This week All our GPUs are sold at cost price — zero margin on the server. See the cost sheets

This week Every GPU sold at cost price
Sign in Get started

Service status

All systems operational.

Every server is probed every 60 seconds from two regions, and every component below is checked the same way. These are the measurements that pay the SLA credits, so this page and the ledger can never disagree. Machine-readable at /status.json, subscribable as an Atom feed.

All systems operationalLast check 00:11:17 UTC · 45 s ago · next in 60 s · 99.9% SLA per server, credited ×10 automatically
How we measure
  • 4regions, 18 monitored components
  • 1,440probes per server per day, from two regions
  • 10incidents in the last 90 days, all regions
  • 2outages credited automatically, 90 days
  • 3maintenance windows announced, next 30 days

Regions, last 90 days

  • ReykjavíkREY · EU-North · since Mar 2024
    operational
    • Network & transit
    • Power & cooling
    • GPU nodes
    • Provisioning
    90 days ago99.997%today
    Probe RTT from AMS, 24 h28.4 ms
    27.4–32.1 ms4 min of outage in 90 days, credited

    Last incident: Core switch supervisor failover in Reykjavík, 5 Sep 2026.

  • HelsinkiHEL · EU-Nordic · since Feb 2025
    operational
    • Network & transit
    • Power & cooling
    • GPU nodes
    • Provisioning
    90 days ago100.00%today
    Probe RTT from AMS, 24 h26.6 ms
    26.0–30.9 msNo outage in 90 days

    Last incident: Hardware fault on one GPU node in Helsinki: card replaced, 20 Aug 2026.

  • AmsterdamAMS · EU-West · since Jun 2025
    operational
    • Network & transit
    • Power & cooling
    • GPU nodes
    • Provisioning
    90 days ago99.992%today
    Probe RTT from HEL, 24 h26.6 ms
    26.0–34.3 ms11 min of outage in 90 days, credited

    Last incident: NVLink fabric manager restart on 4 nodes in Amsterdam, 5 Sep 2026.

  • SingaporeSIN · APAC · since Feb 2026
    operational
    • Network & transit
    • Power & cooling
    • GPU nodes
    • Provisioning
    90 days ago100.00%today
    Probe RTT from AMS, 24 h157.8 ms
    155.6–187.5 msNo outage in 90 days

    Last incident: Transit provider session flap in Singapore, 30 Jun 2026.

Shared services

  • Console & APIdedigpu.com, /app and /api/v1
    operational
  • Crypto top-upsInvoice creation and on-chain confirmation
    operational

Scheduled maintenance

Announced at least 72 hours ahead.

Windows are at night in the region's local time, one device at a time, under four hours a month. Affected tenants see the same notice in the console.

  1. Thu 22 Oct01:00–01:57 UTC
    NVLink firmware update on H200 nodes in HelsinkiHelsinki · GPU nodes · Each node reboots once; no re-image, data untouched.Announced 14 Oct
  2. Thu 22 Oct01:15–01:45 UTC
    NVLink firmware update on B200 nodes in ReykjavíkReykjavík · GPU nodes · Each node reboots once; no re-image, data untouched.Announced 14 Oct
  3. Tue 3 Nov01:45–02:25 UTC
    Switch software upgrade in ReykjavíkReykjavík · Network & transit · Up to 60 seconds of packet loss per server during the failover, twice.Announced 27 Oct

Monthly availability

Per region, per calendar month.

Outage minutes divided by the minutes in the month. The current month is to date.

MonthREYHELAMSSIN
Oct 2026100.00%100.00%100.00%100.00%
Sep 202699.991%100.00%99.975%100.00%
Aug 2026100.00%100.00%100.00%100.00%
Jul 2026100.00%99.996%100.00%100.00%
Jun 2026100.00%100.00%100.00%99.984%
May 2026100.00%100.00%100.00%100.00%
Apr 2026100.00%100.00%100.00%100.00%
Mar 2026100.00%99.962%100.00%100.00%
Feb 2026100.00%99.968%100.00%100.00%
Jan 2026100.00%100.00%100.00%—
Dec 2025100.00%100.00%100.00%—
Nov 2025100.00%100.00%99.988%—

A month below 99.9% for a region means every server there received credits for that month, automatically. Regions show “—” before they opened.

