跳到正文
TechPowerUp· DrUsmanSaleem·· 13 小時前AI 評分34

Phison 展示可在 SSD 內運行 AI 框架的設計

Phison Previews an SSD That Runs AI Frameworks on Its Own NPU

AI 導讀

Phison 展示一款搭載 NPU 的企業級 SSD 設計,可在硬碟本身運行 AI 框架。該設計配備 64 GB 板載記憶體,速度為 130 GB/s,支援 PyTorch、vLLM、OpenVINO 和 llama.ccp;Phison 表示,適合的工作負載可減少 10 至 100 倍數據傳輸。產品仍在開發中,現階段應視為概念驗證。

正文

Everyone's going all in on AI, so SSD manufacturers were not going to be left behind. Phison has detailed an enterprise SSD design that can run AI frameworks on the drive itself. At the Supercomputing 2026 (SC26) event in Chicago, the company showed off the computational storage design that hosts both an NPU and 64 GB of onboard memory running at 130 GB/s. It also supports PyTorch, vLLM, OpenVINO, and llama.ccp. Phison says this SSD could cut data transfers by 10 to 100 times for suitable workloads. It achieves that by scanning and processing the dataset in parallel, so it only sends back the relevant results instead of moving the entire dataset over the host. The company uses DMIPS to rate its NPU performance instead of using TOPS, which is a bit unusual.

Phison has shown off other AI-focused storage projects before, and this SSD is the latest one in that chain. At the same event, Phison showcased a Mixed media design that removes the need for dedicated cache drives by putting high-endurance pSLC and QLC flash behind a single 16-channel controller. Phison says that losing one drive in a 24-drive server would cost only 4% of the cache instead of the usual 25%. Its showcase also included the PCIe Gen 6 X3 controller that was first shown at Computex. As for the SSD with the built-in NPU, it is sparse on details for now. We don't know what kind of controller, interface, form factor, or memory type it uses or plans to use. These are products that are still in development, so it's best to treat them as proof of concepts for now.

來源:TechPowerUp · techpowerup.com