16 ms·
Chris Lattner to Lead SiFive Platform Engineering Team
- m0zg 7y agoSo what's gonna happen to MLIR and TensorFlow Swift now? Dead?
- eklavya 7y agoDepends on the community, more and more communities are proving to be more reliable when it comes to core tech pieces.
- DiffProg 7y agoBy far the #1 committer and founding engineer of Swift for Tensorflow also recently left the project: http://rxwei.me/about/ http://rxwei.me/about/ . From the committer list, it looks like Dan Zheng is the only individual who is actively working on it, and he's was an intern until he formally started January 2019. I don't know the ins and outs of Google culture, but it looks at least like every senior engineer has moved onto something else.
- auggierose 7y agoDoesn't mean Swift for Tensorflow is dead. Could mean Swift for Tensorflow will properly support Metal soon :-D
- jph00 7y agoThere are medium sized teams at Google for each of MLIR and s4tf. There's a weekly open design meeting for each of them. Many different folks have presented their work at both meetings.
- rezahussain 7y agoYea but they cancelled this week's meeting, and I guess this is why
- fmap 7y agoWild speculation, but my guess is that both will continue to be developed. As far as I understand, the MLIR project predates its machine learning application and was originally intended as a new IR for clang. In that capacity it makes a lot of sense. MLIR is also currently experimental in Tensorflow, although I have no idea how mature the implementation is. Similarly, there has been significant investment into Swift for Tensorflow, so it's probably here to stay. On the other hand, from a language design perspective Swift is not a particularly good choice for automatic differentiation and translation into Tensorflow graphs (imperative, exposing many details of the underlying machine, etc.). Without a lot of investment into this project it might just be overtaken by a better engineered competitor, or more likely, fail to gain sufficient mind-share over the "good enough" python solution that already exists.
- nochance 7y agoNot clear with MLIR, since it seems to have some traction beyond S4TF. But yes, I think S4TF has had zero interest among people actually doing ML and is probably dead. It might take Google a while to actually kill it off, though. From conversations I’ve had with people doing ML, if they know about S4TF at all they actively dislike certain aspects (for example that it’s based on a statically typed language), or are dismissive of others (like support for differentiation, which is a neat parlor trick but saves a tiny bit of effort).
- m0zg 7y agoNot "zero" IMO. I, for one, was excited about its potential, I was just waiting for it to mature (wisely in retrospect). Python is really bursting at the seams for ML/DL at this point and the ecosystem is in need of a proper compiled language for implementing deep learning systems. This language must be easier on the newcomers than C++. Imagine a future where you could tell if your shit's broken by recompiling it, and your "deployment" would be just putting some RPC in front of your existing code.
- jph00 7y agoMost people I know that understand s4tf well (including me) are excited about it. Chris Lattner and I co-taught two video lessons on it - have a look at those and see what you think: https://course.fast.ai/videos/?lesson=13 https://course.fast.ai/videos/?lesson=13 .
- Joky 7y agoI wouldn't be worried for MLIR. MLIR is getting used internally more and more inside TensorFlow, but also by separate team in different projects, like IREE for example (https://github.com/google/iree https://github.com/google/iree ). The TensorFlow lite converter has been replaced by the new MLIR-based one, similarly for the Edge TPU. But more than that, what makes me confident is the traction we are getting outside Google. First, we landed MLIR in LLVM last month: https://github.com/llvm/llvm-project/commit/0f0d0ed1c78f1a80139a1f2133fad5284691a121 https://github.com/llvm/llvm-project/commit/0f0d0ed1c78f1a80... The LLVM Fortran frontend (f18/flang) which will merge soon in the LLVM monorepo is using MLIR for their own IR. It'll be exiting to develop a non-ML MLIR-based frontend within LLVM! In particular Flang is opening an HPC perspective that could be leveraged by other DSLs later. They are adding an OpenMP dialect to MLIR right now: https://llvm.discourse.group/t/rfc-openmp-dialect-in-mlir/397 https://llvm.discourse.group/t/rfc-openmp-dialect-in-mlir/39... Intel has been actively porting their nGraph/PlaidML framework to be based on MLIR (search for "The Stripe dialect" and "nGraph Dialect" here: https://mlir.llvm.org/talks/ https://mlir.llvm.org/talks/ ). I'm less familiar with S4TF, but I know they recently got some nice new hires, including https://twitter.com/DaveAbrahams/status/1207690883782467584 https://twitter.com/DaveAbrahams/status/1207690883782467584
- m0zg 7y agoEdge TPU will eventually be canceled as well, I'm pretty sure, which is why, as exciting as it is otherwise, I'm not using it for anything practical. There's just no way to make $1B/yr with it, so it's officially below Google's executive interest threshold.
- enos_feedler 7y agoChris Lattner actually gave a talk at the last MLIR open design meeting. This may or may not ease worrying about MLIR's future.
- Joky 7y ago"2020-01-24: Thoughts on Tensor Code Generation in MLIR" here: https://mlir.llvm.org/talks/ https://mlir.llvm.org/talks/ for reference.
