Jump to content

Data processing unit

From Wikipedia, the free encyclopedia

This is an old revision of this page, as edited by 2601:647:4802:2340:51f:2a92:ce76:3e43 (talk) at 21:51, 20 July 2020 (nvidia gpu). The present address (URL) is a permanent link to this revision, which may differ significantly from the current revision.

A data processing unit (DPU) is a programmable electronic component that processes streams of data.[1] The data is transmitted to and from the component as multiplexed packets of information. DPUs have the generality and the programmability of central processing units but are specialized[2] to operate efficiently on networking packets, storage requests or analytics requests.[3][4]

A DPU differentiates itself from a CPU by a larger degree of parallelism (required to process lots of requests) and from a GPU by a MIMD architecture rather an SIMD architecture (required as each request needs to make different decisions and follow a different path through the chip).[5][6]

References

  1. ^ "A Peek Inside Fungible, A Data Center Startup From Juniper Cofounder Pradeep Sindhu - Packet Pushers". packetpushers.net. Retrieved 2019-08-06.
  2. ^ "To handle the data tsunami, Juniper's founder turns to custom silicon". SiliconANGLE. 2018-11-30. Retrieved 2019-08-06.
  3. ^ 20190012350, Sindhu, Pradeep; Frailong, Jean-marc & Noureddine, Wael et al., "Data Processing Unit for Stream Processing", issued 2019-01-10 
  4. ^ Hennessy, John. “The End of Moore's Law, CPUs (as We Know Them), and the Rise of Domain Specific Architectures.” AI Hardware Summit. Stanford University. September 29, 2019.
  5. ^ "The renaissance of silicon will create industry giants". TechCrunch. Retrieved 2019-08-16.
  6. ^ https://blogs.nvidia.com/blog/2020/05/20/whats-a-dpu-data-processing-unit/