Resources

Practical, vendor-neutral guidance for sizing and buying AI and HPC infrastructure. Every piece ends with a clear next step.

AI Compute · 6 min readH100 vs H200 vs B200: which GPU fits your clusterMemory, bandwidth, power and form factor side by side, plus which one suits training, fine-tuning or inference.Read AI Compute · 5 min readH200 vs B200: Hopper vs BlackwellProven, air-coolable Hopper versus Blackwell’s ~2.5× throughput and mandatory liquid cooling — which fits, and when.Read AI Compute · 5 min readGB200 NVL72 vs H200: rack-scale vs discrete GPUAn integrated 72-GPU liquid-cooled ~120kW rack versus a flexible 700W air-coolable GPU — matched to trillion-parameter scale or node-level builds.Read AI Compute · 4 min readNVIDIA H100: specs, power, price & lead time80GB HBM3, 3.4 TB/s, 700W Hopper — what the H100 is for, its power and cooling, and how to source it.Read AI Compute · 4 min readNVIDIA H200: specs, power, price & lead time141GB HBM3e at 4.8 TB/s on the same 700W Hopper board — the memory-upgraded drop-in for inference and long context.Read AI Compute · 4 min readNVIDIA B200: specs, power, cooling & lead timeBlackwell’s 192GB HBM3e at 8 TB/s and 1.8 TB/s NVLink — frontier training throughput, built around direct liquid cooling.Read AI Compute · 5 min readNVIDIA GB200 NVL72: specs, power & cooling72 Blackwell GPUs + 36 Grace CPUs as one NVLink domain at ~120kW per rack with integrated liquid cooling — and how to source it.Read Cooling · 5 min readSizing liquid cooling for GPU racksHow to match CDUs, rear-door exchangers and direct-to-chip to rack power density without over- or under-buying.Read Power Infrastructure · 7 min readPower planning for a 5 MW GPU hallTransformers, switchgear, UPS and PDUs, and why power is the longest lead item to lock first.Read AI Compute · 4 min readNew, used or refurbished GPUs: when each makes senseWhat changes for warranty, lead time and price, and when refurbished is the smart call.Read Cooling · 5 min readAir vs liquid cooling for AI workloadsWhen direct liquid cooling pays off, and how it changes power and facility planning.Read Power Infrastructure · Interactive toolAI GPU power calculatorEstimate IT power, facility power, cooling load and annual energy cost for H100, H200, B200 and GB200 NVL72 deployments.Read Cooling · Interactive toolAI rack cooling calculatorSize cooling capacity and heat rejection, get an air-vs-liquid recommendation and CDU sizing for H100/H200 HGX, B200 and 100 kW+ GB200 racks.Read Modular Data Centers · Interactive toolAI data center readiness checkerScore your facility for GPU/AI deployment across power, cooling, redundancy and network — see your blockers, required equipment and upgrades.Read AI Compute · Interactive toolAI data center BOM generatorTurn a GPU cluster spec into a full, quote-ready bill of materials — servers, racks, power, cooling, network, monitoring and safety.Read AI Compute · Interactive toolGPU cluster cost estimatorEstimate the cost to build an AI cluster — equipment capex, power/cooling infrastructure, annual energy, and cost per rack and per usable kW.Read Cooling · Interactive toolData center PUE calculatorCompute Power Usage Effectiveness, cooling + distribution overhead, annual energy and cost, and the savings versus a legacy 1.6 PUE baseline.Read Cooling · Interactive toolRack density calculatorWork out kW per rack from units × power or a direct entry, get the required cooling approach, and how many racks a target load needs.Read Power Infrastructure · Interactive toolUPS sizing calculator for GPU racksSize required real power (kW), apparent power (kVA) and the number of UPS units from your IT load, power factor and N / N+1 / 2N redundancy.Read Power Infrastructure · Interactive toolRack PDU sizing calculatorCompute 3-phase line current, per-feed current with dual A+B feeds, and the derated PDU amp rating and count for GPU racks at 208V or 415V.Read Network & Fiber · Interactive toolAI cluster network plannerSize a GPU cluster leaf/spine fabric — leaf and spine switch count, transceivers and MPO fiber trunks — from GPU count, NICs per GPU and oversubscription.Read Cooling · 6 min readDirect-to-chip liquid cooling for AI GPU racksHow cold-plate DLC works, when you need it above ~40 kW/rack, CDU pairing, and the equipment to source for GB200-class density.Read Cooling · 5 min readRear-door heat exchangers for high-density racksPassive rear-door HX cools GPU racks up to ~40 kW without plumbing into servers — when to use it versus direct-to-chip, and what to source.Read Cooling · 6 min readCoolant distribution units (CDUs) for liquid coolingHow liquid-to-liquid CDUs isolate a clean technical loop for direct-to-chip racks, N+1 sizing, and the 1.3 MW units to source.Read Power Infrastructure · 5 min readRack PDUs for high-density GPU racksMetered vs switched, 208V vs 415V, dual A+B feeds and per-outlet metering — how to pick and size rack PDUs for 40 kW+ GPU racks.Read Power Infrastructure · 6 min readData center UPS systems for AI power trainsSizing kW vs kVA, N+1 vs 2N, double-conversion online topology and modular 1500 kVA UPS units for GPU halls.Read Network & Fiber · 6 min readAI data center networking for GPU clustersDesigning a lossless 800G leaf/spine fabric — RoCE Ethernet vs InfiniBand, 51.2 Tb/s switches and MPO/OM4 fiber — for multi-node GPU training.Read
Planning a build and want a second opinion?

Tell us what you're scoping and an engineer will help, usually within a business day.

Talk to an engineerBrowse equipment

Want a second opinion on a build?

Our engineers scope power, cooling and compute together. No payment, and quotes come back in about a business day.

Talk to an engineer