HS-SU / On-site service
On-site support,
after install.
A Helmstone workstation ships with a written support contract and a published SLA — not a “contact your reseller” line. This page is what the buyer reads before they sign the quote, so they know what happens after the unit is on its pallet.
Last updated . Reading time roughly four minutes.
HS-SU-01 / Coverage
What the contract
actually includes.
Three buckets — once a Helmstone engineer has signed the per-unit benchmark sheet in the box, the contract covers thermal re-checks, driver / firmware updates, and parts replacement against the SKU, on-site, in the buyer’s room.
The SLA window each bucket maps to sits in the response-window matrix below.
Coverage 01
On-site thermocouple checks against the 48-hour burn-in baseline, in the buyer’s room.
Within the first 30 days after install, a Helmstone engineer returns to the buyer’s site and re-clips thermocouples to the GPU die hotspot, VRAM, exhaust, and the frame inlet. We re-run the same workload loop for 90 minutes and compare against the per-unit signed benchmark sheet in the box. Any drift is escalated under through-put warranty, not billed hourly.
Coverage 02
Pinned, regression-tested, scheduled — never surprise rolls onto a live unit.
Each SKU ships against a pinned NVIDIA Studio driver set, and that set is what the 48-hour burn-in is regression-tested against. Updates land against the unit’s own SKU as a batch, then are qualified by the same bench rig before they are offered. Buyers are notified one week before a rollout reaches their machine, with the revert path attached.
Coverage 03
On-site swap, cross-shipment, or depot repair — whichever is fastest to spec.
GPU, host memory, storage, PSU, and chassis fans are all warranty-bound under the SLA class the buyer selected at quote time. On-site engineering hours are part of the contract, not a billed line item, and cross-shipment swaps are honored at no engineering charge when the depot cannot return a unit to spec within five business days.
HS-SU-1.5 / Support tiers
The three SKUs
a unit ships against.
Every Helmstone workstation arrives with the Standard tier — there is no bare warranty path. Priority and Mission-Critical are add-on tiers a buyer promotes at quote time, depending on the metro they’re in and how on-call they want their model loop to be.
The window each tier promises is the same numerical matrix as the response-window table below — this block names which tier moves each dial.
HS-SU-ST
The contract every Helmstone shipment ships with.
Coverage
Included with every shipment
Included
What the contract includes
- →Thermal validation follow-up on-site within the first 30 days.
- →Pinned, regression-tested driver / firmware batches — no surprise rolls.
- →Three-year on-site warranty on GPU, host memory, storage, PSU, fans.
- →On-site engineering hours included at no extra billable rate.
Response window
- P1 · metro
- Same-day remote, on-site within 24 business hours.
- P1 · non-metro
- 48-hour remote mitigation; on-site if mitigation fails.
- P2 · metro
- Next-business-day remote triage.
- P3 · metro
- Scheduled in the next qualified batch (weekly).
RMA / replacement
Five-business-day depot swap; cross-shipment on request.
HS-SU-PR
Same-day P1 in priority metros, second-touch on the calendar.
Coverage
Add-on · any SKU class
Add-on · quoted per metro
What the contract includes
- →P1 paging to a named on-call engineer, escalation ladder written down.
- →Pre-staged service parts in the buyer’s metro same-day.
- →Cross-shipment swap already authorised before the depot loop closes.
- →Annual thermal re-validation visit, included.
Response window
- P1 · metro
- On-site engineer dispatched within 4 business hours.
- P1 · non-metro
- 48-hour remote mitigation; on-site within 72 hours if it fails.
- P2 · metro
- Next-business-day remote triage; on-site within 24 hours if hardware.
- P3 · metro
- Schedule-bound to the buyer’s release window.
RMA / replacement
Five-business-day cross-shipment privilege, pre-authorised.
HS-SU-MC
Dedicated pager, named engineer, contracted on the calendar.
Coverage
Add-on · enterprise tier only
Add-on · quoted per site
What the contract includes
- →Named on-call engineer, SMS pager, 24/7 response across all priorities.
- →Quarterly burn-in re-bench, delivered with a signed throughput sheet.
- →Pre-positioned dual-spare kit on the buyer’s premises (optional).
- →Driver / firmware rollouts gate on buyer release window — never mid-job.
Response window
- P1 · metro
- On-site engineer on the buyer’s premises within 2 business hours.
- P1 · non-metro
- Within 6 business hours, plus direct engineering contact first.
- P2 · metro
- Same-business-day remote triage; on-site within 4 hours if hardware.
- P3 · metro
- Schedule-bound to the buyer’s release window, with named owner.
RMA / replacement
Cross-shipment within 48 hours, pre-authorised before depot triage.
HS-SU-02 / Response windows
Priority × metro.
What “on-site” actually means.
Same-day on-site for a thermal fault in a priority metro; 48-hour remote mitigation outside it. P2 and P3 windows are written so a buyer can quote them directly into a procurement request, without a sales call.
