<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Hacker News: adityazero</title><link>https://news.ycombinator.com/user?id=adityazero</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Fri, 07 Aug 2026 05:59:13 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=adityazero" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[New comment by adityazero in "Changes at Google DeepMind: Demis Hassabis from CEO to Chair, Jeff Dean departs"]]></title><description><![CDATA[
<p>All of these would happily come back if Sundar left.</p>
]]></description><pubDate>Thu, 06 Aug 2026 00:37:50 +0000</pubDate><link>https://news.ycombinator.com/item?id=49190939</link><dc:creator>adityazero</dc:creator><comments>https://news.ycombinator.com/item?id=49190939</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49190939</guid></item><item><title><![CDATA[New comment by adityazero in "Changes at Google DeepMind: Demis Hassabis from CEO to Chair, Jeff Dean departs"]]></title><description><![CDATA[
<p>> sundar, softly: “we can create a permission working group.”<p>so true. i have attended meetings to decide on meeting topics for the next half.</p>
]]></description><pubDate>Thu, 06 Aug 2026 00:37:17 +0000</pubDate><link>https://news.ycombinator.com/item?id=49190935</link><dc:creator>adityazero</dc:creator><comments>https://news.ycombinator.com/item?id=49190935</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49190935</guid></item><item><title><![CDATA[New comment by adityazero in "[dead]"]]></title><description><![CDATA[
<p>NVIDIA runs the other way. The instruction set is documented, but the `ptxas` compiler is not. And NVIDIA's documentation stops higher up than people assume: PTX is a virtual ISA, but SASS is published simply as a list of opcode names.<p>So I went through the open compiler sources, OpenXLA and the Mosaic dialect inside JAX. They name the parts of the machine they generates code for, and there is useful information in there for anyone writing TPU kernels.<p>`tpu.log` is `printf`. A tag string plus whatever values you hand it. Next to it sit log_buffer, and trace_start/trace_stop for timing regions. I feel kernel debugging on TPU has a worse reputation than it deserves.<p>`tpu.weird` takes an `f32` and returns a `bool`. JAX's lowering evaluates `<a href="http://lax.is_finite" rel="nofollow">http://lax.is_finite</a>` is its negation. So the TPU has a hardware predicate for "this float is weird."<p>And the MXU is a FIFO. Push the weights, stream the activations, pop the result. Weight-stationary, more than one per core, 32-bit accumulator enforced by the verifier. "bf16 in, fp32 accumulate" is the only shape the unit offers.<p>90 operations, eight memory spaces, and a lot of guidance to developers.</p>
]]></description><pubDate>Tue, 28 Jul 2026 22:55:15 +0000</pubDate><link>https://news.ycombinator.com/item?id=49091097</link><dc:creator>adityazero</dc:creator><comments>https://news.ycombinator.com/item?id=49091097</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49091097</guid></item></channel></rss>