Two New Product Lines Go Live: AI Compute Servers and Junhangclaw Inference Workstations
JH Semiconductor's website now hosts two new product lines: AI compute servers and Junhangclaw AI inference workstations. The compute spectrum runs from 50 TOPS on a personal desktop to rack-scale liquid-cooled training clusters, joining the existing enterprise storage, memory, and GPU supply-chain businesses into one-stop compute-plus-storage supply.
50 TOPS or 2,070 TFLOPS: A Selection Guide to the Four Junhangclaw Inference Workstations
The W50, W120, W280, and W2000 map to four user profiles: individuals, small teams, government/enterprise xinchuang, and desktop flagship. This guide compares them by compute tier, chip route, and data boundary, and explains why the W2000's 128GB LPDDR5x unified memory matters for running large models locally.
After 600W per Card: Where Air Cooling Ends and Cold-Plate Liquid Cooling Begins for 8-10 GPU Servers
With GPUs drawing up to 600W per card, cooling selection for 8-10 GPU servers is no longer a matter of preference. This post maps the boundaries of JH Semiconductor's three cooling routes: the preconditions under which the RTX 6000D platform's independent CPU/GPU airflow ducts and the Thor T5000 platform's high-volume air cooling hold up, the three things cold-plate liquid cooling (full CPU/GPU coverage, CDU connection, supply/return water monitoring) buys in high-power-density and low-PUE scenarios, and which three kinds of users the 8-GPU air-cooled platform — positioned to lower the liquid-cooling retrofit barrier — actually fits.
PC-Farm 4U4H/6U6H: Pluggable PC Nodes in a Standard Rack — Maintenance Granularity, the 750W Backplane and Redundant Power Design
JH Semiconductor's PC-Farm distributed node servers pack four (4U4H) or six (6U6H) pluggable PC nodes into a standard rack chassis. This post explains the workload shape this form factor targets: why node hot swap plus fully independent power control cuts maintenance granularity and the failure domain down to a single node, what the two-layer power design — a 750W-peak-per-node backplane and four redundant CRPS 1100W supplies — actually means, and why one GPU card per node happens to match count-scaled businesses such as render farms and batch inference.
The Three Ledgers of J960C: Capacity Gains, Latency Costs, and Model Boundaries of In-Drive Transparent Compression
The transparent-compression model J960C places its compression engine below the FTL, trading host-invisible pipeline stages for effective-capacity and write-amplification gains. This article breaks down the latency and throughput costs, the boundary between compressible and incompressible data, and why it exists as a dedicated TITAN 9 model rather than a line-wide default.
A Second Expansion Path Beyond DDR Channels: How CXL 2.0 Memory Expansion Lands in E3.S Bays
DDR channel counts lock a server's memory capacity and bandwidth into a slowly growing product, while CXL 2.0 memory expansion inserts a byte-addressable capacity tier between local DRAM and SSDs. This article gives a qualitative account of the memory wall, the engineering logic of memory tiering, and the deployment convenience of reusing storage bays through the E3.S form factor.
Component Procurement or Full-Rack Delivery: Five Links a Rack-Level Liquid-Cooled AI Cluster Must Close, and the Spares Chain After Handover
For the same batch of GPU compute, the line between component procurement and turnkey delivery is who carries the integration risk and integration time. This post breaks down the five links a rack-level liquid-cooled AI cluster must close in sequence — cabinet, CDU, water loop, coolant supply/return monitoring and rack commissioning — how this differs from the delivery boundary of an air-cooled single node, and how the delivery process connects to spare-parts and expansion services afterwards.
Machines Still in Service, Drives Already EOL: The Long-Supply Positioning of LEGACY 3 (J360/J380) and Change Control
Rail transit, industrial control and vertical-market systems stay in service far longer than the generational cadence of storage products, and the EOL of legacy-interface drives forces a三-way replacement dilemma of stockpiling, redesign or spot-market sourcing. This post explains the positioning of the LEGACY 3 series J360/J380 as long-supply Gen3 models — installed platforms need drives that stay the same, not drives that get faster — and how the four-layer SKU/PN/Revision/ECN change-control framework behind the frozen model list underwrites multi-year BOM stability.
How AI Data Centers Are Changing Storage Selection: From Peak Bandwidth to End-to-End Data Paths
AI storage planning cannot rely on sequential peak performance alone. This guide maps data preparation, checkpoints, inference retrieval and operations into a practical evaluation framework.
PCIe 4.0 vs PCIe 5.0 Enterprise SSDs: Six Questions to Answer Before Upgrading
PCIe 5.0 raises raw per-lane signaling rate, but benefits depend on CPU lanes, backplane, thermals, I/O behavior and software. Use this server-side upgrade framework.
