Full Fault Coverage, Remote to On-Site

Remote diagnosis, advance spare parts, on-site repair and scheduled inspection

24/7 Online Technical Support

Remote login diagnosis, boot and downtime handling, driver recognition, BMC/BIOS and firmware upgrade, CUDA environment support

Round-the-Clock Remote Fault Handling

Phone, online chat and enterprise IM channels, minute-level access for critical faults, unattended night-shift diagnostics

Advance Spare Parts

Same-day delivery in metro areas, next-day nationwide; advance parts carry manufacturer-equivalent warranty; faulty parts returned for repair

On-Site Repair

On-site replacement of GPU, motherboard, CPU, HDD, memory, NIC and mechanical parts, with commissioning and acceptance

Scheduled Inspection

Hardware status, data-center environment, dust removal and cabling, system and backup verification, with a printed inspection report

On-Site Team & Repair Base

Repair base within 5–10 km of the data center, 3–5 resident engineers, local fast repair of faulty parts

Three Service Tiers for Different Business Needs

Silver / Gold / Platinum — step-up response time and inspection frequency

Silver Maintenance

For non-critical workloads that can tolerate short downtime

5×8×NBDAdvance parts · next business day

24/7 online technical support
24/7 fault intake and response
Remote handling of level-2 faults
5×8×NBD advance spare parts
5×8×NBD on-site repair
Remote repair of level-1 critical faults
Scheduled on-site inspection

Ask about Silver

Recommended

Gold Maintenance

For production workloads requiring 24/7 continuity

7×24×NBDCritical faults · next business day

Everything in Silver
Immediate remote repair of level-1 critical faults
24/7 dedicated remote support
7×24×NBD advance spare parts
7×24×NBD on-site repair
Quarterly on-site inspection
Resident fast response and preventive maintenance

Ask about Gold

Platinum VIP

For core training clusters with the highest availability requirements

7×24×4HParts on site within 4 hours

Everything in Gold
7×24×4H advance spare parts
Monthly on-site inspection
Local spare parts warehouse
On-site repair base and resident engineers
Customized service support plan
Customized annual service report

Ask about Platinum

GPU Server Repair & Upgrade

From system inspection to chip-level repair, covering A / H / B series GPU platforms

Repair Capability

GPU not recognized / dropped card localized within 1 hour
19-point system inspection covering CPU, GPU, NVLink, NVSwitch, PCIe, power and storage
L1–L4 diagnostic tests with full traceability
Factory-grade precision repair environment and professional test fixtures
SXM module, NVSwitch board and GPU tray level repair

Upgrade & Optimization

GPU platform upgrade and system configuration fit
Memory, NVMe and NIC expansion with RAID configuration
BMC / BIOS / firmware upgrade
NVLink topology check and interconnect bandwidth tuning
Air / liquid cooling inspection and optimization

Common GPU Faults and Repairability

Fault Type	Share	Service Standard
GPU not recognized / dropped card	~55%	Repairable, warranty 1 year+
HBM fault / ECC error	~15%	Batch partial-disable repair, customer approval required
Cooling system fault	~5%	Repairable, warranty 1 year+
PCIe / NVLink interface fault	~5%	Repairable, warranty 1 year+
Firmware and configuration fault	~5%	Repairable, warranty 1 year+
Server-level associated fault	~10%	Repairable, warranty 1 year+
GPU core fault	~5%	Not repairable, chip replacement required

Four-Level Diagnostic Testing

L1

Seconds

Software deployment, hardware basic information, GPU recognition and driver status

L2

Under 2 minutes

PCIe bandwidth, memory and VRAM basic tests, NVLink speed

L3

Under 30 minutes

Hardware diagnostics, memory bandwidth, compute stress, power and topology tests

L4

1–2 hours

Extended compute and memory stress, pulse testing and stability verification

B300 & H200 Dedicated Maintenance and Repair

Platform-specific inspection, repair and spare-part coverage for high-power Blackwell Ultra and Hopper GPU systems

B300 (Blackwell Ultra)

