jlebar

35 ms·

331.
by jlebar 9y ago
> Hi, I'm the guy who wrote the original VCBlog post. Hi from one compiler engineer to another! > Chandler is the guy driving Spectre in LLVM. I actually sit about 25ft away from him; we talked about his tweet before lunch today.
332.
by jlebar 9y ago
Yeah, even this footnote really oversells it imo. I have trouble imagining a person who would want the level of (non) protection delivered by this flag. Kind of reminds me of https://digitizor.com/internet-explorer-9-caught
333.
by jlebar 9y ago
> But for some reason everyone is comparing a cluster of 4 TPUv2 chips to a single V100 chip. I agree that some people are doing that. Marketing, I suppose. But that comparison is explicitly not the point of my parent post. I'm
334.
by jlebar 9y ago
V100 has 900gb/s memory bandwidth [0]. TPUv2 has 600gb/s per chip x 4 chips, so 2400gb/s [1]. As we've discussed elsewhere [2], comparing TPUv2 to V100 on a per chip basis doesn't make much sense. Who cares how
335.
by jlebar 9y ago
This is in comparison to 900gb/s for V100.
336.
by jlebar 9y ago
Has Google published data on the memory bandwidth of TPU v2 (aka "cloud TPU")? I'm having trouble finding it. In any case I agree, we shouldn't be looking at the stated peak compute of either of the chips. (Disclaimer:
337.
by jlebar 9y ago
> A "single TPU" is 4 ASICs. It is not clear if it makes sense to compare a "single TPU" to a "single GPU." Why does the number of chips matter? Put another way, suppose Google tomorrow announced Cloud TPU v
338.
by jlebar 9y ago
I know people don't know what to expect from tpu performance, but does anyone actually get 100tops out of Volta? I thought you'd have to spin the tensorcores and never touch memory, which is...not realistic. I know you hedged by
339.
by jlebar 9y ago
Most just come with some "creative" story about how they are persecuted in their original country [...] while they probably killed or have contributed to the death of many Are you really saying that a majority, or even a large f
340.
by jlebar 9y ago
I know what you mean, but e.g. every time the Google Assistant talks to you, the consumer, you're using TPUs. https://deepmind.com/blog/wavenet-launches-google-assistant/
341.
by jlebar 9y ago
Likewise. Frankly I didn't enjoy The Left Hand of Darkness when I read it, but I think about The Dispossessed all the time.
342.
by jlebar 9y ago
Correct, one of the people listed as teaching this course.
343.
by jlebar 9y ago
If any Stanford students are reading this wondering if they should take this class: Engler was one of my favorite professors at Stanford. He's kind, sharp, well-prepared, and consistently delivered great lectures. One of the things th
344.
Retpoline: a software construct for preventing branch-target-injection(support.google.com)
8 points by jlebar 9y ago|0 comments
345.
by jlebar 9y ago
Without speaking to the capabilities of TPUs, note that most ML models today are mostly convolutions and matrix multiplies.
346.
by jlebar 9y ago
> BSDs tend to have excellent documentation as do some individual projects (git) Should we really point to git as an example of great documentation? I mean, this exists and has to have a giant banner at the top saying "these are NO
347.
by jlebar 9y ago
If we were to slow down the progress of self driving cars, would that not also be a bet on human drivers?
348.
by jlebar 9y ago
> There is evidence that psychoactive compounds have been used since at least 3000BCE in Mesoamerica. Looking into it it seems lead was first mined around the same time in Greece but the lead sweetener idea wasn't popular until the
349.
by jlebar 9y ago
> Humans used psychoactive substances for centuries, usually under the supervision of a shaman or some type of religious figure to undergo a journey. I don't have a position on legalizing psychoactive mushrooms, but I don't thi
350.
by jlebar 9y ago
I haven't been privy to any work with the ISPC folks. It's totally possible they're using the LLVM NVPTX backend, I dunno.
351.
by jlebar 9y ago
nvcc installed on a Mac seems tied to the current clang and the latest clang's don't support CUDA development so I have to retrograde my clang to an older version to use CUDA. Why is nvcc tied to clang? To be clear, there are tw
352.
by jlebar 9y ago
Running code on the GPU usually isn't as easy as "compile and go", but...yeah, if you can emit LLVM IR, you can get PTX which you can run on the GPU.
353.
by jlebar 9y ago
We'd like this too. Unfortunately a lot of the special cases can't be eliminated because we have to interface with ptxas, the closed-source PTX -> SASS (GPU machine code) optimizing assembler.
354.
by jlebar 9y ago
> This is in part because of the work by Google on the NVPTX LLVM back-end. I'm one of the maintainers at Google of the LLVM NVPTX backend. Happy to answer questions about it. As background, Nvidia's CUDA ("CUDA C++?&quot
355.
by jlebar 9y ago
Imagine you had the money today (you'd already, automatically sold all the stock). How much would you buy? Once you've figured that out sell everything except the amount you would buy. This is the correct answer. When your pu
356.
by jlebar 9y ago
I think what OP meant is, if you have a page that's backed by disk, then when you're low on RAM, the OS can simply drop that page -- it's sitting there on disk if you ever need it back. But if you have a page that's not
357.
by jlebar 9y ago
Surprised nobody has mentioned Heinlein, "The Moon is a Harsh Mistress". I would say it's harder sci-fi than "I, Robot" (which I also love).
358.
by jlebar 9y ago
Since abseil doesn't promise ABI compatibility over time, I think the notion is you don't ship libabseil.so.0 as part of a distro. (I'm a Googler, but this is not my project.)
359.
by jlebar 9y ago
Critically missing is Webshit Weekly: http://n-gate.com/
360.
by jlebar 9y ago
I think your post actually proves the point of the person you were disagreeing with. If you, someone experienced with "real" camera gear, were "astonished" by the iphone 7+ pictures and thought they were taken by &quo

More ›