History, last 90 days

Every event longer than a minute, with its timeline.

Outages count against the SLA and are credited; degradations and announced maintenance are listed for the record. Events longer than half an hour carry a post-incident review. Older events are in the feed.

  1. degraded

    Payment processor maintenance: new invoices paused

    2026-10-13 21:51 UTC · 37 min · Crypto top-ups

    New top-up invoices could not be created for a short time; open invoices kept confirming.

    1. investigating2026-10-13 21:51

      Our payment processor is in maintenance; creating a new invoice fails with a clear message. Open invoices and confirmations continue.

    2. identified2026-10-13 22:04

      Processor back; invoice creation working from every probe.

    3. monitoring2026-10-13 22:18

      Watching for an hour.

    4. resolved2026-10-13 22:28

      Resolved.

  2. degraded

    Elevated latency on the console and API

    2026-10-02 21:13 UTC · 23 min · Console & API

    Console pages and API calls slower than usual; servers unaffected.

    1. investigating2026-10-02 21:13

      Console and API responses are taking one to three seconds instead of under 200 ms.

    2. identified2026-10-02 21:21

      A database index rebuild is contending with live traffic. The rebuild is being throttled.

    3. monitoring2026-10-02 21:30

      Latency back to normal; the rebuild finishes in the background.

    4. resolved2026-10-02 21:36

      Resolved. Servers and billing were not affected.

  3. maintenance

    Provisioning host OS patching in Reykjavík

    2026-09-30 01:30 UTC · 32 min · Reykjavík · Provisioning

    New orders in Reykjavík queue for the window; running servers unaffected.

    1. scheduled2026-09-26 13:30

      Announced for 30 September, 01:30–02:02 UTC. Affected tenants notified in the console. New orders in Reykjavík queue for the window; running servers unaffected.

    2. in progress2026-09-30 01:30

      Window open. Work proceeds one device at a time.

    3. completed2026-09-30 02:02

      Completed in 32 minutes, inside the announced window. Announced maintenance under the monthly allowance: no SLA impact.

  4. maintenance

    Payment processor upgrade

    2026-09-29 02:15 UTC · 25 min · Crypto top-ups

    New invoices paused for the window; open invoices keep confirming.

    1. scheduled2026-09-21 18:15

      Announced for 29 September, 02:15–02:40 UTC. Affected tenants notified in the console. New invoices paused for the window; open invoices keep confirming.

    2. in progress2026-09-29 02:15

      Window open. Work proceeds one device at a time.

    3. completed2026-09-29 02:40

      Completed in 25 minutes, inside the announced window. Announced maintenance under the monthly allowance: no SLA impact.

  5. maintenance

    Transit port migration in Singapore

    2026-09-15 18:45 UTC · 23 min · Singapore · Network & transit

    No downtime expected: traffic drains to the second provider first.

    1. scheduled2026-09-12 05:45

      Announced for 15 September, 18:45–19:08 UTC. Affected tenants notified in the console. No downtime expected: traffic drains to the second provider first.

    2. in progress2026-09-15 18:45

      Window open. Work proceeds one device at a time.

    3. completed2026-09-15 19:08

      Completed in 23 minutes, inside the announced window. Announced maintenance under the monthly allowance: no SLA impact.

  6. degraded

    Intermittent errors on the console

    2026-09-08 13:37 UTC · 12 min · Console & API

    Some console requests failed for a few minutes; the API and servers were unaffected.

    1. investigating2026-09-08 13:37

      A share of console requests return an error page after a deployment.

    2. identified2026-09-08 13:41

      Deployment rolled back.

    3. resolved2026-09-08 13:49

      Resolved. The change ships again after review.

  7. outage

    Core switch supervisor failover in Reykjavík

    2026-09-05 20:39 UTC · 4 min · Reykjavík · Network & transit

    Brief loss of connectivity for every server in Reykjavík during the failover.

    1. investigating2026-09-05 20:39

      The active supervisor of a core switch in Reykjavík rebooted unexpectedly; the standby took over.

