HN user
Real-time NVIDIA GPU scheduling researcher.
https://jbakita.me
Location: Chapel Hill, NC, USA
Remote: In-person preferred, open to remote.
Willing to relocate: Yes.
Technologies: Efficient Real-Time Systems on GPUs
Résumé/CV: https://jbakita.me/CV.pdf
Email: (see CV)
I'm currently finishing my PhD in CS at UNC-CH, and interested in academic or industry research positions starting after May 2025.
Yes. The kernel has access to data from every workload, and so technically a bug in _anything_ running at kernel level could result in data leakage.
No official word, but the communication antenna on the tower appears to be damaged [1]. Seems likely that played a role.
This law does not prevent Americans from reading, viewing, or saying anything. You will still be free to visit TikTok's web client and do that. This prevents US companies from doing business with the company (eg. ads, app distribution, etc).
Broadcast news and radio are limited to at most 20% foreign ownership by default [1]. Applying a similar requirement to large internet news distributors seems reasonable if they want to do business in the US (even if "banned" they could still distribute content, they'd just be restricted in making money).
[1] https://www.foster.com/newsroom-publications-The-Road-Map-Fo...
Although there is a type of peer review that includes redoing experiments and analysis: artifact evaluation. All of the top conferences in my field (real-time embedded systems) include this as an opt-in option, and papers get a special badge if they also pass artifact evaluation. I strongly believe that other fields in computer science would benefit by including and normalizing this process.
Besides the reproducibility benefits, artifact evaluation forces documentation of the experiments and process; I've found this enormously useful when on-boarding new students to an existing project.
Very nice overview of the Voyager program.
I love the words from the President included on the spacecraft: “We hope someday, having solved the problems we face, to join a community of galactic civilizations. This record represents our hope and our determination, and our good will in a vast and awesome universe.”
It's like a bonus level!
Last commit was from march... why was this posted now and is there any interest/activity left?
I work in an adjacent research group at UNC, and I can assure you that this is a very active project. Unfortunately, because most venues now use double-blind review, the updated code can't be posted until after the associated paper(s) are accepted.
I'd encourage any potentially interested parties to star/watch the GitHub repo to keep an eye on development. I've seen some very impressive benchmark improvements from work currently in the pipeline.
As of fairly recently, the average US household had 1.9 vehicles. [1]
[1] https://www.bts.gov/archive/publications/highlights_of_the_2...
Per [1], the comment appears to be partially correct. The data prior to 1950 does not agree with [1], but hurricanes were not tracked in the same way prior to 1950. The author may just using a different source for those estimates.
[1] https://en.m.wikipedia.org/wiki/List_of_United_States_hurric...
From what I understand, they just lengthened the submarine to increase its displacement enough to counter the extra weight.
It also looks like the hull of the first submarine is now complete, [1] so there's some possibility they'll meet their new timeline.
[1] https://www.globalsecurity.org/military/world/europe/s-80.ht...
Tangentially related, does anyone know if there is a DisplayPort-to-USB-C/Thunderbolt cable one could use to connect a GPU (DisplayPort) to a monitor (USB-C/Thunderbolt)?
Edit: See other comment. Looks like someone actually did manage to create a cheap adapter, and was able to power it from DisplayPort as well. I'd love to hear if anyone knows how well this works.
I think this would be quite hard to do inexpensively, as USB-C hosts have to be configured by the client to send Display Port signals on the SuperSpeed lanes. [1] If the monitor (client) doesn't detect a host, it's probably not going to try to interpret the SuperSpeed signals as DP. It seems to me that an adapter would likely have to emulate a USB-C host supporting DP alt mode. This would at least require external power (unless your monitor's USB-C port is powered), and would more likely require the adapter to use one of your USB 3 ports so that it can also support devices on any monitor USB hubs.
[1] https://www.totalphase.com/blog/2019/11/how-displayport-alt-...
(Disclaimer: I'm a PhD student in real-time operating systems at UNC.)
Each program has its own strengths, and simply comparing between similarly titled classes is an unfortunately poor metric. (Eg. At UNC, OS is really an introduction to C and OS concepts class. The Operating System Implementations class sounds more similar to CMU's OS Class.) Academic programs need to be compared holistically, and only if one is consistently worse than the other do you have an answer. (Otherwise I could go and claim that UNC is better than CMU because our real-time group is much better.)
Of course, that's too much analysis for most people (including me), so I would argue that looking at where graduates work is typically an okay proxy for undergraduate educational quality.
A large part of that competence comes from their increased manpower. It's almost funny how many more developers they have. I heard from an ex-Google-intern that Google has over 1,000 people working on Chrome. Compare that with EdgeHTML's core dev team, which was well under 100 people (really closer to 50 if my memory serves me right).
Oh, I could certainly see that. I don't recall encryption/signing working very well in the best of cases. If you haven't yet, I'd suggest submitting a bug report (I think there's a button somewhere in the O365 web interface). ActiveX is very deprecated and likely will be removed from Edge as part of the move to Chromium (just my personal speculation - I don't know if that's the official plan).