U.2, E1.S, E3.S or M.2: Enterprise SSD Form Factors Are More Than Dimensions
SSD form factor affects serviceability, thermal path, bay density and future expansion. Compare U.2, M.2 and EDSFF with this engineering selection checklist.
DWPD, TBW and Write Amplification: How to Read Enterprise SSD Endurance
DWPD and TBW are useful endurance metrics, but cannot be compared without warranty term, capacity, write profile and workload. Build an actionable endurance model with this guide.
전체 기사 목록(135개)
- Two New Product Lines Go Live: AI Compute Servers and Junhangclaw Inference Workstations
- 50 TOPS or 2,070 TFLOPS: A Selection Guide to the Four Junhangclaw Inference Workstations
- After 600W per Card: Where Air Cooling Ends and Cold-Plate Liquid Cooling Begins for 8-10 GPU Servers
- PC-Farm 4U4H/6U6H: Pluggable PC Nodes in a Standard Rack — Maintenance Granularity, the 750W Backplane and Redundant Power Design
- The Three Ledgers of J960C: Capacity Gains, Latency Costs, and Model Boundaries of In-Drive Transparent Compression
- A Second Expansion Path Beyond DDR Channels: How CXL 2.0 Memory Expansion Lands in E3.S Bays
- Component Procurement or Full-Rack Delivery: Five Links a Rack-Level Liquid-Cooled AI Cluster Must Close, and the Spares Chain After Handover
- Machines Still in Service, Drives Already EOL: The Long-Supply Positioning of LEGACY 3 (J360/J380) and Change Control
- How AI Data Centers Are Changing Storage Selection: From Peak Bandwidth to End-to-End Data Paths
- PCIe 4.0 vs PCIe 5.0 Enterprise SSDs: Six Questions to Answer Before Upgrading
- U.2, E1.S, E3.S or M.2: Enterprise SSD Form Factors Are More Than Dimensions
- DWPD, TBW and Write Amplification: How to Read Enterprise SSD Endurance
- DDR5 Server Memory Selection: RDIMM, ECC, Capacity and Platform Compatibility
- Enterprise vs Client SSDs: Capacity and Sequential Speed Are Not Enough
- Enterprise SSD Acceptance Checklist: Compatibility, Firmware, Thermals and Recovery
- From DDR5 to NVMe: Capacity Planning That Avoids Memory-Storage Mismatch in AI Servers
- Industry Outlook: Enterprise SSD Competition Is Moving from Spec Sheets to Systems Engineering
- Two Form Factors in FLEX 9: Thermal, Power-Loss-Protection and Backplane Trade-offs Between E1.S and E3.S on Gen5 High-Density Nodes
- Choosing Among Four DDR5 Server Memory Branches: RDIMM 4800/6400, ECC UDIMM 5600, VLP and 3DS 256GB
- Doing the QLC Write-Amplification Math: Endurance Mechanics and the Dual-Port Trade-off in Q9 J960Q/J960QD
- From Capacitor Energy Budget to NVMe-MI: Four Implementation Details in Our Enterprise Feature Baseline
- JH-ZS101 Test Equipment: Throughput and Evidence Chains Behind 600-Drive Parallel Burn-In
- Controller Roadmap: External Gen4 Silicon Plus In-House Firmware, and Converging Interfaces for JHC400
- Three Drive Tiers in an Inference Cluster: X9 for KV Cache, Q9 for Checkpoints and Very Large Arrays, TITAN 9 for Online Databases
- When the L2P Table No Longer Fits in DRAM: Tiered Mapping Caches and Indirection Granularity
- The Cost Model Behind Garbage Collection: Valid Page Ratio, Over-provisioning and Steady-state Write Amplification
- Lifetime Is Set by the Upper Tail: Static Wear Levelling and the Cost of Block Swaps
- Queue-to-Core Binding and Interrupt Coalescing: NVMe Scheduling Trade-offs on Gen4
- After the Write Completion: Durability Semantics and Power-Up Reconstruction in Enterprise SSD Power-Loss Protection
- Online Firmware Update Without Interrupting I/O: Dual-Bank Images, the Activation Window, and Rollback Limits
- Atomic Commit of the FTL Mapping Table: Shadow Updates, Log Replay and the Power-Loss Window
- Hot-Swap and Surprise Power Loss: Two Reset Paths and What Ready Time Is Made Of
- The 10 µs Write Acknowledgement Is a Contract: Where It Lands, What Flush Costs, and the Steady-State Gap
- Decomposing a 65 µs Read: From SQ Fetch to CQ Post
- Deallocate at the FTL Level: How Trim and Write Zeroes Change GC Copy Volume
- After the Layer Count Rises: What 3D NAND Density Gains Leave for Drive Makers