    2. resolved2026-09-05 20:43

      Resolved. 4 minutes of downtime credited automatically to every server in Reykjavík. The faulty supervisor module was replaced the same day.

  8. outage

    NVLink fabric manager restart on 4 nodes in Amsterdam

    2026-09-05 10:17 UTC · 11 min · Amsterdam · GPU nodes

    Tenants on 4 SXM nodes in Amsterdam lost NVLink for the restart; single-card servers unaffected.

    1. investigating2026-09-05 10:17

      The fabric manager on 4 HGX nodes in Amsterdam stopped responding; NVLink links reported down.

    2. identified2026-09-05 10:21

      Service restarted on the affected nodes; links renegotiating.

    3. resolved2026-09-05 10:28

      Resolved. The affected servers were credited automatically.

    Post-incident review

    The fabric manager service on 4 nodes hit a known deadlock after a driver update and stopped answering; NVLink went down for 11 minutes on those nodes. The affected tenants were credited. The fix was to pin the fabric manager to the driver version it was validated with, and to add a watchdog that restarts it within 60 seconds.

  9. maintenance

    Annual UPS battery test in Reykjavík

    2026-09-02 02:00 UTC · 61 min · Reykjavík · Power & cooling

    None expected. Servers stay on utility power throughout.

    1. scheduled2026-08-29 09:00

      Announced for 2 September, 02:00–03:01 UTC. Affected tenants notified in the console. None expected. Servers stay on utility power throughout.

    2. in progress2026-09-02 02:00

      Window open. Work proceeds one device at a time.

    3. completed2026-09-02 03:01

      Completed in 61 minutes, inside the announced window. Announced maintenance under the monthly allowance: no SLA impact.

  10. degraded

    Elevated latency on the console and API

    2026-08-29 00:07 UTC · 18 min · Console & API

    Console pages and API calls slower than usual; servers unaffected.

    1. investigating2026-08-29 00:07

      Console and API responses are taking one to three seconds instead of under 200 ms.

    2. identified2026-08-29 00:13

      A database index rebuild is contending with live traffic. The rebuild is being throttled.

    3. resolved2026-08-29 00:25

      Resolved. Servers and billing were not affected.

  11. maintenance

    Console deployment

    2026-08-25 02:00 UTC · 10 min · Console & API

    Console read-only for under a minute; servers and API unaffected.

    1. scheduled2026-08-21 13:00

      Announced for 25 August, 02:00–02:10 UTC. Affected tenants notified in the console. Console read-only for under a minute; servers and API unaffected.

    2. in progress2026-08-25 02:00

      Window open. Work proceeds one device at a time.

    3. completed2026-08-25 02:10

      Completed in 10 minutes, inside the announced window. Announced maintenance under the monthly allowance: no SLA impact.

  12. maintenance

    Transit port migration in Amsterdam

    2026-08-25 01:30 UTC · 26 min · Amsterdam · Network & transit

    No downtime expected: traffic drains to the second provider first.

    1. scheduled2026-08-19 15:30

      Announced for 25 August, 01:30–01:56 UTC. Affected tenants notified in the console. No downtime expected: traffic drains to the second provider first.

    2. in progress2026-08-25 01:30

      Window open. Work proceeds one device at a time.

    3. completed2026-08-25 01:56

      Completed in 26 minutes, inside the announced window. Announced maintenance under the monthly allowance: no SLA impact.

  13. degraded

    Hardware fault on one GPU node in Helsinki: card replaced

    2026-08-20 07:11 UTC · 181 min · Helsinki · GPU nodes

    One server offline while a failed card was swapped from local spares; the tenant was credited.

    1. investigating2026-08-20 07:11

      Monitoring flagged an uncorrectable memory error on one card of a node in Helsinki; the server was taken offline with the tenant informed.

    2. identified2026-08-20 08:14

      The card is being swapped from the spares shelf in Helsinki.

    3. monitoring2026-08-20 09:21

      Card replaced, node back online and burn-in passed.

    4. resolved2026-08-20 10:12

      Resolved. Downtime credited ten times over to the affected server. No other tenant was involved.

  14. degraded

    Confirmation delays on Litecoin top-ups

    2026-08-10 09:58 UTC · 42 min · Crypto top-ups

    Top-ups in Litecoin took longer than usual to confirm on chain; other coins unaffected.

