AI Computing Servers
4 / 8 / 10-GPU · air & liquid cooling · PCIe 5.0 · cluster delivery
JH AI computing servers span edge inference, enterprise inference, training clusters and high-density nodes: the 8-GPU RTX 6000D platform, the 10-GPU Thor T5000 high-density inference platform, 8-GPU air-cooled and 8-10 GPU liquid-cooled Intel/AMD platforms, plus PC-Farm distributed-node servers — delivered air- or liquid-cooled up to rack-scale clusters, qualified in one compatibility loop with our enterprise SSDs and DDR5 RDIMM.
Product Models
Specification Comparison
| Model | Type | Interface | Config | Key Specs |
|---|---|---|---|---|
| 8-GPU AI Server (RTX 6000D Platform) | 8×GPU | 4U/5U rack-mount | 8× GPU · 600W | 8× dual-width GPUs · 600W each · 11× PCIe 5.0 |
| 10-GPU High-Density Inference Server (Thor T5000) | 10×GPU | High-density rack | 10× T5000 | Up to 10 dual-die T5000 · 10× PCIe 5.0 x16 · 4× CRPS |
| 8-GPU Air-Cooled AI Server (Intel / AMD) | 8×GPU | 4U/5U rack-mount | 8× GPU · 600W | 8× dual/quad-width blower GPUs · 12× PCIe 5.0 · 12× LFF |
| 8-10 GPU Liquid-Cooled AI Server | 8-10×GPU | 4U/5U rack-mount | 8-10× GPU · 600W | 8-10 triple-width GPUs · full cold-plate coverage · 12× PCIe 5.0 |
| PC-Farm Distributed-Node Server 4U4H | PC-Farm | 4U rack-mount | 4× Node | 4 hot-plug PC nodes · 750W per node · 4× CRPS 1100W |
| PC-Farm Distributed-Node Server 6U6H | PC-Farm | 6U rack-mount | 6× Node | 6 hot-plug PC nodes (1 GPU each) · 750W per node · 4× CRPS 1100W |
Specifications as of ; see each model page and the official product guide for details.
Frequently Asked Questions
Air-cooled or liquid-cooled: which should I choose?
Choose the 8-GPU air-cooled server for fast deployment in standard rooms without a liquid-cooling retrofit; choose the 8-10 GPU liquid-cooled platform for high power density, low PUE and scale-out clusters, with full CPU/GPU cold plates and CDU integration.
How many GPUs are supported, and at what power per card?
The 8-GPU RTX 6000D and 8-GPU air-cooled platforms take eight dual-width GPUs, the liquid-cooled platform 8-10 triple-width GPUs, all up to 600W per card; the 10-GPU Thor T5000 platform takes up to ten dual-die GPUs.
What workloads are these servers intended for?
The 8-GPU RTX 6000D targets LLM inference, AIGC and vision computing; the 10-GPU Thor T5000 high-concurrency API serving and enterprise agents; the air- and liquid-cooled platforms model training; PC-Farm distributed inference and cloud desktops.
What validation and services are included with delivery?
Delivery includes BIOS/BMC, driver, CUDA and firmware compatibility validation, burn-in, NCCL / GPUDirect RDMA and multi-node stress testing, plus solution design, integration testing, on-site deployment and O&M support in one package.
Warranty & Maintenance
JH servers come with four service tiers — Basic Care, Standard+, Gold+ and Platinum+ — covering the system and all covered hardware delivered with it (CPU & memory, GPU/accelerators, storage & networking, platform & fittings, per the delivery BOM). Third-party sourced parts are likewise warranted uniformly by JH.
Compared with Basic Care, the advanced tiers respond faster on parts and on-site service, and Gold+ / Platinum+ additionally include a support plan and service reports.
| Service item | Basic Care | Standard+ | Gold+ | Platinum+ |
|---|---|---|---|---|
| Hotline intake | 7×24 | 7×24 | 7×24 | 7×24 |
| Remote troubleshooting | 7×24 | 7×24 | 7×24 | 7×24 |
| Online technical support | 7×24 | 7×24 | 7×24 | 7×24 |
| Firmware / driver support | Within licensed scope | Within licensed scope | Within licensed scope | Within licensed scope |
| Advance parts | 5×10×NBD dispatch | 7×10×ND delivery | 7×24×4H delivery | 7×24×2H delivery |
| On-site hardware replacement | 5×10×NBD | 7×10×ND | 7×24×4H | 7×24×2H |
| On-site fault resolution | 5×10×NBD | 7×10×ND | 7×24×4H | 7×24×2H |
| Defective-part retrieval | Included | Included | Included | Included |
| Support plan & reports | Service tickets | Service tickets | Included | Included |
SLA notes
- 1.Service windows are in Beijing time: 5×10 means Mon–Fri 8:00–18:00 (excluding statutory holidays); 7×10 means daily 8:00–18:00; 7×24 means around the clock, all year.
- 2.NBD / ND / 4H / 2H are dispatch, delivery or arrival time limits for parts and engineers, counted from remote-diagnosis confirmation; they are not commitments on repair completion, business recovery or cluster availability.
- 3.The hourly 4H / 2H SLAs of Gold+ and Platinum+ apply to P1 and P2 faults; P3 follows ND and P4 is scheduled per work order. Equipment must be within 50 km of a service center.
- 4.Hotline and remote support are 7×24 for all tiers: for P1–P3 faults a qualified engineer contacts the customer and starts handling within 15 minutes; P4 is answered within one business day.
- 5.Unless an advanced tier is selected in writing, Basic Care applies by default. The system warranty term follows the service order and is counted per unit from the date it passes delivery acceptance.
What's included
Severity levels (P1–P4)
A server or critical compute workload is down with no viable alternative; a single GPU/accelerator failure that fully halts a multi-card job may also be treated as P1.
Some hardware or functions are unavailable and compute capacity is clearly degraded, with a risk of interruption, but operation can continue via degraded or alternative measures.
Business is largely normal while some functions, redundancy or management capabilities are limited — for example a redundant part alarms but current operation is not visibly affected.
Business is unaffected; the request concerns product functions, specifications, operation, configuration or routine maintenance.
Server Advanced Warranty Service Guide
PDF · 6.8 MB · Full service scope, execution standards and liability terms are governed by the guide and the mutually confirmed service order.
Other Product Lines
Need Product Selection or Sample Testing?
Our engineering team provides selection advice, compatibility validation, and customized solutions.
Contact Us