GPU Open Analytics Initiative

GOAI strives to create common data frameworks enabling developers and statistical researchers to accelerate data science on GPUs.

[GPU Open Analytics Initiative @gpuoai]

More details on our announcement here https://t.co/aC0uBkpnzU. Thank you to everyone involved, and we look forward to working with @ApacheArrow @ChainerOfficial and hopefully many more communities! Interoperability is key to innovation.

[Matt Harrison @__mharrison__]

Want to speed up your Pandas code by 10-1000x?

With no code change?

The folks from @nvidia have created cuDF pandas accelerator mode. By using this line in Jupyter, you automatically leverage your GPU to run Pandas code:

%load_ext cudf.pandas

From command-line:

python -m cudf.pandas https://t.co/E9XbowBV0h

(It also accelerates 3rd party libraries that leverage Pandas.)

Check out this demo notebook: https://t.co/mS5cTK9Rve

I've had a chance to use the pre-release version and am very impressed!

[RAPIDS AI @RAPIDSai]

100% pandas coverage at up to 150x with 0 code change. Introducing cudf.pandas - for instantly accelerating your current code on #NVIDIA #GPUs. #DataScience . https://t.co/umvVwUv3Ao

[Voltron Data @VoltronData]

Want to leverage #GPUs and optimize workflows for #analytics and/or #MachineLearning? We set out to see if we could achieve both throughput AND high compression ratio without compromising one or the other..…and we did. Read this post to learn how!
https://t.co/YCVMAxDcBI

[Voltron Data @VoltronData]

#OpenSource gives you more choices. #Standards help you make better choices.

This is a key theme in The #ComposableCodex.

“For data systems developers, open source was a lightbulb moment… it meant that engineering teams were not spread so thin trying to innovate across the entire surface area of the data system. Instead, they could focus on innovating in targeted, high-impact areas.

Second, it expanded their choices…[but] the downside of having more choices is that ultimately you do have to make choices.

An evergreen problem with open source is: how do you choose? How do you choose which user interface, query optimizer, or database connector to build with and depend on? That is where standards come in.”

Lots of choices. A good one to make now: Read The Codex Chapter 00 “A New Frontier”! https://t.co/hOyBDEdX0L

#ComposableDataSystems #OpenStandards #OSS

[Josh Patterson @datametrician]

The collective work on the #ComposableCodex is 🔥! @VoltronData is stronger than the sum of its parts. Modular composable accelerated systems start w/ @ApacheArrow @RAPIDSai & more; the future is connected/now. Building bridges not walls! Can’t wait for the next chapters to drop.

[Jiwei Liu @JiweiLiu]

Accelerating Vector Search: Using GPU-Powered Indexes with RAPIDS RAFT | NVIDIA Technical Blog https://t.co/cHfv1jmYPS

[Josh Patterson @datametrician]

Another gift brought to you by the @RAPIDSai team…

[Jeremy Howard @jeremyphoward]

There's an amazingly convenient way to install the *full* NVIDIA CUDA dev stack on Linux, that I've never seen mentioned before.

It's all done with conda!

I just tried it and it worked perfectly.🧵
https://t.co/4GHineaag4

[Josh Patterson @datametrician]

Modularity and compatibility is the future of data systems. If you’re interested in @RAPIDSai @IbisData @ApacheCalcite @ApacheArrow or @substrait_io and at @VLDBconf, you should check out the CDMS talks. @VoltronData will have several there and presenting 2 talks!

[Voltron Data @VoltronData]

Our CEO, Josh Patterson (@datametrician) talks using #OpenStandards to augment data systems on the latest episode of @implydata's podcast. Get his take on @ApacheArrow @IbisData @duckdb and the importance of connectivity in the #DataAnalytics ecosystem. https://t.co/X62rtEKD7l

[Voltron Data @VoltronData]

Preview release of the nanoarrow-based @SnowflakeDB Connector for #Python is available! This connector is ~10x smaller in size and removes a hard dependency on a specific version of #PyArrow. Learn how the nanoarrow integration makes it possible. https://t.co/kWkOZSj0vS

[HPC Guru @HPC_Guru]

DGX GH200: Nvidia ties 256 Grace-Hopper Superchips by 36 NVLink Switches to provide >1 EF FP8 (or ~9PF of FP64)

o 144TB unified memory

o 900 GB/s GPU-to-GPU bandwidth

o 128 TB/s bisection bandwidth

o End-of-year availability

https://t.co/NEIlCG7x5O

#HPC #AI via @HPCwire

[HPC Guru @HPC_Guru]

@HPCwire To illustrate the potential speedups, @nvidia shared internal benchmarking projections, showing improvements from 2.2x (for the 1T GPT3) all the way to 6.3x (for the 40TB Distributed Join)

#HPC #AI #DGX_GH200

[CuPy @CuPy_Team]

📢 CuPy v13.0.0a1 & v12.1.0 is here with more enhancements and improved performance. Highlights include expanded coverage of cupyx.scipy.signal and cupyx.scipy.interpolate APIs, faster random number generation, and more.
https://t.co/Fu0ranWahA

[Voltron Data @VoltronData]

Inside this blog, we share a talk from @gilforsyth at #PyData NYC that breakdowns @IbisData and demos how to get the power of #python interacting with optimized database engines: https://t.co/vMVNCE2SQ7

[anton @abacaj]

How? Use cuml pairwise_distances and warm it up https://t.co/0nDoDJO8kh

[anton @abacaj]

Everyone is out here deploying complex vector databases for semantic search or thinking they need to scale to billions of queries on day one

You can run cosine sim for 150k records (768 emb size) on a single GPU in just ~3ms

[Josh Patterson @datametrician]

This is a great time to plug the @VoltronData summary blog about this. https://t.co/UtWXQJeoZ0 reminder a single @nvidia V100 (now 3 generations old) out performs all CPU implementations by over 2.5x and Spark by 20x… again… on a single GPU.

[Yann Collet @Cyan4973]

TIL : NVidia's nvcomp now supports Zstandard compression and decompression in GPU : https://t.co/XOLFrEbxVI

[Josh Patterson @datametrician]

To new beginnings! So excited to join @wesmckinn and others to help drive the future of @ApacheArrow! Woot woot! Join us!
https://t.co/Vd54mTu9Pc