paresh

Product management - Artificial Intelligence, High Performance Computing, GPU

[David Bennett @DavidBennett__]

https://t.co/VcdUENlt2s

[alex @kvamme]

Hard to describe until you try it. I would tag the Taalas team but they are so locked in they don't have X accounts.

[Sally Ward-Foxton @sallywf]

AI chip startup Taalas @taalas_inc is showing off a chip that can do 16,000 tps/user on Llama3.1-8B, many multiples of its nearest competitor. The catch? The chip ONLY runs Llama3.1-8B, and a model like DeepSeekR1-671B would need 30 separate tapeouts:
https://t.co/IJuprQZqaE

[Bryan Catanzaro @ctnzr]

Language models do just keep scaling!

Today we’re announcing Megatron-Turing NLG: a 530B parameter language model.

Joint work from @NVIDIA and @Microsoft. Trained using Megatron and DeepSpeed on DGX SuperPod.

https://t.co/xFN7c8xYQR

[Bryan Catanzaro @ctnzr]

Megatron running on Selene sustains 140 TFlops per GPU on 1024 GPUs while training GPT-3.

That's 6X faster than the numbers reported in this paper, at about the same energy consumption.

Give Megatron a try and lower your CO2 emissions.

https://t.co/6Ht56c71lB

[paresh @pareshkharya]

MLPerf Inference 1.0 results prove the incredible efficiency of A100 MIG in boosting infrastructure utilization for AI. A single A100 ran all seven MLPerf Inference 1.0 tests in offline mode simultaneously while providing 98% of the performance of a single MIG running alone.

[paresh @pareshkharya]

Q&A along with vangel Bojaxhi from Inspur looking into AI in 2020 and expectations for 2021 and beyond.

https://t.co/jv8weuJeSb

#EdgeComputing #AI

[James Wang @draecomino]

Nvidia is now worth more than Intel + AMD combined.

That's the market's way of saying: x86 is a legacy architecture.

The workload of the future is SIMULATION.

Light -> raytracing
Physics -> n-body, CFD
AI -> neural nets

..and these will run on GPUs.

[paresh @pareshkharya]

Congrats Microsoft team on great results achieved using V100 GPUs. Reducing time-to-train and cost-to-train continues to be vital for progress of AI. Cannot wait to see what A100 GPUs will enable for these workloads at scale.

https://t.co/fkomPIznHE

[paresh @pareshkharya]

Now available our whitepaper providing detailed architecture information on our newly launched NVIDIA Ampere Architecture and A100 GPUs
https://t.co/4OEOtMPwRm

[NVIDIA GeForce @NVIDIAGeForce]

PC Gamers, let’s put those GPUs to work.

Join us and our friends at @OfficialPCMR in supporting folding@home and donating unused GPU computing power to fight against COVID-19!

Learn more → https://t.co/EQE4u7xTZT

[paresh @pareshkharya]

Eagerly awaited MLPerf Inference results are out. NVIDIA Turing GPUs, Xavier SoC emerge toppers in their respective categories. Read our blog to learn more.

https://t.co/DU1YptEcLd

[paresh @pareshkharya]

NVIDIA research team trained the largest ever language model based on transformers with a novel model parallel approach using Pytorch. The code that implements this approach is published on GitHub.
https://t.co/lKvRyUhysV

[paresh @pareshkharya]

A huge milestone in natural language processing. Software optimizations used to accomplish these breakthroughs in conversational AI available to developers.

https://t.co/9PxEFqLMJv

[paresh @pareshkharya]

Our Safety Force Field (SFF) vehicle software is designed specifically for collision avoidance. Watch how it double-checks vehicle controls and vetoes unsafe actions. #DRIVELabs https://t.co/0EDS9vm9eH

[paresh @pareshkharya]

NVIDIA has set new records in #MLPerf, breaking 8 performance records- 3 at maximum scale and 5 per accelerator. https://t.co/R1Bu5lTt4O

[paresh @pareshkharya]

Learn how to annotate, build, and adapt #AI models for #medicalimaging with the Clara Train SDK in this technical blog. https://t.co/5AthH30EMt

[paresh @pareshkharya]

Important milestone in GPU accelerated Inference. NVIDIA TensorRT plugins, parsers, & samples are now open source. Easy to extend now with your custom models & layers. https://t.co/GrOxw3U7ZP

[paresh @pareshkharya]

Learn how to use NVIDIA's NGC Container Replicator to automatically download containers, ready-to-run on Singularity. #ISC19 https://t.co/BEhvSMVa1j

[paresh @pareshkharya]

World's fastest Summit supercomputer achieves almost half an exaflop of performance. Tripling its performance record. #ISC19 https://t.co/NOZz6f7ZJX