Urgent.News

What's breaking now, across thousands of outlets.

Tech

H100 PCIe, SXM or NVL: which one are you renting, and how to check from the shell

A listing that says "H100 80GB" can mean two different boards, and "H100 94GB" can mean two others. They differ in memory type, bandwidth, power and how the GPUs talk to each other, and so does the price. This post sets out the variants from NVIDIA's own documents, then shows how to tell which one you have from a shell, with real output. I run VoltageGPU, which rents H100s by the hour . The H100…

Three variants of NVIDIA H100 GPUs are available for renting: H100 SXM, H100 NVL, and H100 PCIe. Each variant differs in memory size, bandwidth, power consumption, form factor, and GPU-to-GPU connection. The H100 NVL (94GB) and H100 SXM (80GB) have higher bandwidth compared to the H100 PCIe (2,000 GB/s), making them more suitable for tensor-parallel work across many GPUs.

The H100 SXM can draw up to 700W, while the H100 NVL and H100 PCIe have 350-400W and 310W (in power-connector mode) power limits, respectively. The PCIe variant costs $1.99 per GPU-hour, the H100 NVL $2.59, and the H100 SXM $2.69 on demand.

Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at dev.to →

More in Tech

One Rust Patch, Two Workloads: Catching Regressions with CodSpeed

You’re reviewing a Rust pull request that changes how a batch job counts unique event IDs. The tests pass, and on the first benchmark the patch uses about 24% fewer instructions than the original…

  • Rust patch replaces BTreeSet with small Vec, reducing instructions by 24%
  • Patch incurs 3.28x more instruction references with unique IDs benchmark
  • Lab tests patch on two batches: 2,048 IDs each, one with repeats, one unique

More from Friday 20 November →