8× B300 288GB HBM3E, ~2.1TB GPU memory per system, DGX B300 / HGX B300 platforms
~1,400W per GPU: dedicated inspection of power delivery, PDB, liquid-cooling loop and air cooling
NVLink 5 and NVSwitch link errors, link-down and dropped-GPU localization
HBM3E ECC errors, reduced memory capacity and GPU throttling handling
SXM module, GPU tray, cold plate and quick-disconnect replacement

H200 (Hopper)

8× H200 141GB HBM3E, 4.8TB/s per GPU, HGX H200 4-GPU and 8-GPU systems
Power and cooling fit for H200 SXM up to 700W and H200 NVL 600W
NVLink 900GB/s link degradation, NVSwitch and PCIe Gen5 fault diagnosis
H200 NVL dual-slot PCIe system maintenance and spare-part support
BMC / BIOS / firmware, driver and CUDA environment maintenance

B300 / H200 Service Scope and Delivery Standards

Service Item	B300 8-GPU System	H200 8-GPU System
Platforms	DGX B300 / HGX B300 8-GPU	HGX H200 8-GPU / H200 NVL
GPU memory	288GB HBM3E × 8, ~2.1TB per system	141GB HBM3E × 8, ~1.1TB per system
System power	~1,400W per GPU, liquid or high-capacity air cooling required	Up to 700W (SXM) / 600W (NVL) per GPU
Fault localization	GPU / NVLink / power within 1 hour	GPU / NVLink / HBM within 1 hour
Repair level	SXM module, GPU tray and NVSwitch board level	SXM module, GPU tray, NVSwitch and PCIe card level
Spare parts	Graded stock of systems, modules, cold plates, power supplies and mechanics	Graded stock of systems, modules, power supplies, NICs and mechanics
Service tiers	Silver 5×8×NBD / Gold 7×24×NBD / Platinum 7×24×4H	Silver 5×8×NBD / Gold 7×24×NBD / Platinum 7×24×4H
Repair warranty	1 year+ on repaired parts, extendable by contract	1 year+ on repaired parts, extendable by contract

Note: B300 draws ~1,400W per GPU, so adequate rack power and cooling are required; liquid-cooled systems are serviced by certified engineers. H200 systems have been leaving original warranty from 2026 — third-party maintenance preserves the original service standard.

Request a B300 / H200 Quote

Service Matrix

Response time, inspection frequency and value-added services across three tiers

Category	Service Item	Silver	Gold	Platinum
Online technical support	Basic remote fault diagnosis	24/7	24/7	24/7
	Operating system support	24/7	24/7	24/7
	Environment support	24/7	24/7	24/7
	Standard online service	24/7	24/7	24/7
24/7 remote fault handling	Fault intake and response	24/7	24/7	24/7
	Level-1 critical fault	None	24/7	24/7
	Level-2 routine fault	24/7	24/7	24/7
	24/7 remote assurance	None	24/7	24/7
Advance spare parts	Advance parts delivery	5×8×NBD	7×24×NBD	7×24×4H
Advance spare parts	Faulty parts collection	√	√	√
On-site maintenance	On-site emergency repair	5×8×NBD	7×24×NBD	7×24×4H
	Combined response mechanism	5×8×NBD	7×24×NBD	7×24×4H
	Scheduled inspection	None	Quarterly	Monthly
Resident service	Local parts warehouse	None	None	√
	Fast response	None	√	√
	Preventive maintenance	None	√	√
Plans & reports	Customized support plan	None	None	√
Plans & reports	Customized annual report	None	None	√

Spare Parts Management

Configuration ratio, ledger management and tiered dispatch

0.3%

Spare parts configured at 0.3% of cluster quantity, covering complete systems and GPU, motherboard, CPU, memory, HDD, NIC, PSU and mechanical parts

Billed at signing-day market price × 0.3% / unit / year

Ledger & Dispatch

Spare parts ledger with safety stock and monthly stocktaking
Ledger updated daily and uploaded automatically, customer portal for real-time query
Tiered dispatch: Silver 5×8×NBD, Gold 7×24×NBD, Platinum 7×24×4H engineers on site with parts

