raphlinus

I think we're talking past each other to some extent. Putting aside the question of how misleading it is to market a 16 bit multiply as a "TF32" operation, this is all about tradeoffs. …

Thanks for the matmul sizes, but the question I am more interested in is precision. Matrix multiply throughput can be dramatically higher for tensor cores[1] than normal shader ALU, specifically for r…

I have questions! What is the size of the matrix elements? My understanding is that float32 is super-easy to represent in SPIR-V, and that float16 is possible but the path is not as smoothly paved. Of…

So... separating front-end from back-end in and of itself is not necessarily a problem, but using async to do it adds complexity that is not worth it. Originally, I was hoping that a CRDT or similar…

I'm happy to see them cite my Rope Science series. Looking back, there were some things that xi did wrong, most importantly separating the front-end from the back-end with async communication, bu…

https://twitter.com/gidmk/status/1422044335076306947 is probably the best source on this. GidMK has made something of a name for himself debunking this stuff, for better or …

If you dig into that story, it's just a lab result that IVM shows antiviral activity against Omicron. It does not support your implied claim that Japan is actually using IVM. Here's a source…

A lot of it is just straight-up fraud. Gideon Meyerowitz-Katz has documented this in some detail. Also, ivnmeta itself is just weird. It can only be characterized as a propaganda site. It's well …

You rang? [3148]: https://slashdot.org/~raph

That should just render correctly thanks to the BiDi algorithm. The "override" control characters are a heavy hammer, and are extremely rarely needed. In fact, at this point I think it'…

> (you can’t even transpile OpenCL C 1.2 to it fully, with quite heavily limited pointers…) Is that still true as of Vulkan 1.3? We have buffer device address now, which I think counts as a real po…

It's not just easier to use, but more powerful in many ways. A huge advantage of CUDA is the ecosystem of high quality libraries, including cuBLAS, cuDNN (machine learning), CUB, and Thrust. In t…

Adequate for what purpose? My personal feeling is that if you are sufficiently motivated, you can get your work done in them, but they are painful and tedious enough that in most cases your GPU is sit…

While it has roots in C, I think it's better to consider it a separate language, especially in more modern forms (Metal is C++14). Metal is locked into Apple's proprietary platforms, but the…

Lots of problems with this ( https://xkcd.com/386/ ). "All shader languages are very similar to C". Even counting only mainstream shader languages, MSL is explicitly base…

I look forward to reading that report!

Makes you wonder why they ended the collaboration in November and didn't continue to retain him to validate that the new stuff lives up to its claims, doesn't it?

It's funny how the comments here are polarized, some of them claiming that Jepsen slaughtered RDX, others that it proved that the consensus layer is rock solid. Let's appreciate this for wha…

Hi Anthony! Yes, I'm really pleased with the way it started as an individual project and turned into a collaboration. And I think it's likely that Ken's detailed and excellent investiga…

It's not bad, but I've heard that the person who started it is a jerk who constantly posts self-serving comments on HN, and always seems to pop up on threads about the DX7.

Really well said, thank you. I think the lesson here is that one shouldn't listen either to antivax grifters or to greedy pharma execs who oversell their products because of the enormous incent…

This is a thoughtful and thought-provoking blog post. I think it's worth asking similar questions of computer science. I think you'll find some math-like patterns -- there's basically n…

To me, that's actually one of the more interesting questions. It's possible to grade the output of the AI against objective criteria, like does it run, and resources consumed (RAM, CPU time,…

This is correct, though I can certainly see how someone would come to the conclusion that starting colors are not identical, as default painting mode is less than full opacity, so the mixed color is g…

This is a great paper, will likely move the state of the art forward. I've been aware of Kubelka-Munk representations of painting surfaces since [1], which used a six-dimensional space (scatterin…

In the spirit of being charitable, I would say that the article is highlighting a place where the guarantees are incompletely documented. And not having a solid specification and clear documentation i…

I don't think I come across as an alien. For example, I'm very deeply involved in my local Quaker meeting, and I don't think any of that is particularly different than others. That said…

Unusually, we have pretty good documentation about Prof Tao's background, childhood, and mathematical education. I found this a good read. http://math.fau.edu/yiu/Oldwebsites…

Danny is correct here. I was aware of it at the time (having been on Android 2012-2016), but won't leak any further details. I recommend taking the word of a highly credible person who has first …

That is already possible with just a weight axis. The benefit is a lot more flexibility in design and layout, plus just a preference. The Iosevka font also being discussed here today is on the narrow …