    1. investigating2026-08-10 09:58

      Litecoin transactions are confirming slowly because of congestion on the network itself. Invoices stay valid and will credit once confirmed.

    2. identified2026-08-10 10:13

      Congestion easing; confirmations arriving with a delay of about an hour.

    3. monitoring2026-08-10 10:28

      Backlog cleared; new payments confirm at the usual pace.

    4. resolved2026-08-10 10:40

      Resolved. Every delayed payment was credited at the invoice rate.

  15. degraded

    Payment processor maintenance: new invoices paused

    2026-08-02 22:35 UTC · 40 min · Crypto top-ups

    New top-up invoices could not be created for a short time; open invoices kept confirming.

    1. investigating2026-08-02 22:35

      Our payment processor is in maintenance; creating a new invoice fails with a clear message. Open invoices and confirmations continue.

    2. identified2026-08-02 22:49

      Processor back; invoice creation working from every probe.

    3. monitoring2026-08-02 23:04

      Watching for an hour.

    4. resolved2026-08-02 23:15

      Resolved.

  16. maintenance

    Provisioning host OS patching in Reykjavík

    2026-07-28 01:45 UTC · 30 min · Reykjavík · Provisioning

    New orders in Reykjavík queue for the window; running servers unaffected.

    1. scheduled2026-07-24 08:45

      Announced for 28 July, 01:45–02:15 UTC. Affected tenants notified in the console. New orders in Reykjavík queue for the window; running servers unaffected.

    2. in progress2026-07-28 01:45

      Window open. Work proceeds one device at a time.

    3. completed2026-07-28 02:15

      Completed in 30 minutes, inside the announced window. Announced maintenance under the monthly allowance: no SLA impact.

  17. maintenance

    Provisioning host OS patching in Amsterdam

    2026-07-28 01:45 UTC · 50 min · Amsterdam · Provisioning

    New orders in Amsterdam queue for the window; running servers unaffected.

    1. scheduled2026-07-24 09:45

      Announced for 28 July, 01:45–02:35 UTC. Affected tenants notified in the console. New orders in Amsterdam queue for the window; running servers unaffected.

    2. in progress2026-07-28 01:45

      Window open. Work proceeds one device at a time.

    3. completed2026-07-28 02:35

      Completed in 50 minutes, inside the announced window. Announced maintenance under the monthly allowance: no SLA impact.

  18. degraded

    Cooling pump swapped on a loop in Helsinki

    2026-07-25 11:50 UTC · 70 min · Helsinki · Power & cooling

    None. Inlet temperatures stayed within range on the redundant pump.

    1. investigating2026-07-25 11:50

      A pump on one cooling loop in Helsinki reported a bearing fault; the redundant pump took the load.

    2. identified2026-07-25 12:15

      Facility engineers are replacing the pump. Inlet temperatures are within range.

    3. monitoring2026-07-25 12:40

      Pump replaced, loop back to N+1.

    4. resolved2026-07-25 13:00

      Closed. No thermal event, no downtime.

  19. maintenance

    Provisioning host OS patching in Amsterdam

    2026-07-22 01:00 UTC · 37 min · Amsterdam · Provisioning

    New orders in Amsterdam queue for the window; running servers unaffected.

    1. scheduled2026-07-18 08:00

      Announced for 22 July, 01:00–01:37 UTC. Affected tenants notified in the console. New orders in Amsterdam queue for the window; running servers unaffected.

    2. in progress2026-07-22 01:00

      Window open. Work proceeds one device at a time.

    3. completed2026-07-22 01:37

      Completed in 37 minutes, inside the announced window. Announced maintenance under the monthly allowance: no SLA impact.

Downtime starts from the first minute in which both probes fail two consecutive checks and ends at the first successful check; hardware faults count from detection. Announced maintenance up to four hours a month is excluded. Every minute of outage beyond the monthly allowance is credited ten times over, to the balance, without a ticket. The full service level agreement.

Rent a server on a network you can audit.

Same probes, same numbers, same credits for everyone. One month at a time, at cost.

See the servers Read the SLA