Fault Handling Process

01

Report

Phone / chat / enterprise IM intake

02

Remote Diagnosis

IPMI / BMC logs confirm hardware fault

03

Fault Determination

Issue a hardware fault statement

04

Parts Dispatch

Same-day metro, next-day nationwide

05

On-Site Swap

Engineer replaces and tests the part

06

Faulty Part Repair

Log, inspect, repair, test report

07

Verify & Close

Remote business recheck, dual acceptance

Maintenance Pricing

Charged as a percentage of unit market value; tiers and terms are flexible

Equipment Age	Silver / year / unit	Gold / year / unit	Platinum / year / unit	Spare Parts / year / unit
0 – 3 years	1% of unit value	1.2% of unit value	1.4% of unit value	0.3% of unit value
4 – 5 years	1.4% of unit value	1.6% of unit value	1.8% of unit value	0.3% of unit value

Note: Based on the signing-day market price of the complete system, covering A, H and B series. If the market price fluctuates by more than 10%, the average price of the preceding three months may be applied.

Payment: three months deposit plus quarterly payment; the next quarter is due 30 days before expiry. Terms can be negotiated.

Get a Custom Quote

Maintenance FAQ

Equipment coverage, response time, spare parts and pricing

Which devices are covered?

NVIDIA A, H and B series GPU servers, Oracle servers, storage, switches and AI cluster supporting equipment. Full-system coverage or component-level coverage for GPU, motherboard, memory and drives.

What is the response time of each tier?

Silver is 5×8×NBD (next business day on site), Gold is 7×24×NBD, Platinum is 7×24×4H (parts on site within 4 hours). All tiers include 24/7 online support and fault response.

How does advance spare parts work?

We first confirm the hardware fault remotely via IPMI / BMC logs and issue a fault statement. Parts ship same-day within the metro area or next-day nationwide. The customer or our engineer swaps the part, and the faulty part is returned for repair after service restoration.

Can out-of-warranty equipment be covered?

Yes. A series GPUs have a large installed base in China with service life extending to 5 years; H series warranties expire from 2026; B and GB series enter the market with no manufacturer warranty. Third-party maintenance is the practical way to keep clusters running.

What facility conditions does a B300 system require?

B300 draws ~1,400W per GPU, so a fully loaded 8-GPU system needs roughly 12–14kW of rack power and cooling. Before signing we assess rack power, PDB headroom, liquid or high-capacity air cooling and floor loading, then issue racking and cooling recommendations.

Can out-of-warranty H200 systems still get OEM-standard service?

Yes. H200 follows the Hopper diagnostic and repair framework. We provide 24/7 response, advance spare parts and on-site repair to OEM service standards, covering both H200 SXM at 700W and H200 NVL at 600W.

Can dropped or unrecognized GPUs be repaired?

GPU not recognized / dropped card accounts for about 55% of faults and is usually repairable, with localization within 1 hour. HBM faults and ECC errors account for about 15% and can be handled with batch partial-disable repair (customer approval required). GPU core faults account for about 5% and require chip replacement.

How is maintenance priced?

As a percentage of unit market value per year: for equipment aged 0–3 years, Silver 1%, Gold 1.2%, Platinum 1.4%; for 4–5 years, Silver 1.4%, Gold 1.6%, Platinum 1.8%; spare parts service is 0.3% for both.

Which regions are covered?

On-site service is available in major cities including Beijing, Shanghai, Guangzhou, Shenzhen, Chengdu, Hangzhou, Nanjing and Wuhan. Remote operations combined with emergency on-site visits cover other regions. Platinum customers can have a repair base and resident engineers within 5–10 km of their data center.

Keep Your Compute Cluster Online

Send your equipment list and service age to receive a dedicated maintenance plan and quote

Submit Your Requirement Call 400-865-8652

GPU Maintenance

GPU Compute Center Maintenance & Repair