FAQ

With full hardware testing plus third-party maintenance they can run production databases. Risks: OEM discontinued, parts shrinking year by year, firmware no longer updated, no OEM support.

Oracle has stopped new SPARC development and OEM new stock is extremely scarce; the market is mostly inventory new machines or second-hand pull-out equipment.

T7/T8: mid-range for OLTP databases and legacy replacement; M8: high-end for core large databases with high concurrency and reliability.

Some high-end models are subject to export controls. Confirm model compliance before purchase and prefer domestic in-stock channels.

Yes — a 7×24 hotline is available, and production clusters can sign an emergency maintenance agreement.

Within-province generally within 24 hours; outside the province 4–48 hours depending on distance, with remote first-aid handling available beforehand.

Third-party maintenance generally includes firmware assessment and upgrade operations, with risks assessed in advance; production system changes require customer confirmation.

Annual maintenance is quoted as a percentage of hardware value, parts included; single repairs are quoted as call-out fee plus parts plus labor, resolving the fault in one visit.

Remote fault troubleshooting, on-site inspection, parts replacement, basic firmware tuning, simple cluster inspection, and a post-incident review report.

Run nvidia-smi status plus ECC log review; extended stress tests (e.g. cuda stress); model-load testing; review factory logs and temperature curves; inspect boards for repair traces.

Fine for testing and non-core training; not recommended for production-scale training. Risks: long full-load aging of GPUs, hidden VRAM defects, latent ECC issues, and no OEM warranty.

Short projects, temporary training and tight budgets: rent. Long-running models, stable business and a 3-year-plus horizon: buying is more economical.

Keep adapter speed consistent with the switch (e.g. 200G IB); match cluster node count to switch ports; for training clusters prefer RDMA-capable models.

Hardware does not carry software licenses. CUDA and base drivers are free; commercial large models, enterprise software and SDKs require separate licenses.

Large-model training: prefer 8 GPUs. Model fine-tuning and mid-scale inference: 4 GPUs. Small-model inference, testing and AI development workstations: 2 GPUs. Single-card VRAM, cluster size and budget decide together.

Certainly. ZhongLianDa supports test-first purchasing — buy one, evaluate, then scale up — and even short-term rental (daily or weekly), ideal for validating compute, compatibility or POC scenarios. NVIDIA DGX and H100 servers both support test-then-buy, and rental fees can be credited against purchase. We can also help set up the test environment with technical support so your evaluation goes smoothly.

For volume purchases (5+ units) ZhongLianDa provides basic installation free of charge: OS installation, RAID configuration, IPMI setup and similar. Complex deployments — Oracle database installation, GPU compute clusters, RAC — are quoted per project. Small purchases without free deployment can buy a deployment service package as needed. Our deployments respect business continuity: executed inside downtime windows, with complete documentation and operator training afterwards.

Three core advantages. First, 23 years focused on server service, with deep repair experience on Oracle and NVIDIA GPU systems where others mostly swap parts. Second, a 3,000 m² parts warehouse with extensive OEM parts in stock — repairs don't wait on sourcing. Third, engineers holding Oracle OCE and NVIDIA DGX certifications who handle the full chain from hardware to database. Our service is trusted long-term by customers in finance, manufacturing and internet industries.

NVIDIA DGX is the officially pre-configured integrated option — turnkey but pricier; a self-built H100 cluster is more flexible on price and can be composed freely by budget. DGX brings certified optimal configuration, integrated scheduling software and whole-system warranty; self-built clusters bring lower cost, configuration freedom and node-by-node expansion. ZhongLianDa supports both, advises by your budget, team and workload, and can also provide a hybrid (official GPU servers + custom network and storage).

Three core differences. First, it requires Oracle-qualified engineers — general hardware repair engineers cannot handle database-level issues. Second, the scope is broader: beyond hardware it includes database-level inspection, backup verification and performance monitoring. Third, the response standard is higher: critical failures need a database expert engaged remotely right away. ZhongLianDa covers both Oracle hardware and database maintenance — one provider instead of coordinating several.

Absolutely. ZhongLianDa is a duly registered company issuing VAT special invoices (6% rate) or general invoices. Provide your invoicing details at signing and we complete the process before service starts. For full reimbursement packs we can also supply service confirmation forms and acceptance reports to fit internal approval flows. All financial processes are standard and transparent.

Yes. Upgrading from 8-hour to 2-hour response, or from remote support to plans including on-site visits, only requires paying the difference — no re-signing. Temporary upgrades are supported too, such as a priority-response channel during business peaks, billed by day. Downgrades are equally supported with pro-rated refunds. ZhongLianDa builds maintenance plans around customer needs.

Yes, the device list comes with the maintenance contract — it is the basis of your dedicated maintenance file. It should include at least: brand and model, serial numbers, configuration, machine-room address and business criticality. The more detail, the faster our response and the more precise our parts. Add or retire equipment during the term and we update the file and pricing free of charge. Your list is kept strictly confidential.

ZhongLianDa maintenance contracts are primarily annual, since longer terms bring steadier service and better pricing. Monthly contracts are supported too — ideal for short projects, transitions or limited budgets. Monthly maintenance carries the same response levels and service quality as annual, at a slightly higher price. Multi-year contracts earn bigger discounts — contact us to discuss. We also offer one-off inspections and emergency repair options.

Yes. ZhongLianDa operates a 3,000 m² spare-parts warehouse stocked with large quantities of brand-new OEM and genuine pull-out components: memory, drives, RAID cards, power supplies, network adapters, GPU accelerator cards and more. We always disclose part origin (new OEM / genuine pull-out / compatible replacement) and warranty key components. Single and volume purchases share the same pricing, with discounts at volume. Check stock online or call for a quote.

Prices for NVIDIA H100 8-GPU servers vary widely by brand, configuration, OS and warranty. ZhongLianDa stocks H100 systems, selling whole units or configured per-GPU, with transparent, negotiable pricing. We also provide one-stop service from selection and deployment to ongoing maintenance, with extra discounts for volume purchases. For a fast quote, submit your requirements online or call our hotline — we respond with a detailed plan within 2 hours.

Each has its strengths depending on your scenario. New servers suit production core workloads with full factory warranty; second-hand servers offer better value for test environments, non-core workloads or tight budgets. ZhongLianDa runs power-on self-test, stress testing and component health checks on every second-hand unit, discloses condition and warranty status honestly, and provides warranty service as well. Choose by budget and business criticality — we can also help prepare a selection plan.

Yes. Every second-hand unit we sell passes strict testing (power-on self-test, stress testing, component health checks). We disclose condition and warranty status honestly and provide matching warranty service. Second-hand equipment offers strong value for budget-limited, test or non-core scenarios — buy with confidence.

Replaced parts carry a warranty — OEM parts are generally covered for 3 months to 1 year depending on part type — and the repair work itself is warranted too. The same failure recurring within the warranty period is repaired free of charge. The exact warranty term is clearly stated on the repair order.

Maintenance pricing is based on device count, brand and model, and service level (response time, on-site coverage, parts handling). We offer annual whole-system maintenance, per-repair service and one-off inspections, and can prepare a maintenance plan and quote free of charge from your equipment list — no hidden fees. Submit your device list online for a quote.