Data Centre
21 articles
-
Why Experienced AI Hardware Suppliers Matter When Purchasing NVIDIA GB300 and NVL72
The execution mistakes on GB300 and NVL72 projects: wrong tier, unready sites, missing networking, no commissioning plan, unclear RMA, and what to ask.
Wei Fei
·
-
Buying NVIDIA GB300 for a Data Centre? Prepare These Documents Before Requesting a Quote
A pre-quote readiness pack for NVIDIA GB300 NVL72 in Singapore: commercial details, rack power, liquid cooling, floor loading, network and compliance papers.
Wei Fei
·
-
NVIDIA GB300 End-User Certificates: Why Suppliers Need to Know Where Your AI Server Will Be Installed
Why suppliers ask where an NVIDIA GB300 system will be installed, who uses it and what it runs, with scenarios for colocation, shared use and later moves.
Wei Fei
·
-
AI Workstations Are Getting Smaller, but Data Centres Aren't Going Away
GB10, Mac Studio and RTX Spark put 128–512 GB AI on a desk, while hyperscalers spend ~US$720B+. Which AI work belongs where for Singapore firms.
Wei Fei
·
-
The Death of the Traditional Server Room? Why SMEs Are Rethinking On-Premise IT
Hybrid, not dead: what Singapore SMEs keep on-premise, what moves to SaaS or cloud, and what repatriation surveys really show.
Wei Fei
·
-
The Silent Danger of Overloaded Server Racks: Are You Monitoring Your PDUs?
Why racks trip when one feed fails, how much a 16A or 32A PDU really carries, and what metered and switched PDUs show. With Singapore pricing.
Wei Fei
·
-
The Problem With Buying Refurbished Enterprise Hardware for Production Systems
Refurbished servers, switches and storage: licence transfer fees, firmware access, drive wear, NEA e-waste rules and when used gear is fine to deploy.
Wei Fei
·
-
Power Strips vs Rack PDUs: Why Data Centres Don't Treat Them as the Same Thing
Power strips vs rack PDUs: Singapore SAFETY Mark rules, C13/C19 outlets, capacity, daisy-chaining and metering — and when a strip is still fine.
Wei Fei
·
-
NVLink vs InfiniBand: Why NVIDIA Uses Both in the Same AI Factory
NVLink scale-up at 900GB/s per direction vs 800Gb/s InfiniBand scale-out in GB300 NVL72 AI factories: roles, switches, traffic and Ethernet options.
Wei Fei
·
-
72 GPUs, One AI Supercomputer: Why NVIDIA GB300 NVL72 Needs NVLink, Not Just More GPUs
How 18 NVSwitch chips give each of 72 Blackwell Ultra GPUs 1.8TB/s in one hop, what 130TB/s aggregate means, and why NVL72 needs NVLink.
Wei Fei
·
-
InfiniBand vs Ethernet: Are AI Data Centres Paying More for Performance They Don't Need?
Quantum InfiniBand vs Spectrum-X Ethernet for AI training and inference: switch costs per Tb/s, SHARP, RoCE tuning, market share and how to choose.
Wei Fei
·
-
Why Your Next AI Data Centre Needs More Than Powerful GPUs
NVLink, NVSwitch, InfiniBand, storage, power and liquid cooling: what a GB300 NVL72 AI data centre needs beyond GPUs, with Singapore capacity and costs.
Wei Fei
·
-
The Hidden Cost of AI Networking: Why Cables and Switches Can Make or Break Your GPU Investment
What DACs, 800G optics and switches cost in an AI cluster, how many a 1,024-GPU fabric needs, and how cable quality affects GPU uptime.
Wei Fei
·
-
37TB of Memory and 130TB/s NVLink: Is NVIDIA GB300 NVL72 the Ultimate AI Factory?
GB300 NVL72's 37TB memory, 130TB/s NVLink and 1.44EF FP4: model capacity, training, inference, 132-142kW power and whether it is the ultimate AI factory.
Wei Fei
·
-
NVIDIA GB300 NVL72 vs 72 Individual GPUs: Why Connecting Them Changes Everything
NVL72 NVLink at 900GB/s per direction vs 8-GPU servers on 800G networks and PCIe GPUs: bandwidth, all-reduce time, power, reliability and when each fits.
Wei Fei
·
-
800Gbps InfiniBand vs 130TB/s NVLink: Why These Numbers Don't Tell the Whole Story
Convert 800Gb/s and 130TB/s to GB/s per GPU per direction: bits vs bytes, aggregate vs per GPU, bidirectional, bisection, NCCL busbw and latency.
Wei Fei
·
-
NVLink vs InfiniBand vs PCIe: Is Your AI Supercomputer Only as Fast as Its Weakest Connection?
NVLink 900GB/s, InfiniBand 100GB/s and PCIe Gen6 121GB/s per direction compared. When the slowest link really limits AI training and inference.
Wei Fei
·
-
From DGX Station to GB300 NVL72: How NVIDIA Scales AI From Your Desk to the Data Centre
How to build and debug AI on a DGX Station GB300 and scale to GB300 NVL72: what carries over, what needs testing in the rack and a step-by-step...
Wei Fei
·
Wei Fei
