Walkthrough: Design a 1024-GPU Foundation-Model Pretraining Cluster

A 1024-GPU cluster is the modern minimum for serious foundation-model pretraining — enough to train a 70B-parameter dense model on 5-15T tokens in 30-90 days, or a Mixture-of-Experts (MoE) model with 400B+ total / 30-50B active in similar wall-clock. Below 512 GPUs you cannot afford the data-parallel batch size to reach the Chinchilla-optimal token ratio in a reasonable schedule; above 8192 GPUs you are doing what xAI Colossus (200K H100, expanding to 1M H200), Meta’s 24K H100 Llama 4 training cluster, Anthropic’s Project Rainier (multi-hundred-thousand Trainium2 with AWS), and Microsoft Stargate (the announced $100B+ joint OpenAI-Microsoft buildout) are doing — a different cost class and a different operations problem.

This walkthrough builds the 1024-GPU reference: hardware selection across NVIDIA / AMD / Google / AWS silicon, the fabric, the storage, the orchestrator, the training framework, the observability surface, the cooling and power envelope, and the financial model. CAPEX lands at $70-110M depending on accelerator choice; OpEx $10-25M/yr. Reference builds: Meta Grand Teton + RSC, Microsoft Mt Hood + Azure Cobalt100, Tesla Dojo D1, Google Trillium TPU v6e and Ironwood TPU v7, AWS Trainium2/3 UltraServers.


1. What 1024 GPUs gets you

WorkloadWall-clock at 1024× H100Wall-clock at 1024× B200
70B dense, 2T tokens (Chinchilla 2x)~55 days~22 days
70B dense, 15T tokens (Llama 3.1 70B class)~410 days (impractical)~165 days
405B dense, 15T tokens (Llama 3.1 405B class)infeasible at this scale~950 days (infeasible)
MoE 400B-total / 50B-active, 8T tokens~95 days~38 days
8B dense, 3T tokens (Llama 3 8B class)~7 days~3 days

A 1024-GPU pod is the right size for: pretraining 7-70B dense and 50-150B-active MoE; high-throughput finetuning of frontier-scale checkpoints; full data-parallel ablation sweeps on architecture choices. Bigger frontier models (200B+ dense, 1T+ MoE) need a multi-pod build.


2. Accelerator selection: NVIDIA vs AMD vs Google TPU vs AWS Trainium

SiliconPer-GPU peak (BF16/FP8)HBMInterconnectList price (2025-26)Per-GPU $/TFLOPS·yr (3yr amort)
NVIDIA H100 SXM5 80 GB989 TF BF16 / 1979 TF FP880 GB HBM3NVLink 4 (900 GB/s) + IB NDR$25-30Kbest documented
NVIDIA H200 SXM 141 GB989 TF BF16 / 1979 TF FP8141 GB HBM3eNVLink 4 + IB NDR$32-40K+memory for inference
NVIDIA B200 SXM 192 GB2.25 PF BF16 / 4.5 PF FP8 / 9 PF FP4192 GB HBM3eNVLink 5 (1.8 TB/s) + IB XDR or Spectrum-X$40-55Kbest 2025 dense perf/$
NVIDIA B300 / Blackwell Ultra2.5 PF BF16 / 5 PF FP8 / 15 PF FP4288 GB HBM3eNVLink 5 + Quantum-X800$60-75K (early 2026)best 2026 perf/$
NVIDIA GB200 NVL72 (rack-scale)720 PF FP8 per rack (36×B200 pairs)13.4 TB HBM3e/rackNVLink 5 fully meshed in rack$3.0-3.5M per rackrack-scale TCO leader
NVIDIA GB300 NVL72 / NVL576 (2026)1.4 EF FP4 per NVL7220 TB HBM3e/rackNVLink 5 + Quantum-X800 800G$4.5-5.5M per rackfrontier 2026 platform
AMD MI300X 192 GB1.31 PF BF16 / 2.61 PF FP8192 GB HBM3Infinity Fabric (896 GB/s) + RoCE$15-20K~30-40% cheaper than H100, lags FP8 maturity
AMD MI325X 256 GB1.31 PF BF16 / 2.61 PF FP8256 GB HBM3eInfinity Fabric + UALink (roadmap)$22-28Kinference-leaning
AMD MI355X (CDNA 4, mid-2025)2.3 PF FP8 / 4.6 PF FP4288 GB HBM3eInfinity Fabric + UALink$30-40Kfirst credible AMD training challenger
Google TPU v5p459 TF BF16 / 918 TF INT895 GB HBMICI 4800 Gb/s in 8960-chip podnot sold — $2-4/chip-hr GCPbest TCO at scale, GCP-locked
Google TPU v6e Trillium918 TF BF16 / 1836 TF INT832 GB HBMICI in 256-chip pod$2.70-4.20/chip-hr GCPinference + small training
Google TPU v7 Ironwood (2026)4.6 PF FP8192 GB HBM3eICI in 9216-chip podnot yet retailGoogle’s frontier-class TPU
AWS Trainium21.3 PF BF16 / 2.6 PF FP896 GB HBM3NeuronLink in UltraServer 64-chip~$1.30/chip-hr us-east-1Anthropic Project Rainier silicon
AWS Trainium3 (2026)~2x Trainium2, FP4128 GB HBM3eNeuronLink-v2 + EFAnot yet retailfollows Anthropic offtake

