Asit Mishra @asitkumarmishra

AI/ML at Nvidia United States Joined June 2009

Tweets

142
Followers

71
Following

90
Likes

104

Asit Mishra @asitkumarmishra

4 months ago

If you are looking to train in 8bits (MXFP8) on Blackwell then here are some recipe details and what Nvidia’s SW stack (cuBLAS, TE, CUTLASS, etc.) does: arxiv.org/abs/2506.08027. We proved the recipe on 15T tokens for 8B transformer model.

0 0 3 57 0

Tim Cook @tim_cook

2 years ago

WWDC is always one of my favorite moments of the year — and this one is going to be our best ever! Tune in for our special event at 10 a.m. PT

706 3K 31K 3.0M 239

Download Video

Bryan Catanzaro @ctnzr

3 years ago

The Megatron team at NVIDIA is hiring! We're looking for people who can contribute to any aspect of foundation model research and development: nvidia.wd5.myworkdayjobs.com/en-US/NVIDIAEx…

1 28 116 39K 30

Vinod Grover @vinodg

3 years ago

Presenting this at ASPLOS on Monday. :)

0 1 32 5K 0

Download Image

Geoff Langdale @geofflangdale

4 years ago

I wish VP2INTERSECT{D,Q} was fast. uops.info is claiming 46 uops for VP2INTERSECTD on zmm values (around latency 36-41 and throughput 23).

1 1 3 0 0

Asit Mishra @asitkumarmishra

4 years ago

MLPerf 1.0 training results are out and Nvidia is still ruling (no surprises there). TPUv4 looking good too (especially scale out numbers). Most others not even close. mlcommons.org/en/training-no…

0 0 2 0 1

Jonathan Frankle @jefrankle

4 years ago

NEW WORKSHOP: Sparsity in Neural Networks: Advancing Understanding and Practice (July 8-9, 2021). This workshop will bring together members of the many communities working on neural network sparsity to share their perspectives and the latest cutting-edge research (Deadline: 6/15)

4 83 328 0 63

Download Image

Peter McMahon @peterlmcmahon

4 years ago

*How many photons do you need to perform a multiplication operation optically?* It turns out that you can get away with less than 1 photon per scalar multiplication if you're doing large matrix-vector multiplications, as are used in neural networks. 1/3

7 80 492 0 79

Download Image

Asit Mishra @asitkumarmishra

5 years ago

Accelerating Sparse Deep Neural Networks: arxiv.org/pdf/2104.08378… Very detailed discussion on sparsity support in Nvidia Ampere GPU arch and software workflow on how to prune a dense model without losing accuracy and run on sparse TC.