Eh, half the problem is that there are far too many options to maintain, many of them legacy. Smart card authentication has worked great in Edge on Office 365 for years without any plugins at Microsoft, so I imagine your employer has got the wrong box checked somewhere triggering a fall back to ActiveX.
Finally, reboots are necessary on all major operating systems, even linux...
Sure, but there's the question of regularity. For the most part, I can do live updates of every part of my linux system (including the kernel if livepatch is enabled). I only have to reboot if there's something like a major change to the init system. This leaves my server with the latest bits and no downtime in the past 5 months. On the other hand, my Windows machine tries to reboot at least monthly - and that's less often than I'm sure Microsoft would like.
"For example, they may start integrating technologies for which they have exclusive, or at least 'special' access. Can you imagine if all of a sudden Google apps start performing better than anyone else's?"
This is already happening. I very recently worked on the Edge team, and one of the reasons we decided to end EdgeHTML was because Google kept making changes to its sites that broke other browsers, and we couldn't keep up. For example, they recently added a hidden empty div over YouTube videos that causes our hardware acceleration fast-path to bail (should now be fixed in Win10 Oct update). Prior to that, our fairly state-of-the-art video acceleration put us well ahead of Chrome on video playback time on battery, but almost the instant they broke things on YouTube, they started advertising Chrome's dominance over Edge on video-watching battery life. What makes it so sad, is that their claimed dominance was not due to ingenious optimization work by Chrome, but due to a failure of YouTube. On the whole, they only made the web slower.
Now while I'm not sure I'm convinced that YouTube was changed intentionally to slow Edge, many of my co-workers are quite convinced - and they're the ones who looked into it personally. To add to this all, when we asked, YouTube turned down our request to remove the hidden empty div and did not elaborate further.
And this is only one case.
Note that this does not effect all MIPS-based systems. The standard has included hardware floating point for many years now, it just seems some OEMs have neglected to upgrade or cut it for cost savings.
They revived the app sometime last year and it's back in a usable state now.
Some of the "others," includes pilots, FAA managers, and ex-crash investigators. The article is worth paying for.
I was just poking through some of the code and noticed that there seem to a lot of open bugfix PRs [1]. Have you thought about going through and upstreaming some of them? Many of them seem like quick reviews.
Edit: Also, if you would like some help maintaining it, I imagine plenty of fellow HN-ers (including me) would love to help.
[1] https://github.com/fogleman/primitive/pulls?q=is%3Aopen+is%3...
I spent this summer working on a Windows team at Microsoft, and this article is incredibly insightful for an outside perspective. The dev process internally is basically just waterfall sped up 3x. While my team only permitted code commits after adding to and running the tests, that did not seem to be the norm. Almost weekly one of our sister teams would not only break a feature (that happened pretty much every day) but break the build. Things got so bad that as we approached this most recent release, management (under pressure to reduce the number of failing tests) just disabled the majority of our failing tests.
While it might slow down some of the feature work, I think it's clear that an always release-ready master branch is the path forward for Windows. Some of my more enterprising managers were already working on splitting our subfeature out of the Windows release tree just so that we could do that when I left a couple months ago. I hope they succeed, otherwise I worry for the future.
AMD has a CUDA Transpiler (https://github.com/ROCm-Developer-Tools/HIP) and a multitude of other open source libraries through GPUOpen ( https://gpuopen.com/professional-compute/).
The quality and open source nature of their tools has resulted in much of my research group (real-time vision) increasingly and voluntarily moving to work on AMD platforms (we were previously almost exclusively using CUDA).
According to the linked article, Threadripper 2 uses the equivalent of 4 Ryzen 7s, however they disable the memory busses on two of them which yields only a 2x increase in memory bandwidth. You're correct that Threadripper tends to clock slightly lower, but the cache is on-die so it shouldn't be any different per-core than in a typical Ryzen.
Hey, I work on the web platform team at Microsoft. Do you have any more information about this? I'm not sure this is a bug that we're aware of. If you've personally experienced the bug, could you pull the Windows build number (Settings->System->About) and send it to t-jobak (at) microsoft.com along with the steps to reproduce (if any)? We'd really appreciate it and might be able to get a fix out in the next release.
Yes. Threadripper is essentially just four well-binned Ryzen 7s tied together with a creative interconnect. The memory bandwidth doesn't increase as proportionally as the core count, but each individual core should operate just as fast as in a typical desktop part.
Edit: To answer the second part of the parent question, since the dies are the same, it should have all the same power-saving features available in AMD's laptop and desktop processors.
Just wanted to say that I met Sanketh ^ along with some of the rest of the Smartcar team at a HackDuke last fall and was really impressed. The team was enthusiastic and their pre-production API beat so many of the production APIs that I've encountered in the past. Definitely check them out if their work sounds interesting!
If the datacenters are really too complex for the people running them to understand, I would expect failures like this to drive the airlines towards AWS, Azure, or some similar service.