Recommendation for 1024-GPU greenfield Q3 2025+ build: NVIDIA B200 SXM in 8-GPU HGX baseboards, 128 nodes — $50-65M silicon; or 14× GB200 NVL72 racks (1008 GPUs) at $45-55M for the rack-scale variant. GB200 NVL72 wins on training perf/$ because the 72-GPU NVLink domain replaces 9 InfiniBand hops with NVLink for the tensor-parallel and expert-parallel dimensions — see distributed-systems-fundamentals.

For organizations not GPU-locked, Google TPU v5p / v6e via GCP commitments offers ~20-35% lower $/token-trained at frontier scale but ships you fully onto JAX + MaxText + GCP — a different sovereignty story.


3. Network fabric: the cluster’s other half

A 1024-GPU pod has three logical networks layered on the physical fabric:

NetworkBandwidth targetProtocolVendor stack
Compute backplane (collective ops, AllReduce / AllGather / AlltoAll)400-800 Gb/s per GPUInfiniBand NDR / XDR, or RoCEv2, or NVLink within scale-up domainNVIDIA Quantum-2 QM9700, Quantum-X800, NVLink Switch (NVSwitch)
Storage + management100-200 Gb/s per nodeRoCEv2 over Ethernet, or NFS over IBSpectrum-4 SN5600 (Ethernet), Quantum-2 (IB)
Out-of-band BMC / IPMI1-10 Gb/sEthernetany 1G/10G ToR

Compute backplane choices:

  1. InfiniBand NDR 400 Gb/s (NVIDIA Quantum-2 QM9700, 64 ports of 400G): the default for NVIDIA training clusters. Lossless, RDMA-native, ~700 ns switch latency. 8 NICs per H100/H200 node (one ConnectX-7 per GPU) on a non-blocking rails-optimized fat tree. For 1024 GPUs: 2 layers (128 ToR + 16 spine) of QM9700 — ~144 switches $40-60K each = $6-9M just in fabric.
  2. InfiniBand XDR 800 Gb/s (Quantum-X800 Q3400): 144 ports of 800G in a single chassis; collapses 2 layers into 1 at this scale. 2025 product; ~$120-180K per chassis. Required for B200/GB200 to feed full NVLink-5 egress.
  3. Spectrum-X (NVIDIA Ethernet, 800 GbE) + ConnectX-8 NICs: lossless RoCEv2 with NVIDIA’s adaptive routing + congestion control. Comparable to IB on AllReduce within ~5%, easier to integrate with existing Ethernet ops. Hyperscalers (Meta, Microsoft) are leaning toward Spectrum-X or in-house Ethernet for new builds.
  4. AWS EFA (Elastic Fabric Adapter) — only relevant in AWS-hosted clusters; SRD transport, comparable to RoCEv2.
  5. UALink (Ultra Accelerator Link) + Ultra Ethernet consortium (UEC) — emerging open standards for 2026-27, AMD/Intel/Broadcom-led; not yet deployable.

NVLink + NVSwitch (scale-up domain): within a GB200 NVL72 rack, NVLink-5 connects all 72 GPUs in a single coherent memory domain via 9 NVSwitch trays — 1.8 TB/s per GPU, 130 TB/s aggregate. This collapses what used to be cross-node tensor parallelism into in-rack NVLink, which is ~10× faster than IB.

Rails-optimized topology: for 8-NIC servers, each rail connects to a separate leaf switch, so a single AllReduce uses 8 parallel network paths and avoids hot-spot collision. See realtime-embedded for network-determinism context.