- Two-Tier LDPC: The Latency Budget of Hard Decision and the Real Cost of Soft Decision
- What Read Retry Really Costs: Modelling Threshold Drift and Designing the Recovery Ladder
- Unpacking 7,400 MB/s: The Parallelism Budget Across Channels, Dies and Planes
- The Single-Bit Cache Trade-Off: Why Enterprise Drives Are Specified on the Steady-State Curve
- Retention Loss and What Actually Triggers a Background Refresh
- Read Disturb: Physical Origin, Counter Granularity and Migration Thresholds
- Handling Program Failures by Class: Retry Boundaries, Data Provenance and Bad-Block Growth
- What One Third of the Capacity Buys: The Boundaries of pSLC, and Three Ways to Misuse It
- How Millisecond Erases Eat Your Tail Latency: Staggered GC Scheduling and the Cost of Suspend
- How Wide Should the Stripe Be? Choosing an Allocation Unit and Paying for It Twice
- Where QLC Belongs: Workload Boundaries for the Q7 and Q9 Series
- Under 700 Nanoseconds per Command: Core Partitioning and L2P Sharding in the Dispatch Engine
- NAND Channel Interfaces: DQS Training, Fan-out Loading and Timing Margin
- The DRAM Budget Inside an SSD: Mapping Table, Metadata and Write Buffer
- Where the AES-256 Engine Sits in the Datapath: Architecture Trade-offs for Self-Encrypting Drives
- JH Semiconductor Visits the 2026 South China Smart Manufacturing and Industrial Automation Exhibition
- Omdia Raises 2026 Semiconductor Growth Forecast to 62.7% as AI Storage Demand Takes the Lead
- Blackwell to Top 70% of NVIDIA High-End GPU Shipments in 2026, Reshaping the AI Computing Landscape
- Where the Margin Comes From at 32 GT/s: PCIe Link and PHY Engineering in Enterprise SSDs
- Reading 18/22/6 W Correctly: Power States, DVFS, and the Thermal Ceiling
- Where to Put the Check Points: Placing T10 PI Along the Controller Datapath
- Huawei Ascend 950P Servers Enter Volume Delivery as China's AI Computing Ecosystem Matures
- DDR5 Memory Market in 2026: AI Servers Drive a Demand Surge
- P99.9 Is Not an Extension of the Average: Three Sources of Tail Latency and How to Isolate Them
- JH Semiconductor Team Attends IC China 2026 in Beijing
- SR-IOV Passthrough and Namespace Isolation: Moving VM I/O Out of the Hypervisor
- Enterprise SSD Selection Guide: NVMe U.2 vs. E1.S in AI Data Centers
- Dual-Port Failover: x2+x2 Links, Reservation Semantics and System Bring-up
- Telemetry Log Partitioning and the Version Discipline That Makes It Usable
- Inspur NF5488A7 AI Training Server Now Available at JH Semiconductor
- Getting a Decommissioned Drive Through Sanitisation: Prerequisites, Window Planning and Evidence
- NAND Flash Supply and Demand in 2026: Tight Supply and a Rising Price Cycle
- A Compatibility Matrix Is Not an Exhaustive Sweep: Pruning Dimensions and Defining Pass Criteria
- The AEN Channel: Moving Fleet Monitoring from Polling to Push
- JH Semiconductor Visits the 2026 Shenzhen International Semiconductor and IC Exhibition
- Namespace Partitioning and the Side Effects of Format: Block Formats, Erase Semantics and OP Accounting
- eMMC and UFS in Automotive Systems: From Infotainment to Autonomous Driving
- Why Percentage Used Does Not Travel Across Models: Reading Drive Health Data Correctly
- Three Storage Workloads in a Training Pipeline: Sample Loading, Checkpoints and KV Cache
- Which Line on the Spec Sheet Decides Vector Search P99: The Arithmetic of a Serial Read Chain
- Reliability Design of Industrial SSDs for Rail Transit and Energy Applications
- Choosing SSDs for Object Storage Back Ends: EC Read Amplification, Density and Rebuild Windows
- No Engineer On Duty at the Edge: Letting the Cost of a Site Visit Define the Drive's Feature List
- JH Semiconductor Technical Team Visits SEMICON China 2026 in Shanghai
- RDIMM vs. LRDIMM Server Memory: Choosing the Right Option for AI Workloads
- What the Drive Must Add After Disaggregation: Target-Side Demands NVMe-oF Places on Enterprise SSDs
- Two Write Paths in OLTP: Commit-Path Logging Versus Page Flush
- Profiling VDI Storage: Boot Storms, Steady-State Writes and Where Selection Breaks Down