- zapnuk 7y agoMy guess is that Swift+Tensorflow on hold. Last I checked, Swift is still years away from 1st class Windows support and adequate tooling. Without those in place, Swift+Tensorflow is still a very niche product for Google and I cannot think of a good enough reason why they should heavily invest into it.
- seanmcdirmid 7y agoWhy would windows matter so much for tensorflow? Isn’t most ML work these days done under Linux?
- pjmlp 7y agoBecause that is what we get on our desks, managed by IT. Python Tensorflow, Julia, ML.NET, Tensorflow for C++, Pytorch, DL4J do all pretty well on Windows.
- jeffshek 7y agoA lot of work from researchers are using *conda on Windows. ML research is happening across all fields, not just computer science.
- eklavya 7y agoI just saw the last commit is from Eugene Burmako, who has previously done awesome work with Scala compiler. I think the project is in good hands.
- chrislattner 7y agohttps://twitter.com/JeffDean/status/1222033368700706816 https://twitter.com/JeffDean/status/1222033368700706816
- arbhassan 7y agoReally loved listening to his recent podcast with Lex Fridman[1]. [1] https://m.youtube.com/watch?v=yCd3CzGSte8 https://m.youtube.com/watch?v=yCd3CzGSte8
- WanderPanda 7y ago"Recent" seems relative
- hinkley 7y agoWhen I read that SiFive is working on custom silicon, my first thought was to wonder what would happen if custom hardware and custom programming languages co-evolved together, instead of languages adapting to old hardware that's adapted to older programming languages. .. and here he is talking about compilers. I might have to keep an eye on SiFive in addition to Oxide. There are some other people talking about him not staying long at places. In this talk he mentions how he intended to stay at UIUC for one year and got 'nerd-sniped' into staying for 5 years building LLVM. After an experience like that, I could see how someone might feel claustrophobic and tend to take any opportunity on offer - if it's interesting enough.
- chrislattner 7y agoYeah, I don't expect people to understand how transformative SiFive is. Give it a couple years and it will make sense :-)
- OrangeMango 7y ago"error bootstrapping app" Quite a worldchanging blog entry.
- dang 7y agoPlease don't post unsubstantive comments here. Also, please don't snark. https://news.ycombinator.com/newsguidelines.html https://news.ycombinator.com/newsguidelines.html
- forgotmyhnacc 7y agoI wonder what's going to happen to tensorflow Swift now, seemed like Chris was the main champion there.
- pjmlp 7y agoIt will most likely fade away. Swift for Tensorflow could never be taken seriously outside Apple community. On Linux, Foundation barely works and one still needs to selectively do either import Darwin or import Glibc for basic IO stuff. Then we are already at Swift 5.1, and Windows version has to be built from source will lots of caveats. How can it even be taken seriously against Julia, Tensorflow for C++, ML.NET all of which work across macOS, Linux and Windows as of today, and offer the same strong typing benefits?
- throwlaplace 7y agosadly i agree with you. initially i was really excited about S4TF because swift is a fantastic language and it would be a fantastic replacement for python as the defacto ML language. but then i realized tensorflow is much worse than pytorch and the S4TF team was too small to build anything substantive enough to win market share.
- flipgimble 7y agoWhile I agree there is an even chance that Google will allow S4TF to fade away after Lattner's departure, that is more of a reflection on the company having no commitment or consistency to good ideas: ex: https://gcemetery.co https://gcemetery.co However S4TF should be taken seriously if you understand what they are trying to accomplish and how deeply they designed machine learning support into the language. Take a look at http://fast.ai http://fast.ai new course offerings using S4TF. Swift has always been a long bet. If it doesn't work as you want it, is still short sighted to discount it in the future. also: https://twitter.com/JokerEph/status/1221831507351748608 https://twitter.com/JokerEph/status/1221831507351748608
- pjmlp 7y agoI will count on it, when it isn't used to sell Google VMs and works on my machine as easy as Julia and ML.NET work today. I wasn't even aware of http://fast.ai's http://fast.ai's existence.
- october_sky 7y agoFYI, more comments over on this post: https://news.ycombinator.com/item?id=22160226 https://news.ycombinator.com/item?id=22160226
- dang 7y agoWe merged them hither.
- DSingularity 7y agoAfter apple Lattner went from to tesla to google to sifive in no time flat.
- GrayTextIsTruth 7y agoThat’s what I was thinking but his LinkedIn shows he was at google for over 2 years.
- CalChris 7y agoHe went from Tesla to Google in no time flat, less than 3 months. But he was at Google from August 2017 until now which is significantly more than one ISO no time flat unit.
- bglusman 7y agoIs there some connection to CarbonFive[0], the consultancy, or did they just.... um... completely duplicate their logo by accident or something? [0]https://www.carbonfive.com/ https://www.carbonfive.com/
- bityard 7y agoDid CarbonFive completely duplicate the HTML5 logo by accident or something? https://www.w3.org/html/logo/ https://www.w3.org/html/logo/
- bglusman 7y agoI don't think that's nearly as close to Carbon5's logo as the SiFive logo is to Carbon5's, but fair point, there's always overlap in these... when I posted that I actually couldn't see the difference between their logos and was genuinely a bit confused if they were connected, but I see a small difference at the bottom of the Pentagon now... so it seems unlikely there's an intentional connection, but was genuinely confused at first!
