Tech IndustryDec 8, 2023
RivianOwoU27

AI Accelerators

What are the top companies with AI accelerators? I know Ambarella and Nvidia. Meta, Apple, Amazon, Google and Microsoft are always there for their in-house thing. but how is Qualcomm and AMD ? How are startups in this space ? #tech #AI @Qualcomm @AMD

Cisco sancisco Dec 8, 2023

What is ai accelerator?

Intel NwNc78 Dec 10, 2023

Hardware accelerator specifically designed/optimized for use w/ AI/ML workloads/computations. Generally a compute focused GPU.

Cisco sancisco Dec 10, 2023

Is that what pensando does?

Apple welltodo Dec 8, 2023

Excuse me what?

Qualcomm gfs⚛️ Dec 8, 2023

No one else comes close to Ambarella in this area.

NVIDIA Nvidia_ Jan 31

Why do you think so?

Confluent weqre Feb 15

It does not seem Ambarella is making money.

Rivian scarjey Dec 8, 2023

Sambanova, graphcore, mythic

Cisco sancisco Dec 9, 2023

What about Rivos?

AMD pdpi Dec 8, 2023

AMD MI300X - Highest performance accelerator in the market rn. Yes, even better than H100 in Hardware. We lack the software stack. Then there's Intel Haban Gaudi2 and Gaudi3. These are the best ones in the market for Generative AI due to their huge HBM3e memory.

AMD 🥜 makers Dec 10, 2023

We have HIP! It’s a carbon copy of CUDA!

AMD neh02 Dec 9, 2023

AMD has a mi300 and AIE accelerator (xilinx).

Qualcomm gfs⚛️ Dec 9, 2023

My understanding (please comment/correct) - AMD MI300X, Intel Gaudi2/3 target data-center / training workloads …… architecture is GPU based. So, these are trying to replace Nvidia with amazing commercialization & software advantage (CUDA). I think only large companies like Intel/AMD have a slim chance ……. Better odds of success maybe for in-house accelerator from big software companies (Meta, Google etc). Can’t see AI HW startups succeeding at all. Others (including Qualcomm) target on-device ….. mainly for inference (always on, or on demand)…..they opt for NPU based architecture to maximize power efficiency (rather than just performance). Again here as well I think companies with established products have insurmountable advantage over startups. Only AI startups likely to succeed are software/compiler for AI workloads.

Samsung lfD41j Dec 9, 2023

SambaNova, Graphcore, Groq, Cerebras, Tennstorrent, Esperanto.ai, Ventana