- Industrial-Grade eMMC and UFS Storage Chips Now Available
- Domestic Substitution in Memory Chips: Opportunities and Challenges
- 70/30 Mixed Workloads: How Much Read Performance Writes Actually Take Away
- Reading a Gen5 Spec Sheet: Without Stated Conditions, the Numbers Are Not Comparable
- Three Engineering Gates Above 30TB per Drive
- HBM High Bandwidth Memory: The AI Chip's Essential Companion and Its Supply Chain
- What MTBF and AFR Actually Measure — and What They Do Not
- Choosing Between E3.S and U.2: Density, Power Ceiling and the Pace of Chassis Refresh
- U.3 Compatibility Runs One Way: Backplanes, Power Budgets and Selection Mistakes
- 1 DWPD Versus 3 DWPD: Reading the Write Budget When Two Capacity Ladders Do Not Line Up
- Embedded Storage Selection: eMMC vs. UFS vs. SPI NOR Flash by Application
- Auditing a Performance Number: Stated Conditions, Internal Consistency, Distribution
- Gen4 to Gen5 Knock-On Effects: Power, Thermals and Form Factor
- NVIDIA L40S Data Center GPUs Now Available at JH Semiconductor
- From 22 W to E1.S Airflow: How an Enterprise SSD Thermal Budget Is Allocated
- Gen4 Is Not Retiring Yet: How CORE 7 and TITAN 7 Divide the Work
- Data Center Storage Evolution: The Migration Path from SAS HDD to All-Flash NVMe
- From EVT to Volume: Gate Criteria, Rework Traps and Test Throughput
- Designing Failure Analysis Before Any Drive Comes Back
- One Part Number, One Frozen BOM: Building Traceability Before Mass Production
- Firmware Freeze Is Not a Code Stop: Setting Production Gates and Exception Paths
- JH Semiconductor Completes Intake and Listing of Enterprise NVMe SSDs
- From Samples to the AVL: Four Gates in Enterprise SSD Customer Qualification, and Where It Stalls
- Board-Level Margin at 32 GT/s: Stackup, Connectors and Production Consistency
- DDR5 RDIMM Sub-Channels and On-Module Power: What the Structural Changes Buy
- Why the DDR5 Capacity Branch Runs Through 3DS: Drawing the RDIMM / LRDIMM Line
- Numbers Before Steady State Do Not Count: Preconditioning and Reporting Discipline in SSD Performance Testing
- Sequencing Reliability Validation: From Media Characterisation to System Bring-up
- New DDR5 RDIMM Server Memory Products Now Available
- JH Semiconductor Completes 2025 Annual Warehouse Stocktake
- JH Semiconductor Visits the 2025 Shenzhen International Touch and Display Expo
- JH Semiconductor Attends IC China 2025 in Beijing
- JH Semiconductor Receives and Stocks New Batch of DDR5 Server Memory
- Engineering Application Toolkit Fully Upgraded: Intelligent Part Selection and Online Simulation
- JH Semiconductor Takes Part in ES SHOW 2025 in Shenzhen
- JH Semiconductor Visits Hannover Messe Shanghai Industrial Shows 2025
- Domestic Substitution in Analog ICs: Breakthroughs in Op Amps and Power Management ICs
- JH Semiconductor Attends the 2nd Bay Area Semiconductor Industry Ecosystem Expo in Shenzhen
- Building Semiconductor Supply Chain Resilience: Diversified Sourcing and Risk Management
- JH Semiconductor Launches Digital Supply Chain Platform
- JH Semiconductor Exhibits at the 27th China International Optoelectronic Exposition (CIOE 2025)
- JH Semiconductor Completes ESD Environment Upgrade of Its Shenzhen Warehouse
- JH Semiconductor Completes Update of Server and GPU Product-Line Technical Documentation
- JH Semiconductor Engineering Team Completes NVIDIA Data-Center GPU Product Training
- JH Semiconductor Technical Team Visits electronica China 2025 in Shanghai
- JH Semiconductor Visits SEMICON China 2025, Tracking Advanced Packaging and Memory Manufacturing
- JH Semiconductor Completes 2025 Annual Supplier Review and Tiered Evaluation
- JH Semiconductor Completes Internal ERP Inventory System Upgrade with Real-Time Stock Sync
- JH Semiconductor Visits NEPCON China 2025, Tracking Packaging, Test and SMT Advances
- JH Semiconductor Team Visits the 2025 Shenzhen International Electronics Show and Huaqiangbei Electronics Fair
- JH Semiconductor Visits LASER World of PHOTONICS CHINA 2025 in Shanghai
135개 기사 · 1/12 페이지