‹ BackHN Continuity

Thread

How GLM built its own inference infrastructure

411 points · 285 comments · whiteros_e

  1. zicohacks · · focus · HN ↗
    US chip export restrictions may actually be an advantage for China's AI Infrastructure. Chinese companies are forced to speed up developing their own AI chips
    1. menaerus · · focus · HN ↗
      It was evident that this will happen.

      > Compared with our initial baseline on the same hardware, we achieved a 3× improvement in end-to-end serving performance, reaching hardware efficiency and per-token cost comparable to mainstream NVIDIA GPUs. This demonstrates that Chinese chips can support frontier-model inference efficiently and economically at scale.

      1. HSO · · focus · HN ↗

        [dead]

        1. ux266478 · · focus · HN ↗
          The mainstream consensus on PRC silicon has been "the gap exists, but is closing" for a long time[1][2]. I think you might be confusing the algorithmically-cast shadows on the cave wall that you're sat in front of for a reasonable and unbiased sample. Can you provide any reason to believe that your data points are actually generalizable? From your tone it sounds more like you're a victim of getting outrage baited into engagement, and mistaking observations made in that envelope for a high quality data source. You gotta watch out for that mate.

          [1] - <a href="https:&#x2F;&#x2F;www.eastwestcenter.org&#x2F;sites&#x2F;default&#x2F;files&#x2F;private&#x2F;iegwp001_1.pdf#:~:text=move%20from%20catching-up%20to%20forging-ahead%20in%20semiconductors" rel="nofollow">https:&#x2F;&#x2F;www.eastwestcenter.org&#x2F;sites&#x2F;default&#x2F;files&#x2F;private&#x2F;i...

          [2] - <a href="https:&#x2F;&#x2F;www.usitc.gov&#x2F;publications&#x2F;332&#x2F;journals&#x2F;chinese_semiconductor_industrial_policy_prospects_for_success_jice_aug_2019.pdf#:~:text=China&#x27;s%20gap%20with%20leading%20international%20semiconductor%20firms%20has%20consistently%20narrowed%20over%20time" rel="nofollow">https:&#x2F;&#x2F;www.usitc.gov&#x2F;publications&#x2F;332&#x2F;journals&#x2F;chinese_semi...

          1. HSO · · focus · HN ↗

            [dead]

Open on Hacker News to reply ↗

Unofficial Hacker News client; not affiliated with Y Combinator.