4. Storage tier

Foundation-model training reads training data (~1-10 PB tokenized) and checkpoints every 30-120 min (each checkpoint 1-3 TB for a 70B model, 8-20 TB for a 400B MoE). Three tiers:

TierCapacityThroughputVendorsCost
Hot scratch (parallel filesystem)1-5 PB NVMe1-5 TB/s aggregateVAST Data, WekaIO, DDN ES400NVX2 / EXA6, Pure FlashBlade//S, IBM Storage Scale (GPFS), Lustre on Cray ClusterStor$1.50-3.50/GB raw
Warm dataset (object)5-50 PB100-500 GB/sMinIO, Cloudian HyperStore, Ceph (RADOS), VAST (unified)$0.30-0.80/GB raw
Cold archive50-500 PB1-10 GB/sAWS S3 Glacier Deep Archive ($0.00099/GB-mo), Azure Archive ($0.00099/GB-mo), tape (LTO-9 $15/TB media)$0.001-0.10/GB-mo

Recommended baseline: VAST Data Universal Storage (3 PB usable, 2 TB/s) for hot — $6-8M; Cloudian or Ceph for warm bulk dataset; S3 Glacier for archive. VAST and WekaIO are the two architectures actually shipping at >2 TB/s sustained for AI training in 2025; DDN’s EXA6 ships in the same class for the HPC-native customers. Pure FlashBlade//S is the easiest to operate but tops out lower in raw bandwidth.

Checkpoint write strategy: distributed checkpointing (PyTorch DCP, MosaicML Streaming, Megatron checkpoint sharding) so each rank writes only its shard — 1024 ranks each writing 1-3 GB completes in 20-60 s versus 10-30 min for a serialized rank-0 dump.


5. Cluster orchestration + scheduling

Two camps: HPC-native (Slurm) and Kubernetes-native (Volcano + KubeRay).

StackStrengthWeaknessUsed by
Slurm 24.05 + Pyxis (NVIDIA container plugin) + Enrootdense per-job GPU scheduling, gang scheduling first-class, decades of HPC ergonomicsweak multi-tenancy, no native autoscaling, awkward for servingOpenAI training, Meta RSC, Anthropic, most national labs
Kubernetes + Volcano + KubeRayunified train+inference platform, integrates with cluster autoscaler, MLOps-friendlygang scheduling bolt-on, GPU topology awareness immatureCoreWeave, Lambda Labs, internal hyperscaler training-as-a-service
Determined.ai (HPE)research-team UX, hyperparameter sweeps, integrated experiment trackingsmaller community, vendor riskenterprise R&D teams
MosaicML Composer + MCLI (now Databricks)turn-key recipes, optimized DDP/FSDP, Streaming datasetDatabricks-tiedDatabricks-hosted training
Ray + KubeRayflexible, Python-native, good for RL + agent pipelinesnot ideal for pure synchronous pretrainAnyscale, RL pipelines

Recommendation: Slurm 24.05 + Pyxis for the pretraining workload itself (gang scheduling, MPI integration, predictable performance), with a separate small Kubernetes pool (KubeRay) for ablation jobs, eval harnesses, and inference smoke tests. This is the architecture Meta, Anthropic, Mistral, and most research labs converge on.

Containers: Apptainer (formerly Singularity) for HPC sites; Enroot for NVIDIA-native sites; both are rootless and avoid Docker daemon. NVIDIA NGC containers (PyTorch + CUDA + NCCL + cuDNN preinstalled) are the lingua franca.


6. Training framework + parallelism

For a 70B+ dense or 400B+ MoE on 1024 GPUs, raw PyTorch DDP is not viable — communication overhead dominates. Pick one:

FrameworkParallelism dimsBest forNotes
Megatron-LM / Megatron-Core (NVIDIA)TP + PP + DP + EP + CPdense + MoE, NVIDIA-tunedReference for NVIDIA-stack pretraining; Llama 4 trained on Megatron-derived stack
DeepSpeed (ZeRO 1/2/3) + DeepSpeed-MIIZeRO sharding + 3Ddense, memory-constrainedMicrosoft; integrates with HuggingFace Trainer
PyTorch FSDP2 (2024 GA)fully-sharded data parallel + TPflexible, native PyTorchReplaces FSDP1; better composability with TP via DeviceMesh
ColossalAI3D + ZeRO + Gemini offloadopen research, China-popularLess production-hardened
Levanter (Stanford CRFM, JAX)FSDP + TP via jax.shardingTPU + GPU, reproducibility-firstBit-exact resumption, used for Marin
MaxText (Google)JAX-native, TPU-optimizedTPU podsReference for TPU training
GPT-NeoX (EleutherAI)3D parallel via Megatron-DeepSpeedopen-source pretrainingPythia, OpenLLaMA derivative work

