Hacker News: Bimos

New comment by Bimos in "Chinese carmaker patents voice-controlled 'in-vehicle toilet'"

Bimos — Fri, 17 Apr 2026 09:11:51 +0000

> The toilet also features a rotating heating element that evaporates urine and dries other waste.

Electrical cars got exhaust gas now, in a worse way.

New comment by Bimos in "Rust Threads on the GPU"

Bimos — Tue, 14 Apr 2026 09:17:08 +0000

> a block of Rust threads that are properly programmed to take advantage of the vector processing by avoiding divergence

But the library is using a warp as a single thread

New comment by Bimos in "Show HN: The King Wen Permutation: [52, 10, 2]"

Bimos — Tue, 24 Mar 2026 02:14:09 +0000

You didn't explain or prove there is a reason for it.

New comment by Bimos in "Xiaomi launches next-gen SU7 with 902 km range and Lidar, still undercuts Tesla"

Bimos — Fri, 20 Mar 2026 02:54:28 +0000

My car is 605km in CLTC and 492km in WLTP. And it runs for about 450km before battery totally die, so I have to charge for every about 400~420km. A bit cold and totally no hill here in Shanghai.

So for 900km CLTC, I think it will go like at least 600~650km easily.

New comment by Bimos in "Exposure Simulator"

Bimos — Thu, 12 Feb 2026 03:56:48 +0000

I like the fact that the photos seem to be real rather than simulated.

New comment by Bimos in "ASCII Clouds"

Bimos — Wed, 14 Jan 2026 05:34:23 +0000

Yeah, with all these colors, any single character should be enough.

New comment by Bimos in "China is eating the world"

Bimos — Fri, 29 Aug 2025 02:54:20 +0000

Look at "Four Asian Tigers", people don't just constantly become rich for granted

New comment by Bimos in "Making Explainable Minesweeper"

Bimos — Thu, 10 Jul 2025 07:39:18 +0000

I thought no guessing mode means the map is generated in such way that guessing is not needed to solve the whole board? (To those who didn't know the mode, it also forces you a starting point)

As an experiment, I tried clicking on a random position after clicking the starting point. And it is possible to that the position is safe. So I don't think the map in no guessing mode is dynamic.

Also in the help section: "In this mode, a starting position is provided, and you never need to guess to complete the board."

New comment by Bimos in "Show HN: Pixel Art Generator Using Genetic Algorithm"

Bimos — Tue, 08 Jul 2025 05:36:25 +0000

It looks like an unconstrained least square problem to me, which is convex and naive. Similar projects like reconstructing an image with transparent triangulars seems more interesting.

New comment by Bimos in "Binary Wordle"

Bimos — Wed, 04 Jun 2025 06:03:04 +0000

Same here.

New comment by Bimos in "Display any CSV file as a searchable, filterable, pretty HTML table"

Bimos — Thu, 22 May 2025 02:11:55 +0000

Yeah but since it claims "any CSV file", and CSV files are widely known to be variate, I didn't expect it fails to work on edge use-cases.

New comment by Bimos in "China strikes back with 125% tariffs on U.S. goods as trade war intensifies"

Bimos — Fri, 11 Apr 2025 09:37:10 +0000

> “Even if the U.S. continues to impose higher tariffs, it will no longer make economic sense and will become a joke in the history of world economy,” the ministry said in a statement, according to a CNBC translation.

New comment by Bimos in "Reciprocal Tariff Calculations"

Bimos — Thu, 03 Apr 2025 07:53:09 +0000

> (x - m) / (εφm)

> ε, was set at 4.

> φ, is 0.25.

What a coincidence!

New comment by Bimos in "63 Chinese Cuisines: The Complete Guide (2024)"

Bimos — Mon, 24 Mar 2025 05:53:45 +0000

I would assume Guilin Noodles are cousin of Vietnam Pho, based on the similarity of their ingredients and their geographic positions.

New comment by Bimos in "DualPipe: Bidirectional pipeline parallelism algorithm"

Bimos — Thu, 27 Feb 2025 02:58:17 +0000

Maybe add Chimera as well?

https://arxiv.org/pdf/2107.06925

New comment by Bimos in "DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling"

Bimos — Thu, 27 Feb 2025 01:59:40 +0000

I heard that it is possible to achieve better performance than cuBLAS using CUTLASS? I thought they chose the better one among cuBLAS and CUTLASS as baseline.

New comment by Bimos in "DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling"

Bimos — Wed, 26 Feb 2025 03:09:13 +0000

I edited it.

New comment by Bimos in "DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling"

Bimos — Wed, 26 Feb 2025 03:02:52 +0000

I think most AI players rely on high performance GEMM. But most people would be satisfied with cutlass or cublas, and the others implement gemm themselves, but not necessarily use undocumented features?

New comment by Bimos in "DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling"

Bimos — Wed, 26 Feb 2025 02:06:57 +0000

> FFMA SASS interleaving

> We observe a performance improvement in the CUTLASS FP8 kernel between NVCC 12.2 and 12.3. By comparing the compiled SASS, we discover that one bit in a series of FADD instructions is flipped in an interleaving pattern. After referencing some open-source CUDA assembler implementations, we identified that this bit controls yield, which may enhance warp-level parallelism (just a guess, yielding the current warp and let other warps work).

> To leverage this, we develop a similar script to modify the FFMA instructions in the compiled binary. Besides simply modifying the yield bit, we also flip the reuse bit (registers cannot be reused if the warp is yielded). This adjustment improves performance (10%+ in some cases) for fine-grained scaling FP8 GEMMs by creating more opportunities to overlap MMA instructions with promotion FFMA instructions.

I would say it is really mind-blowing.

New comment by Bimos in "DeepSeek open source DeepEP – library for MoE training and Inference"

Bimos — Tue, 25 Feb 2025 02:44:33 +0000

The PTX instructions they talked about in the tech report should be pointing to the code here?