooling failure causing server thermal shutdown.
- Critical physical events (e.g., fire or fire suppression activation).
Physical Security & Access
- Physical Security: Governed by access control, CCTV recording, and security personnel standards of the DC operator.
- Site Access Authorization: On-site work permits and emergency access follow DC operator SOP procedures and approval timelines.
Exclusions
Following events are excluded from Monthly Facility Downtime.
- Scheduled facility maintenance notified in advance.
- Power over-current/allocation overload or use of unapproved equipment.
- Force majeure or external outages beyond DC contractual scope.
CPU and Storage Server Service Level Agreement
System Availability and Uptime
CPU and storage systems are guaranteed at 99.5% monthly availability
| SLA | SLA Credit | Monthly Downtime |
|---|---|---|
| 99.5% | Guarantee | 3h 39m 8.7s |
| 99.0%–99.5% | 5% | 7h 18m 17s |
| 98.0%–99.0% | 10% | 14h 36m 35s |
| Less than 98.0% | 15% | > 14h 36m 35s |
Demarcation Point
- CPU Infrastructure: Covers hardware integrity of CPU, RAM, Motherboard, Power Supply Units (PSU), and server-level cooling.
- Storage Infrastructure: Covers functionality of connected disk arrays (Local NVMe / SSD / HDD Array or Distributed Storage Volumes) for Read/Write I/O operations.
Downtime Definition
Hardware outage counting begins upon:
- Unplanned server shutdown/crash or OS boot failure caused by hardware failure.
- Uncorrectable ECC Memory Errors or CPU Machine Check Exceptions (MCE) causing system halts.
- Total storage volume unavailability or complete I/O blockage.
Exclusions
Following events are excluded from Monthly Facility Downtime.
- Partial disk degradation in redundant arrays (RAID 1/5/6/10 or Distributed Storage) where normal Read/Write operations continue without I/O interruption.
- Data loss, corruption, or OS crashes/kernel panics caused by user-level software, commands, or applications.
- Scheduled maintenance notified at least 7 business days in advance.
Public Internet & VPN Connectivity Service Level Agreement
Internet Service Availability
Stability and availability of Public Internet and VPN connectivity directly depend on Internet Service Provider (ISP) SLA terms.
Demarcation Point
Demarcation ends at the provider Border Router Gateway Handoff Point. Excludes ISP network outages, latency, fiber loss, or client-side network issues.
Downtime Definition
Outage counting begins upon complete inability to transmit/receive data via assigned Public IP addresses originating strictly within the provider’s demarcation point.
Exclusions
Following Events are excluded from provider SLA liability:
ISP network failures, fiber loss, BGP routing/peering flaps, core outages, or scheduled ISP maintenance.
Security incidents, DDoS attacks, or temporary Null-Route IP actions taken to protect shared infrastructure.
Misconfiguration of IP addresses, routing tables, VPN settings, or firewall rules by the user.
Client-side ISP issues, client firewall port blocking, or client VPN software errors.
The service provider reserves the right to exclude service disruptions, access restrictions, or connection rejections from Service Downtime calculations in all cases if they result from the user’s own IP address usage behavior, including:
Third-party IP blocking/blacklisting by destination service providers, websites, or external platforms due to misuse or violation of those third-party terms and conditions.
Automated Security Threat Filtering triggered by user traffic patterns, port scanning, or misuse.
Temporary service suspensions resulting from scheduled maintenance operations, network equipment firmware updates (e.g., Firewalls, Network Switches), or planned system infrastructure upgrades that require a system or device graceful shutdown and impact internet connectivity shall be excluded from Monthly Downtime calculations under this SLA, subject to the following conditions:
Prior Notice: The service provider must provide written notice detailing the work plan and affected schedule to the user via the ticket system or email at least 7 days in advance prior to commencing operations.
Maintenance Window: Such operations must be scheduled during off-peak hours according to the provider’s standards to minimize the impact on user operations.
Notice: The Service Provider reserves the right to modify the SLA conditions on the website. The user must be notified in writing at least 14 days in advance.
Note: The full SLA details are subject to the terms and conditions specified in each individual service agreement.