PDP-11🚀 – Telegram
PDP-11🚀
413 subscribers
112 photos
28 files
146 links
AI Hardware & Domain Specific Computing

#FPGA #ASIC #HPC #DNN

@vconst89
Download Telegram
What is in-memory computations
Running on servers in SK Telecom’s data center,
Alveo U250 cards have demonstrated improved
throughput and increased accuracy for Tview theft
detection services, enabling the company to provide
more customers with access to real-time, AI-based
security services, and a more reliable defense against
security threats.
SiMa.ai is a new company developing a product to fit these requirements. Its MLSoC provides a custom ML accelerator that is rated at 10x better power efficiency than leading GPU-based designs and 2-3x better efficiency than the best accelerators available today. The MLSoC also includes a popular ARM-compatible CPU for legacy code and standard DRAM, Ethernet, and camera interfaces to simplify system design. Companies developing camera-based edge systems should consider the MLSoC for adding intelligence to their next design.
Linley2020_Cerebras_v02.pdf
3.2 MB
Linley2020_Cerebras_v02.pdf
Not an ML topic, but still about hardware. Hardware security evaluation of MAX10 intel FPGA. Invasive and non-invasive attacks
Huawei released MindSpor, Tensorflow style AI computing framework, in August 2019.

Announced support of the following hardware platforms: Ascend 910, high performance AI processor for training, and Ascend 310, power efficient AI processor. Probably, it should also include support of Kirin990, Huawei SoC of mobile devices.

On the last week Huawei announced that Mindspor is open source now.
Eetimes podcast

KEVIN KREWELL: Well, to my mind, once again, the Linley Conference (April 6-9) got taken over by AI companies. And in this case, I think the primary highlights were the AI companies. We got to see a new company, Tens Turnt (?), come to light and talk about their solution. Grok finally really opened up the kimono and talked about the internal architecture of their chip. And they were a show, not a no-show this time. The small guys were making progress. The very low power guys like Gray Matter, they’re all moving forward with their products. And I think were a number of IP companies. Actually, interesting, Flex Logic, which has been an IP company for FPGAs, sort of pivoted a bit. So they’re still making their FPGA stuff, but they’re also building dedicated chips for AI using a DSP core they created that works well for machine learning applications. So they actually went ahead and built their own chip, and they’re showing some great promise in terms of performance.

It was a lot of interesting technologies. A lot of the IP guys like Seefa had some really good DSP. ARM talked a little more about their accelerator for MCUs for Ethos U55, which is their machine learning accelerator. Not new, but it’s a key product dealing with ARM, which is a key vendor.

And then something completely different now was the RISC-V guys talking about vector processing (?). That’s the next wave of architecture changes for the RISC-V guys. They’re getting into vector processing and tightly coupling it to the instruction set of RISC-V CPUs.