Recommendation for B200/GB200 + 1024 GPUs: Megatron-Core with TP=8 (within NVL72 NVLink domain or within HGX node), PP=8 (across racks), DP=16, with EP=8 for MoE. FSDP2 for 7-13B dense where TP overhead exceeds memory benefit. See distributed-systems-fundamentals.

Mixed precision: FP8 (Transformer Engine on Hopper/Blackwell) for forward + backward; BF16 master weights; FP4 (Blackwell-only) for inference. ~1.5-2× throughput vs BF16 with ~0.1-0.3% downstream eval delta when calibrated.


7. Data pipeline

Tokenized training data lives upstream of the cluster. Pipeline:

  1. Crawl + collection: Common Crawl snapshots (250-400 TB/dump), curated repos (Wikipedia, ArXiv, StackExchange), licensed data (books, news, code repos under license), synthetic data (model-generated).
  2. Filtering + dedup: AI2 Dolma pipeline, RefinedWeb-style heuristics, FineWeb-Edu (educational filter), MinHash + LSH dedup at document + paragraph + line level. Datology AI, Cleanlab, and Snorkel offer commercial pipelines.
  3. Quality classification: educational-value classifier (Llama 3 ed classifier, FineWeb classifier), toxicity + PII redaction (Microsoft Presidio).
  4. Labeling / annotation (post-train data): Argilla, Scale AI, Surge AI, Lilac (Databricks) for human-feedback labels.
  5. Tokenization: SentencePiece or HuggingFace tokenizers; vocabulary 100K-200K (Llama 3 = 128K, GPT-4o = 200K).
  6. Sharding + streaming: MosaicML Streaming (MDS format), WebDataset (tar shards), Megatron’s IndexedDataset. Stream from object store (Cloudian / S3) with prefetch — avoids loading 10 TB onto every node’s scratch.

A clean 5-10T-token mix takes 2-4 PB of raw crawl input and 3-6 weeks of CPU preprocessing on a 500-1000-core cluster. Plan for it as a parallel workstream, not an afterthought.


8. Cluster software baseline

ComponentVersion (2025-26)Role
OSUbuntu 22.04 LTS or RHEL 9.4NVIDIA-certified
Kernel6.5+ (with NVIDIA Open GPU Kernel Modules)required for B200
NVIDIA driver560+B200/GB200 support
CUDA12.6+FP8 + Transformer Engine
cuDNN9.xconv + attention kernels
NCCL2.22+collectives; SHARP offload on Quantum-2
Mellanox OFED24.07+InfiniBand drivers + libraries
UCX1.17+unified communication transport
Slurm24.05scheduler
Pyxis + Enrootlatestcontainer runtime
Apptainer1.3+alternative container runtime
Python3.11training runtime
PyTorch2.5+ (with Triton 3.1)framework
Transformer Engine1.10+FP8 kernels
Megatron-Core0.9+parallelism

A cluster-config-as-code repo (Ansible + Terraform) is the only sane way to manage 128-256 nodes. Updates are blue/green or rolling-with-drain.


9. Observability

GPUs fail. NCCL hangs. Power supplies trip. Without per-GPU + per-rank + per-job observability, a 30-day training run loses 15-40% wall-clock to silent stalls.

LayerToolWhat it captures
GPU metricsNVIDIA DCGM + dcgm-exporter Prometheusutilization, SM occupancy, HBM ECC errors, NVLink errors, thermal throttle, power
FabricUFM (Unified Fabric Manager, NVIDIA)IB link health, congestion, BER, port flap
ClusterPrometheus + Grafana + Lokimetrics + logs aggregation
Distributed traceJaeger + OpenTelemetryrank-to-rank dependencies
Job profilingNVIDIA Nsight Systems, PyTorch Profiler, Pixieper-step kernel timing
Cloud Native StackNVIDIA Cloud Native Stack 12+bundled reference deployment
CommercialDatadog (DCGM integration), New Relicoptional

Loss curves + gradient norms + token/sec throughput shipped to Weights & Biases or MLflow. SRE-style alerting on: throughput drop >10% from baseline (likely a stuck rank), HBM ECC error rate spike, NCCL timeout, loss-spike detection (NaN trap + autorollback). See observability-stack.


