3 ms·

This page has some details of Tenstorrent's current chips [1]. Looks like a manycore simd design focused on tensor ops from a brief look. Apparently they also h

This page has some details of Tenstorrent's current chips [1]. Looks like a manycore simd design focused on tensor ops from a brief look. Apparently they also have some sort or compression scheme to boost memory bandwidth. 1: https://www.tenstorrent.com/technology/ https://www.tenstorrent.com/technology/

modeless 6y ago
They have a Hot Chips presentation on YouTube with a lot more detail: https://www.youtube.com/watch?v=HLjumOyWj0g https://www.youtube.com/watch?v=HLjumOyWj0g I don't notice anything in particular that stands out vs. the many other AI chips people are making, at first glance. But I'm far from an expert. There are several other technical videos on their YouTube channel as well: https://www.youtube.com/channel/UC7041p6DlAh0r4_Fnlk10pQ https://www.youtube.com/channel/UC7041p6DlAh0r4_Fnlk10pQ
- vnorilo 6y ago
  Thanks for these! My assumption is that hiring Keller means they plan to gain advantage by world class execution rather than some crazy architectural leap of faith.
  - andy_ppp 6y ago
    Watching the video I think turning each tensor into packets is quite clever as you get some ability to send them around a network and organise layers/data manipulation/transforming the layers/compression all as part of the stack. I’m pretty surprised no-one has actually exposed the actor model for parallelising neural networks, it seems it would work quite well and allow you to have a layer per node (or actually many split configurations). Maybe data locality would be an issue with actor based approaches. They seem to be solving this at a lower level but with less knowledge of the actual parallelism in software.
  - jcims 6y ago
    Also Jim was almost certainly the one doing the interviewing, so there’s likely something interesting in there that he felt was worth his time.
BenoitP 6y ago
Their marketing material states: "Facilitating machines to go beyond pattern recognition and into cause-and-effect learning". I wonder what they are referring to. Are they accelerating what SHAP's GradientExplainer [1] does? (namely: crafting inputs at a specific layer, propagating forward to see the influence on class prediction, and sort of backpropagating to pixels) Or is it about something more related to Judea Pearl's work on causality? [1] https://github.com/slundberg/shap#deep-learning-example-with-gradientexplainer-tensorflowkeraspytorch-models https://github.com/slundberg/shap#deep-learning-example-with...
absolutelyrad 6y ago
PiedPiper for machine learning?

modeless 6y ago

They have a Hot Chips presentation on YouTube with a lot more detail: https://www.youtube.com/watch?v=HLjumOyWj0g https://www.youtube.com/watch?v=HLjumOyWj0g I don't notice anything in particular that stands out vs. the many other AI chips people are making, at first glance. But I'm far from an expert. There are several other technical videos on their YouTube channel as well: https://www.youtube.com/channel/UC7041p6DlAh0r4_Fnlk10pQ https://www.youtube.com/channel/UC7041p6DlAh0r4_Fnlk10pQ

vnorilo 6y ago
Thanks for these! My assumption is that hiring Keller means they plan to gain advantage by world class execution rather than some crazy architectural leap of faith.
- andy_ppp 6y ago
  Watching the video I think turning each tensor into packets is quite clever as you get some ability to send them around a network and organise layers/data manipulation/transforming the layers/compression all as part of the stack. I’m pretty surprised no-one has actually exposed the actor model for parallelising neural networks, it seems it would work quite well and allow you to have a layer per node (or actually many split configurations). Maybe data locality would be an issue with actor based approaches. They seem to be solving this at a lower level but with less knowledge of the actual parallelism in software.
- jcims 6y ago
  Also Jim was almost certainly the one doing the interviewing, so there’s likely something interesting in there that he felt was worth his time.

BenoitP 6y ago

Their marketing material states: "Facilitating machines to go beyond pattern recognition and into cause-and-effect learning". I wonder what they are referring to. Are they accelerating what SHAP's GradientExplainer [1] does? (namely: crafting inputs at a specific layer, propagating forward to see the influence on class prediction, and sort of backpropagating to pixels) Or is it about something more related to Judea Pearl's work on causality? [1] https://github.com/slundberg/shap#deep-learning-example-with-gradientexplainer-tensorflowkeraspytorch-models https://github.com/slundberg/shap#deep-learning-example-with...

absolutelyrad 6y ago

PiedPiper for machine learning?