Keith Kraus

🎉CUDA meet-up in NYC on November 13 with @floxdevelopment. 🙌 Be sure to reserve your spot, and we will see you there!

Calling all CUDA developers! The CUDA team will be in London on June 5 to host a free meetup for all skill levels. Chat with the people behind CUDA - we'll answer your questions! @aterrel & I will present on CUDA Python. Register now - space is limited! nvda.ws/45tlrBV

We've announced cuTile, a tile programming model for CUDA! It's an array-based paradigm where the compiler automates mem movement, pipelining & tensor core utilization, making GPU programming easier & more portable. I'm proud of my stellar team for all their hard work on this!

It's the year of CUDA Python.

Curious about the latest in high-performance #Python? Introducing Numbast, a new tool that automates Numba bindings, bridging the gap between CUDA C++ and Python for seamless performance. ➡️ nvda.ws/4f30fVt Dive into our technical blog to see it in action.

Voltron layoffs, super unfortunate. You should hire Felipe.

During my 2y at @VoltronData, I worked on @ApacheArrow C++ (and PyArrow) spec’ing and implementing new types (list-view and run-end encoded arrays), expanding the fs abstraction, improving compute kernels, unblocking contributors by mentoring them on many areas of the project.

As you may have heard, there has been a big layoff at @VoltronData. I was one of the affected people together with 50+ others. It’s an exciting time for Data Engineering and Compilers targeting AI accelerators, so I’m curious to see what opportunities are available.

After 1 year at Voltron Data, I found myself along with my teammates and 50+ people in need of a job I have plenty experience building OSS Python data tools, presenting at conferences and OS community management. If you think I'd be a good fit for a job please let me know

🚀 Heading to @Oracle #CloudWorld 2024? Book a meeting with our execs @datametrician, @rodaramburu & @darrenhaas on Sept 10-11. Spots are limited! airtable.com/appWDPBDUhIP0v… 📍 Visit us at Booth #115 for a deep dive into accelerated compute with @NVIDIA #GPUs. #OCW2024

@krismicinski @RyanMarcus For scaling joins on GPUs: adms-conf.org/2021-camera-re… For less papers and more OSS implementations: * GPU hash maps: github.com/nvidia/cucolle… * GPU DataFrame Library with various join implementations: github.com/rapidsai/cudf * GPU DBMS: github.com/heavyai/heavydb

Encoding and Compression Guide for Parquet String Data Using @RAPIDSai @nvidia developer.nvidia.com/blog/encoding-…

It's my first ever @SciPyConf and I've been blown away at the presenters, audience, and community that keeps it all going. I was honoured to give a talk on @ApacheArrow nanoarrow! Check out the slides and/or repo if you're interested: dewey.dunnington.ca/slides/scipy20…

Join @gilforsyth tomorrow July 12th at SciPy Conference to learn about @IbisData . We've been told there will be a live demo! cfp.scipy.org/2024/talk/CKVV…

This Thursday, @wesmckinn will discuss what he thinks dataframes and data systems will look like in the future. Wes is the creator of some of the most popular data engineering tools, including pandas, Apache Arrow, and Ibis. He’s also a core contributor to Apache Parquet. Wes will start with a short presentation about his work, followed by a Q&A session. The talk will be hosted by @chipro. Let us know what questions/topics you would like Wes to discuss. Time: 12pm PT, Thursday, June 20 RSVP: lu.ma/vkd8h5nu

@mim_djo @RitchieVink @mrocklin This has been implemented! github.com/ibis-project/i…

@RitchieVink @mrocklin damn it, create view is not supported in polars ibis backend 529 def create_view(self, *_, **__) -> ir.Table: --> 530 raise NotImplementedError(self.name) 531 532 def drop_table(self, *_, **__) -> ir.Table: NotImplementedError: polars

TPC-H Cloud Benchmarks: Spark, Dask, DuckDB, Polars Across scales: 10 GiB, 100 GiB, 1 TiB, 10 TiB Hardware: MBP and AWS It was a fun experiment. No project wins uniformly. DuckDB and Dask do pretty well. docs.coiled.io/blog/tpch.html

holly Cow, I just witness a miracle, @IbisData is the most underrated data engineering package WTF is this sorcery !!!

Awesome to see @RAPIDSai 24.04 officially launches cuVS (the library to accelerate vector search on GPU), in addition to pandas 2.0 support, and significant expansions of @networkx_team GPU acceleration! medium.com/rapids-ai/rapi…

Ibis in the browser? Try it out! (Warning: very experimental, doesn't work on iOS, and we'll be working on UX improvements going forward!): ibis-project.org/tutorials/brow…

@rf @hpcprogrammer Yes, many of the queries here are running large hash joins that run larger than both GPU memory and host memory where we gracefully spill to run out of core. Large sorting would work similarly, just with a different work partitioning strategy.

@hpcprogrammer Does Theseus do external-memory joins with hashing/sorting on the GPU? I remember seeing some benchmarks on the Theseus site mentioning good join perf. (Which is good, large joins seem to be a place a lot of otherwise good analytics tools get weird.)