Ted Lee

Computer Science Enthusiast, Engineer by Heart

[Ted Lee @tedthethinker]

Recursion, next ===>

[Ted Lee @tedthethinker]

@gaunernst "quite a gap", but at least 90% of CuBLAS? Wat? 😅

[Ted Lee @tedthethinker]

@mattrothenberg That looks fun, but now imagine needing to do that many times a day for weeks straight 😅

[Ted Lee @tedthethinker]

ARM often wins on efficiency in many products, but I don't think that comes from the ISA alone.

Curious what folks think about x86 SBCs like the LattePanda when discussing x86 vs ARM efficiency 👀

[Ted Lee @tedthethinker]

@RodrigoFMaga @PythonPr Is that a (joke) reference about universal gates? 👀

[Ted Lee @tedthethinker]

@JenMsft :qall!
yy
:wall
i

... wait we're not talking Vim specifically? 😅

[Ted Lee @tedthethinker]

@lcamtuf Nah, `M365Copilot.exe` is "superior" (with emphasis on the quotes)

[Ted Lee @tedthethinker]

@jino_rohit ... Verilog and SystemC so that I can build my own accelerators ofc, 🤣

Yes I'm joking; I'm not actually crazy lmao

[Ted Lee @tedthethinker]

@mejitwo Here's one from Wikipedia; did screenshot instead because I doubt that majority of people on X understand LaTeX 😅

[Ted Lee @tedthethinker]

@MainzOnX Decision Trees boosted or bagged can still be pretty good from what I understand 👀

[Ted Lee @tedthethinker]

@yaboilyrical Now please show us the validation loss over time or even the final test accuracy 👀

[Ted Lee @tedthethinker]

@hiarun02 Wait you're joking right? I remember at least a year ago when I burned through max conversation length in 2 turns with free tier Claude because I was testing it with a custom parallel mergesort asymptotic analysis task, but I had no idea that it was still this bad? 👀

[Ted Lee @tedthethinker]

@gpuwaster @SzymonOzog_ When the meme is underspecified 😅

[Ted Lee @tedthethinker]

@gpuwaster @SzymonOzog_ I'm sorry I don't understand; if "inline PTX is all I need" then won't that essentially imply that I write *all* my device code in PTX?

[Ted Lee @tedthethinker]

@jimkxa Interesting point. But I was just reminded of Intel’s Larrabee though: is that a case where "simpler" looked appealing architecturally, but in practice the complexity just reappeared elsewhere?

[Ted Lee @tedthethinker]

@techbromemes *Mentally flashes back to when I ported a version of a custom CUDA C/C++ mergesort from Linux to Windows and got nearly a 2x "mysterious" performance boost... until I later changed my `long`-declared indices to `long long`* 😅

[Ted Lee @tedthethinker]

DLSS 5 isn't just about image quality; it's about control + ergonomics.

- If it integrates cleanly ➡️ RTX moment
- If it fights artist intent ➡️ NV1 quadratics moment

Curious how this plays out 👀 #DLSS

[Ted Lee @tedthethinker]

@FelixCLC_ You mean fixed point calculations? Aren't CPUs and GPUs already capable of that?

[Ted Lee @tedthethinker]

@tlbflush Heard though that the software stack for these devices weren't very optimized though especially when it came to compiler based auto vectorization 👀