Priority metros are San Francisco Bay, NYC metropolitan, Austin, Greater Boston, and Greater Seattle — see the technician footprint below.
| Priority | Window |
|---|---|
Priority metroNon-metro | |
P1 Thermal fault, system down Thermal fault means any unrecoverable throttle event, sustained over-temperature alarm, or wall-meter power regression detected against the per-unit signed benchmark. | Priority metro Same-day, on-site engineer dispatched within 4 business hours. Non-metro 48-hour remote mitigation; on-site within 72 hours if mitigation fails. |
P2 Performance regression Performance regression covers any sustained-throughput shortfall vs the published per-SKU tok/s band, or any non-throttle wall-meter power drift greater than 10% across a 24-hour window. | Priority metro Next-business-day remote triage; on-site if root cause is hardware. Non-metro Three business-day remote triage; on-site enclosed with next engineer tour. |
P3 Driver / firmware request Driver or firmware requests are batched so the upgrade is regression-tested against the SKU before it reaches any live unit. Buyers are notified one week before the rollout touches their machine. | Priority metro Scheduled within the next qualified update batch (usually weekly). Non-metro Scheduled within the next qualified update batch (usually weekly). |
How P1 is defined
A P1 ticket is one of three things: an unrecoverable throttle event in nvidia-smi, a sustained over-temperature alarm against the per-unit signed benchmark, or a wall-meter power regression detected against the same sheet. Anything else routes into P2 unless the unit is system-down, in which case it still runs P1.
P3 tickets are batched so the upgrade is regression-tested against the SKU before it reaches the live unit — driver and firmware requests never roll onto a machine mid-job.
HS-SU-03 / Footprint
Where the on-site
engineer is already on the calendar.
Five priority metros carry an engineer on the calendar — no dispatch lag, P1 tickets turn into a truck on the road. Outside those metros the contract still covers, but the window collapses to remote mitigation first, on-site second.
See the metro map → /service-area
North America · Tier 1
6 engineers on the calendar
On-site engineer on the calendar, no dispatch lag.
Mon–Fri, 08:00–18:00 PT · 24/7 P1 paging
North America · Tier 1
5 engineers on the calendar
On-site engineer on the calendar, no dispatch lag.
Mon–Fri, 09:00–19:00 ET · 24/7 P1 paging
North America · Tier 1
3 engineers on the calendar
On-site engineer on the calendar, no dispatch lag.
Mon–Fri, 08:00–18:00 CT · 24/7 P1 paging
North America · Tier 1
4 engineers on the calendar
On-site engineer on the calendar, no dispatch lag.
Mon–Fri, 09:00–18:00 ET · 24/7 P1 paging
North America · Tier 1
3 engineers on the calendar
On-site engineer on the calendar, no dispatch lag.
Mon–Fri, 08:00–18:00 PT · 24/7 P1 paging
Worldwide · Tier 2
Direct engineering contact within four business hours
Direct engineering contact within four business hours; remote mitigation in 48 hours, then on-site within 72 hours if remote fails.
Business-day coverage, remotely + scheduled site visits
HS-SU-04 / SLA terms
The contract,
written so each line is quotable.
Seven clauses, the same shape as the signed quote but in the buyer-readable form published on the site. The print button at the top and bottom of this section renders the same terms to a single-page PDF, with the nav stripped out.
SLA clauses
Each clause is reproduced at the level of detail a buyer can quote into a procurement request. Last updated .
- Clause 01
Hardware warranty — three years, on-site
GPU, host memory, storage, PSU, and chassis fans are warranty-bound for three years from the per-unit signed benchmark date, with on-site engineering hours included at no additional billable rate.
- Clause 02
Throughput warranty — 30-day burn-in window
Any sustained-throughput shortfall vs the per-SKU published tok/s band during the first 30 days post-install is corrected under warranty. Buyers do not pay engineering hours for that re-bench.
- Clause 03
Escalation classes — P1 / P2 / P3
P1 covers thermal faults, unrecoverable throttle events, and wall-meter regressions vs the signed sheet; P2 covers any non-fatal performance regression vs the published band; P3 covers driver and firmware update requests, scheduled and batched.
- Clause 04
Cross-shipment privileges
If a unit cannot be returned to spec within five business days, the depot cross-ships a qualified replacement at no engineering charge, and the original unit is returned after the diagnostic loop finishes.
- Clause 05
Parts logistics — pre-staged against the SKU
Spare GPU, host memory, storage, PSU, and chassis-fan kits are pre-staged in the same metro as the buyer’s site. Common service parts are pre-positioned so the swap window is a single engineer visit, not a multi-day procurement.
- Clause 06
Driver / firmware governance
Driver and firmware updates roll against the SKU as a batch, then are regression-tested against the same workload loop before they reach any live unit. Buyers are notified one week before the rollout, with the revert path attached.
- Clause 07
Notice, suspension, and cure
Either party may suspend the SLA in writing with 30-day notice. Suspended units retain their warranty pass-through; cure periods for SLA breach do not exceed 30 business days from written notice, per the terms in the signed quote.
Save this section as the buyer-facing SLA summary. The signed quote is the authoritative version; this page reproduces the same terms in the language buyers read.
HS-SU-05 / Open a ticket
Support escalations
route straight to ops.
A support message is a signed ticket. The signed message below goes to the same ops mailbox a buyer quotes the SLA window against, and an engineer replies on the record. The form is the same one used for general inquiries — what makes the difference is what you put in the message body.
What to include
Workstation serial
The serial on the per-unit signed benchmark sheet — that is the unit’s identity in our depot. Without it, the ticket can only move on model class, which costs at least one round-trip.
Service metro
The metro we dispatch into — the on-site engineer’s calendar is pinned to that metro, and response windows collapse outside it.
Priority
P1 (thermal fault, system down), P2 (performance regression), or P3 (driver / firmware request). The priority pins the SLA window that applies.
What you observed
Wall-meter reading, GPU junction temperature, sustained tok/s, any throttle-state transitions in `nvidia-smi`, the workload running at the time, and the time the symptom first appeared.
Reading via email instead? helmstone-systems@polsia.app.
Open a support ticket
Tell us what you observed.
The four fields below go to ops. An engineer reads the message body in the order it appears — serial first, symptoms last — and replies against the priority and the service metro you name.