# @storagereview on Instagram

- **Type:** Image
- **Original URL:** https://www.instagram.com/p/Db82b1IxGq_
- **Gondola URL:** https://gondola.cc/posts/68970059-storagereview-instagram
- **Thumbnail:** https://img.gondola.cc/tr:w-,h-,fo-auto/postThumbnails/9482b78f84.jpg
- **Posted:** 2026-08-12T18:52:54.000+00:00
- **Account Owner:** StorageReview (@storagereview) — https://gondola.cc/storagereview

## Caption

In hand at Dell Technologies World: an AMD Instinct MI350P engineering sample, the PCIe card at the center of Liqid’s new UltraStack 30. Each one of these carries 144GB of HBM3E and 2.3 PFLOPS of dense FP8 compute in a passively cooled, double-slot card that drops into a standard server, no exotic cooling required. Liqid’s platform pools thirty of them into a single system: 4.3TB of GPU memory and 69 PFLOPS for AI inference, with GPUs allocated on demand instead of stranded in fixed boxes. The company claims up to 3.7x more tokens per second than traditional server configurations. Full story at the link in bio.

#AMD #AIInference #GPUpooling

## Stats

- **Views:** 0
- **Likes:** 127
- **Shares:** 0
- **Comments:** 1

## Tags

amd, gpupooling, aiinference

---
Copyright (c) Gondola