Mark Harris

Software Engineer at NVIDIA. Views expressed are my own, not necessarily NVIDIA's. Software leader; developer; miller; builder; brewer; verber.

[NVIDIA AI Developer @NVIDIAAIDev]

👀 NEW feature: @RAPIDSai cuDF supports up to 2.1B rows of text data.

⚡ Watch #pandas code with large strings get GPU-accelerated up to 30x with zero code changes.

Try the notebook ➡️ https://t.co/8ysziZ1bns

#DataScience, #Python, #RAPIDS

[Mark Harris @harrism]

🐊Learn to maximize GPU memory allocation performance and robustness at #gtc2024! Join my session "More Data Faster: GPU Memory Management Best Practices in Python and C++” and steer clear of dreaded OOM errors. 🐍

https://t.co/yQO0RgXrlI

#CUDA #cplusplus #Python #CUDA

[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!

[Mark Harris @harrism]

My colleagues Michael Wang and Thomson Comer will be presenting at #GTC23 about our work on the @RAPIDSai cuSpatial library. Join for a fascinating discussion on GPU geospatial analytics and a killer #Python demo. Don't miss it! Register now: https://t.co/V9MWbMHbPU

[Bradley Dice @Bradley_Dice]

Stream-ordered memory operations are essential for maximizing CUDA performance. We achieve high GPU utilization in @RAPIDSai libraries with tools like these, enabling scalable data science and machine learning workflows. Watch Mark's talk to learn more! #GTC23

[Mark Harris @harrism]

Excited to be (virtually) back at #GTC23! Join me for a deep dive into concurrent #CUDA code with stream-ordered allocation. Learn how to avoid common pitfalls and achieve safe, lightning-fast performance. C++ examples included! Register now: https://t.co/P28nCOGT30

[Conor Shakory @code_report]

Literally laughed out loud while reading this... credit to Jake Hemstad and @randerzander #chatgpt

[Linus ✦ Ekenstam @LinusEkenstam]

Here are some incredible things people are already doing with GPT-4

It’s been less than 3.5 hours

🧵 A thread

[Mark Harris @harrism]

Hmmm, if I prompt it, it chooses a name similar to my prompt./

[Mark Harris @harrism]

After doing this once, #chatGPT now refuses to invent a new name for itself. But the first time, it chose "Ava Skye Frost". After this, it refuses, claiming

[RAPIDS AI @RAPIDSai]

If your work touches GIS at all, you owe it to yourself to read this new blog by Michael Wang, Thompson Comer, Mark Harris and Ben Jarmak. cuSpatial has made big advances, and it keeps getting better and better. https://t.co/tMbnGLE2m3

[Mark Harris @harrism]

@Peter_shirley Slaughterhouse Five - Kurt Vonnegut
To Say Nothing of the Dog - Connie Willis
Recursion - Blake Crouch

[Mark Harris @harrism]

Do you remember when you joined Twitter? I definitely do not. #MyTwitterAnniversary

[Mark Harris @harrism]

Happy with this find at @citybasement books in Melbourne. I like learning new subjects from classic introductory texts.

[Bryce, the CUDA Colonel @blelbach]

If you use Thrust, CUB, or libcu++, come to our user forum on Monday at #GTC22 today to meet the developers and ask us questions! @AllisonVacanti, @g_evtushenko, @Guriwesu, @ericniebler, @code_report and I will all be there!

https://t.co/vYHU7N9GN0

[Bryce, the CUDA Colonel @blelbach]

Want to write portable code that can be GPU-accelerated? Come to my #GTC22 talk to find out how you can with Standard C++! I'll cover all the latest developments - senders & receivers, mdspan, and linear algebra.

https://t.co/UlWvbd7hMX

[Mark Harris @harrism]

@natedoth @Peter_shirley And I love the fact that the little gap you leave to hide imperfection is called the "quirk".

[Mark Harris @harrism]

Neat.

[Freya Holmér @FreyaHolmer]

reminder that you should use exponential interpolation, not lerp, for multiplicative quantities~

if you want to find the frequency, zoom level or scale, halfway between 2 and 8, then the right answer is 4, not 5

lerp( 2, 8, 0.5 ) = 5
eerp( 2, 8, 0.5 ) = 4

[Freya Holmér @FreyaHolmer]

Linear interpolation will behave in non-linear ways when used to control a scalar such as zooming. After doing some math I'd now like to introduce:

Eerp - Exponential Interpolation 📈

(If someone already invented this I will ignore that and still claim this is now called Eerp)