10. Power, cooling, and physical layout

A 1024-GPU B200 cluster draws ~1.5 MW IT load (1.2 kW/GPU × 1024 + CPU/NIC/storage). A GB200 NVL72 deployment at 14 racks × 132 kW = 1.85 MW IT — plus 0.3-0.4 MW for storage, network, and cluster mgmt. Plan 2.2-2.5 MW total facility load including PUE.

Rack-scale options (OCP-aligned and proprietary):

VendorReference platformForm factor
NVIDIA-MGXopen spec, multiple ODMsmodular
GB200 NVL72 (NVIDIA + ODMs)72-GPU rack, NVLink-meshed132 kW liquid
Supermicro SYS-A21GE-NBRTGB200 NVL72 implementationrack-scale
Foxconn / Hon HaiGB200 NVL72 ODMrack-scale
Inspur NF5688M7HGX H200 8-GPU10 kW/node air
Gigabyte G593-SD0HGX B20011 kW/node liquid
Quanta / QCTHGX B200 + GB200rack + node
Wiwynn ES200G2GB200 NVL72rack-scale
Hyve Solutionshyperscaler customrack-scale
Penguin Solutionsturn-key HPC clustershall-scale
Aivres (formerly Inspur USA)HGX + GB200rack-scale
PegatronGB200 NVL72 ODMrack-scale

Cooling: direct-to-chip cold plates (CoolIT OMNI, Asetek RackCDU, Motivair MCDU, JetCool SmartPlate two-phase) on every GPU and CPU; rear-door heat exchangers (Motivair ChilledDoor, Vertiv Liebert XDR, STULZ CyberRow) for residual ~15-20% air load. CDU per rack pair (Motivair MCDU-1500 at 1.5 MW, ~$120-180K each). Primary loop W32-W45 ASHRAE class to dry-cooler array (Güntner, Modine) or chilled water from Schneider EcoStruxure ETM. See design-data-center-cooling-system.

50-100 kW/rack is the new normal for AI; GB200 NVL72 pushes 132 kW; GB300 NVL576 will push 200+ kW/rack via Vertiv + Schneider integrated solutions. Air-only is not on the menu.


11. CAPEX + OpEx

11.1 CAPEX (greenfield 1024-GPU B200 build, 14 racks GB200 NVL72 variant)

LineCost
14 × GB200 NVL72 racks (~1008 GPUs) @ $3.2M$44,800,000
Or: 128 × HGX B200 nodes @ $420K$53,760,000 (alternative path)
Quantum-X800 IB fabric (compute backplane)$6,500,000
Spectrum-4 storage + mgmt network$1,800,000
VAST Data hot scratch (3 PB usable)$7,200,000
Cloudian warm bulk (15 PB usable)$4,500,000
CDUs (14 × Motivair MCDU-1500 + spares)$2,100,000
Power distribution (busways, PDUs, transformers)$3,500,000
Site build / fitout (white space, dry coolers, fire suppression)$8,000,000
Cluster head nodes + login + slurmctld + storage mgmt$600,000
Software (NVIDIA AI Enterprise sub for 1024 GPUs, ~$4,500/GPU/yr)$4,608,000/yr (Year 1 prepay possible)
Integration + commissioning + Cx$3,500,000
Contingency 10%$8,400,000
Total CAPEX (GB200 NVL72 path)~$95M

For the lower-end H200 path: $70-80M total. For frontier GB300 NVL72/NVL576 buildout (2026): $120-160M. The hyperscaler builds (xAI Colossus $2-3B for 100K H100, Stargate $100B+) scale linearly until you hit grid + cooling constraints.

11.2 OpEx (annual, fully operational)

LineAnnual
Power (2.2 MW × 8760 h × $0.07/kWh blended)$1,350,000
NVIDIA AI Enterprise subscription (renewal)$4,600,000
InfiniBand support + spares$650,000
Storage software support (VAST + Cloudian)$1,200,000
Hardware spares + RMA pool (1.5% of CAPEX/yr)$1,400,000
SRE + cluster engineering (4 FTE × $280K loaded)$1,120,000
ML platform engineering (3 FTE × $300K loaded)$900,000
Data team (preprocessing + curation, 4 FTE × $240K)$960,000
Compliance + security audits$200,000
Facility lease (if not owned, ~$200/kW/mo)$5,280,000
Total OpEx~$17-18M/yr

Owned facility + own-grid power can cut OpEx to $10-12M/yr; full colo at retail rates pushes $22-25M/yr.

