Operated 24/7 from Limassol and Auckland.
Managed operations from $1,500 per GPU node / month. We run updates, monitoring and on-call for your installation, with a 99.95% uptime SLA and 5 min incident response.
12 monitors. Enter opens one, Escape closes it. Every figure on the wall is Demo or Sample.
- Every node watched24/7
- 99.95%uptime SLA
- 5 minincident response
- HandoffLimassol, Auckland
- Model updatesincluded
- Driver updatesincluded
- Response timewatched live
- Batch loadpeak marked
- Request logper endpoint
- Alertsto on-call
- GPU memoryper node
- from $1,500per GPU node / mo
What we watch on your nodes.
The wall above is a simulation, and each panel says whether it is Demo or Sample. The text says what the monitor covers in a real installation.
GPU nodes
DemoEach GPU node under contract is a line on the wall. We watch it around the clock, and a node that drops out pages the on-call engineer.
Uptime SLA
SampleThe SLA is 99.95% uptime. Over a 30-day month that leaves 21.6 minutes of downtime before the commitment is missed. The tile shows a sample month with one short incident.
Incident response
DemoAn incident gets a human response in 5 minutes, at any hour. The ring fills toward 5:00 and stops when an engineer acknowledges.
On-call handoff
SampleLimassol and Auckland pass the on-call rota to each other, which is how coverage reaches 24/7. The map is schematic, not geographic.
Model updates
DemoModel updates are part of the subscription. We apply them to your nodes, so a new release of the open-weight model you run does not become a project for your team.
Driver updates
DemoGPU driver updates are included as well. The sequence on the tile is a demonstration, not your maintenance window.
Response time
DemoWe watch how fast the endpoint answers. A slow endpoint is an incident before it turns into a complaint.
Batch load
DemoBatch load shows how hard the GPUs are working. The amber bar marks the peak in the window.
Request log
DemoThe log tail shows requests reaching the OpenAI-compatible endpoint inside your network: successes, and the 429 responses that signal saturation.
Alerts
DemoMonitoring raises alerts to the on-call engineer. On the radar each amber point is an alert, and the sweep shows which ones are new.
GPU memory
DemoMemory use on each GPU node against a limit. A node close to the line is a capacity conversation before it is an outage.
Cost
SampleManaged ops is from $1,500 per GPU node per month. The bars multiply that price by the number of nodes. It is a sample, not a quote.
The SLA in numbers.
These are the published figures. SLA tiers are part of the subscription, so ask an engineer which tier fits your node count.
4.2 min down17.4 min left
- 99.95%Uptime SLA
- Allows 21.6 minutes of downtime in a 30-day month, or 4.4 hours in a year.
- 5 minIncident response
- From the alert to an engineer working on it, at any hour.
- 24/7Coverage
- On-call rota shared between the Limassol and Auckland offices.
What managed ops includes.
- Model and driver updates, monitoring, on-call, SLA tiers.
- 99.95% uptime SLA and 5-minute incident response.
- 24/7 across Limassol and Auckland.
Managed operations takes over once your installation is live. The production deployment (from $25,000, one-time, hardware separate) puts it there.
from
$1,500
per GPU node / month
4GPU nodes
Estimate 4 × $1,500 from $6,000 per month
Incident response and coverage.
An incident gets a response in 5 min. The line is open 24/7, staffed from two offices on opposite sides of the world.
22 engineers, 100+ clients, enterprise infrastructure since 2000.
An UpSystems practice — enterprise infrastructure since 2000.
Tell us how many GPU nodes.
An engineer scopes managed operations with you, node by node, at from $1,500 per GPU node / month. Not live yet? The 72-hour pilot is the first step.