@gabriel1: 100k h100 datacenter ballpark numbers, so you know the magnitudes rounded to numbers that are easy for quick mental mat…
Summary
A quick breakdown of ballpark numbers for a 100k H100 GPU datacenter, covering GPU costs (~$3B), full datacenter build (~$5B), power consumption (~0.2GW), and annual energy costs (~$50M).
Similar Articles
GPU Compass – open-source, real-time GPU pricing across 20+ clouds [P]
SkyPilot team open-sources a continuously updated catalog that tracks on-demand and spot pricing for 50 GPU models across 20+ clouds, now browsable online.
@sudoingX: this is a laptop running a 31b parameter model at 99% gpu autonomously through hermes agent, 15 tok/s sustained, 22.8 o…
A 31B parameter model runs locally on a laptop via Hermes agent at 15 tok/s, using 22.8 GB VRAM and 94 W power, highlighting fully autonomous, private AI inference without cloud dependencies.
Taiwanese company Skymizer announces HTX301 - PCIE inference card with 384GB of Memory at ~240 Watts
Skymizer announces the HTX301, a PCIe inference card capable of running 700B-parameter LLMs on-premises with high memory and low power consumption.
Efficiency at Scale: NVIDIA, Energy Leaders Accelerating Power‑Flexible AI Factories to Fortify the Grid
NVIDIA and Emerald AI unveiled a collaborative approach at CERAWeek to treat AI factories as flexible grid assets, improving energy efficiency and grid reliability through intelligent power management. The initiative partners with major energy companies to optimize AI workload operations based on grid conditions while maximizing tokens per second per watt.
@outsource_: NEW GLM+ QWEN 18B RUNS ON CONSUMER GPU IT BEATS 35B MoE AT HALF THE VRAM @KyleHessling1 just dropped the healed Qwopus-…
A new 18B merged quantized model, Qwopus-GLM-18B-GGUF, outperforms 35B MoE models while using half the VRAM and running on consumer GPUs.