For comparison: training a 70B-class model on a public cloud (AWS Trainium2, GCP TPU v5p, Azure ND H100 v5, Lambda Labs, CoreWeave, Together AI) lists at $2-8M per pretrain run at this scale — owning the cluster pays back at 3-6 full pretrains, plus retains all post-train + experimentation capacity.


12. Reference builds + what they actually look like

  • Meta Grand Teton + Open Rack v3 + RSC: Meta’s training platform; 24,576 H100 cluster trained Llama 3.1 405B; two clusters reported. Grand Teton chassis fits 8 H100 in liquid-assisted air; RSC is the prior 16K A100 platform. Meta announced two 24K H100 clusters and is building toward 350K H100-equivalent by end-2024. MTIA (Meta Training and Inference Accelerator) v1/v2 silicon for inference workloads, complementing NVIDIA training.
  • Microsoft Mt Hood + Azure Cobalt 100 + Stargate: Mt Hood is the GB200 NVL72 reference; Cobalt 100 is the ARM CPU in Azure for general-purpose; Stargate is the announced $100B+ joint OpenAI-Microsoft buildout reportedly targeting multi-GW frontier-training capacity through 2028.
  • xAI Memphis Colossus: 100K H100 brought online in 122 days (mid-2024), expanding to 200K including H200, with public stated ambition of 1M H200-equivalent. Direct-to-chip liquid on Supermicro reference platform; controversial on-site natural-gas turbines for supplemental power.
  • Anthropic Project Rainier: multi-hundred-thousand-chip AWS Trainium2 cluster announced 2024 as the substrate for Claude pretraining; first-party AWS silicon collaboration.
  • Tesla Dojo D1: in-house 5nm training silicon, 354 TF BF16/FP16 per D1 die, 25-die “training tile”; AI training focused on Tesla FSD video corpus.
  • Google Trillium TPU v6e + Ironwood TPU v7: Trillium v6e ramping through 2024-25; Ironwood v7 announced for 2026 as Google’s frontier-class TPU, ~4.6 PF FP8 per chip in 9216-chip pods.
  • AWS Trainium2 + Trainium3 UltraServers: 64-chip Trainium2 UltraServer, NeuronLink coherence; Trainium3 mid-2026 with FP4 and ~2× throughput per chip.

The pattern: every frontier-lab build separates training (custom silicon or NVIDIA frontier) from inference (lower-cost custom or older-gen NVIDIA), with a custom fabric stack tuned to the model parallelism shape. A 1024-GPU pod is a small instance of the same pattern.


13. Risk register

  • Silicon supply — H200/B200/GB200 allocation queues remain gated by NVIDIA OEM relationships; 6-12 month lead times in 2025. AMD MI300X/MI325X has shorter lead time but immature FP8 software story.
  • Fabric mismatch — IB NDR feeds H100 fine; B200/GB200 want XDR 800G or Spectrum-X to keep collective ops from bottlenecking. Don’t deploy B200 on QM9700 NDR fabric without modeling the AllReduce loss.
  • Power + cooling — 2.2 MW IT in a single hall is a substation conversation in most markets; greenfield sites need 18-30 month grid lead times. Northern Virginia, Ireland, and Singapore moratoria force new builds to Texas, Phoenix, the upper Midwest, or Sweden/Finland.
  • Software drift — CUDA + driver + NCCL + framework versions interact subtly; a working stack today rarely upgrades cleanly. Pin everything; track NVIDIA NGC release notes weekly.
  • Silent corruption — silent data corruption (SDC) in HBM and NVLink under sustained load is real (Google, Meta both published 2023-2024 papers). Run periodic NCCL bandwidth + DCGM diagnostics in-line with training; auto-rollback on NaN.
  • Vendor concentration — NVIDIA + a single ODM + a single storage vendor is fragile. Dual-source where possible.
  • Compliance / export controls — H100, H200, B200, GB200 are export-controlled to a long list of countries (US BIS rules amended 2022, 2023, Oct 2023, Dec 2024). Confirm shipment destination compliance.

14. Adjacent