Dissecting the Graphcore IPU Architecture via Microbenchmarking · HackerPapers discussions

Dissecting the Graphcore IPU Architecture via Microbenchmarking — mentioned by kvathupo

Thu, 01 Sep 2022 04:03:29 GMT

Although this seems to only affect the Chinese and Russian governments (as opposed to researchers), I'm generally against economic brinkmanship of this variety. It just festers resentment, and doesn't seem to achieve much: of what little I could find and scan in an hour, it appears that Nvidia GPUs strong-scale pretty well for neural networks ([1], [2]). Although this depends heavily on the NN architecture and training algorithm, Ampere architectures _seem_ to suffer from similar levels of atomic contention as Volta architectures [3]. That said, the throughput for half-precision floating…

Paper and discussion history. A mention is not an endorsement.

Dissecting the Graphcore IPU Architecture via Microbenchmarking — mentioned by winterismute

Mon, 24 Jan 2022 20:49:59 GMT

It's a tough question, it's not even back-propagation but even just sometimes the "parameters" of the models, for example [1] shows that models such as ResNeXt already perform better on a very different architecture such as Graphcore, for some sizes of convolutions. Older models, or models that get tuned for existing GPUs, do not perform as well. It's tough to come up with a new architecture that can have an advantage on current and future models, at least from a peak perf point of view, from a perf/watt for example instead the scaled-up Apple GPUs seem to show new interesting properties. But…

Paper and discussion history. A mention is not an endorsement.

Dissecting the Graphcore IPU Architecture via Microbenchmarking — HN submission

Thu, 14 Jan 2021 14:24:58 GMT

Dissecting the Graphcore IPU Architecture via Microbenchmarking

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Dissecting the Graphcore IPU Architecture via Microbenchmarking — HN submission

Sat, 11 Jan 2020 15:38:05 GMT

Dissecting the Graphcore IPU Architecture via Microbenchmarking

3 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Dissecting the Graphcore IPU Architecture via Microbenchmarking — HN submission

Fri, 20 Dec 2019 08:58:27 GMT

Dissecting Graphcore's IPU Architecture via Microbenchmarking

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Dissecting the Graphcore IPU Architecture via Microbenchmarking — HN submission

Sat, 14 Dec 2019 14:48:24 GMT

Dissecting the Graphcore IPU Architecture via Microbenchmarking

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Dissecting the Graphcore IPU Architecture via Microbenchmarking — HN submission

Wed, 11 Dec 2019 06:48:09 GMT

Dissecting the Graphcore IPU Architecture via Microbenchmarking

1 point; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.

Dissecting the Graphcore IPU Architecture via Microbenchmarking — HN submission

Tue, 10 Dec 2019 16:50:29 GMT

Dissecting the Graphcore IPU Architecture via Microbenchmarking

2 points; 0 discussion replies on HN. Paper and source history. A submission is not an endorsement.