ML Compilers & Runtime
San Jose, CA
Joined January 2017
- Tweets52
- Following227
- Followers90
- Likes3.1K
Unpopular opinion:
“Everything happens for a reason” is the most successful post-hoc rationalization in human history. We don’t find meaning. We manufacture it, then forget we built the factory.
An A100 finishes the math for a single 70B decode step in 0.45ms. Then it waits 69ms for weights to load from memory. Your GPU is 99% idle during inference. Here's why.
Life is a war between you and your mind. Winning is just refusing to let it beat you, every single day.
Om Thakkar retweeted
Looking forward to working with Intel on the Terafab!
Intel is proud to join the Terafab project with @SpaceX, @xAI, and @Tesla to help refactor silicon fab technology.
Our ability to design, fabricate, and package ultra-high-performance chips at scale will help accelerate Terafab’s aim to produce 1 TW/year of compute to power future advances in AI and robotics.
It was fun hosting @elonmusk at Intel this past weekend!
Om Thakkar retweeted
Intel is proud to join the Terafab project with @SpaceX, @xAI, and @Tesla to help refactor silicon fab technology.
Our ability to design, fabricate, and package ultra-high-performance chips at scale will help accelerate Terafab’s aim to produce 1 TW/year of compute to power future advances in AI and robotics.
It was fun hosting @elonmusk at Intel this past weekend!
Om Thakkar retweeted
Breakthrough performance—read how Intel Gaudi 2 AI accelerators delivered on training speed and inferencing in this new blog post from @StabilityAI. intel.ly/3Vfphts
#Intel #GenerativeAI #LLM
ALT Image displaying statistics on Intel Gaudi 2 accelerators outperform the competition for training on Stable Diffusion 3. 1.5x faster than NVIDIA H100 3.0x faster than NVIDIA A100 Disclaimer: Visit the Stability AI blog for details. Result may vary.
Om Thakkar retweeted
GPUs get the glory for training, but our CPUs quietly shine where it counts—in AI inference. 🤓
Om Thakkar retweeted
Why do people want to use GPUs vs. CPUs for even relatively small LLM models? @VMware shows the difference in performance in an Ice Lake CPU without AMX and a Sapphire Rapids CPU with AMX. It goes from unuseable to useable. @IntelBusiness #AIFD4
Om Thakkar retweeted
TensorFlow 2.16 will be released in a few weeks -- it's the first release of TensorFlow where Keras 3 is the default. If you'd like to stay on Keras 2 past that point, you should make sure to install the tf-keras package. github.com/tensorflow/tensor…
Om Thakkar retweeted
Learn how to take advantage of Intel’s new AI Acceleration Engine called Advanced Matrix Extensions and oneAPI libraries to improve TensorFlow performance on 4th Gen Intel Xeon processors ➡️ goo.gle/3ID4n0M
ALT Intel Xeon 4th Gen Intel Xeon Scalable processor
Om Thakkar retweeted
The Intel team has worked very hard on this and I'm super happy that it's finally out! ☺️ Read how TensorFlow got up to 19x faster on the new Sapphire Rapids CPU with AMX instructions here:
Learn how to take advantage of Intel’s new AI Acceleration Engine called Advanced Matrix Extensions and oneAPI libraries to improve TensorFlow performance on 4th Gen Intel Xeon processors ➡️ goo.gle/3ID4n0M
ALT Intel Xeon 4th Gen Intel Xeon Scalable processor
Om Thakkar retweeted
.@intel and @Google have joined forces to accelerate #AI operations with @TensorFlow on oneDNN integration. @VentureBeat goes in-depth with what this means for data scientists and application developers. intel.ly/3MeJ99s
#oneAPI #datascience
Om Thakkar retweeted
There is a reason why: Amazon, Google, PhonePe, PayTM, CRED, are all fighting for the UPI 2.0 pie.
Is it: Bigger, Better, Faster?
[A thread...]
Om Thakkar retweeted
This is big step forward for scaling AI everywhere!
TensorFlow 2.9 has been released! 🎉
Highlights include:
👉 Performance improvements with oneDNN
👉 Release of DTensor, a new API for model distribution that can be used to seamlessly move from data parallelism to model parallelism
Learn more ↓ goo.gle/3sHhp4A
Om Thakkar retweeted
Indian roads get a fast charge on July 16th.
#PerformanceIsAnAttitude #AudiRS7sportback #AudiRS7
#StayTuned
Om Thakkar retweeted
This candle is lit!
Liftoff, we have liftoff!!! 🚀
After nearly a decade, American astronauts launch from American soil. Such an extraordinary feat of human spirit and ingenuity. Godspeed Bob and Doug! The world is watching, and we are with you. #LaunchAmerica