- jmccarthy 7y agohttps://blog.carbonfive.com/2011/01/19/world-wide-web-now-carbon-five-compliant/ https://blog.carbonfive.com/2011/01/19/world-wide-web-now-ca... :)
- zamadatix 7y agoBoth were probably designed the same way - company name is "word five" so take the first letter and fit it into a 5 sided polygon and bam you've made each logo. SiFive's actually fits this design more even though it came out later. They don't need the added shape at the bottom and the embedded character can be seen as either "s" or "5" while the carbon 5 one needed the extra bit at the bottom and is a bit harder to see the "5".
- icandoit 7y agoThe website looks nice, but I can't find any prices. They won't let you register with a web email (eg gmail) account. Not cool.
- cwyers 7y agoThere's a lot of weird hyphens strewn throughout this post.
- 1MachineElf 7y agoI use lots of hyphens like this too. Is it grammatically incorrect?
- smcl 7y agoI'm not sure whether GP refers to joining phrases up (like "end-to-end" or "idea-to-silicon" or "one-stop-shop") or using it instead of a comma ("I also spearheaded the creation of Swift - a programming language that powers Apple’s ecosystem - and led a team at Tesla that applies a wide range of tech in the autonomous driving space"). The former is just a matter of taste (but IMO has a teensy flavour of "marketing" about it), the latter is totally fine too.
- 52-6F-62 7y agoTo dive down into the pedantic with you all, I think in the latter case the hyphen is incorrect. Normally in that place is an en(–) or em(—) dash. Hyphens are used for joining or breaking words. IIRC
- thewebcount 7y agoThe GP is talking about how the text was written with hyphens inserted where words crossed a line break, but then the text was reflowed so the lines no longer break at the same spot. For example, in the original, it looked something like this: an ex- ample But on GP's (and my) screen it looked like this: an ex-ample
- Khoth 7y agoI think GP is referring to stuff like "pro-vides" and "program-ming" which look like they were hyphenated to line-wrap, but depending on the width of your browser window probably don't actually split across lines
- 7y ago
- leggomylibro 7y agoI can't wait to see more affordable RISC-V microcontrollers on the market! The Kendryte K210 looked very cool, especially with its SIMD-ish "machine learning coprocessor", but it felt like they had rushed the hardware to market without investing in scrutable documentation or software support, last time I checked. The GD32V series looks fantastic, since the current crop of GD32VF103 chips appear to be API-compatible with the venerable STM32F103 workhorse. But I haven't been able to find a source of the raw chips yet, it seems like you can only get them on development boards at the moment. And there are always softcores running on FPGAs, but those sort of highlight how many permutations of "RISC-V" exist. I hope that we don't end up with too many inscrutable compiler flags to juggle as more of these chips become available.
- cculpepper 7y agoNot completely API-compatible. They are /extremely/ similar. Giga Devices did make an ARM STM32 clone, so they probably just did a ctrl-c, ctrl-p on the peripherals. Addresses are slightly different, and things aren't quite the same. I just ran into an issue with the system timer and it's lack of documentation... But good news is that they are pin-for-pin compatible, and you can put a raw chip onto a blue-pill board and use it! You can get raw chips from taobao. I used taobao and a reseller, superbuy to get mine. Not bad at all!
- leggomylibro 7y agoHuh, what do you search for on Taobao to find them? I tried "GD32VF103CB" a week or two ago, but I only got results for the GD32F103 ARM clones. I guess it makes sense that the addresses aren't quite the same; iirc ST has some licensing restrictions on their SVD/header files saying that you can't use them with other vendors' chips anyways. Thanks for the extra information!
- magicalhippo 7y agoI'm guessing you're after the bare MCU, but for others, you can get a nice relatively cheap prototyping board with the chip from Sipeed, for example: https://www.seeedstudio.com/Sipeed-Longan-Nano-RISC-V-GD32VF103CBT6-Development-Board-p-4205.html https://www.seeedstudio.com/Sipeed-Longan-Nano-RISC-V-GD32VF... SeeedStudio has a few others as well: https://www.seeedstudio.com/tag/RISCV-Board.html https://www.seeedstudio.com/tag/RISCV-Board.html Just note that you need a relatively new J-Link (v10 iirc) to program RISC-V cores using J-Link, alternatively for the Sipeed Longan boards, pick up two of them and you can flash one with a provided debugging/uploading firmware.
- bityard 7y agoReminds me of the Silicon Valley TV show sketch where the only goal of every single startup is "to make the world a better place". https://www.youtube.com/watch?v=J-GVd_HLlps https://www.youtube.com/watch?v=J-GVd_HLlps
- kick 7y agoObservational comedy is intended to give the viewer this: without it having a very real basis in reality, the joke wouldn't have been funny (at least, not in the same way).
- azhenley 7y agoSeems like he has been changing companies quite a bit recently. Is this typical for the VP level? Does anyone know how his tenure at Google was viewed by others?
- hn_throwaway_99 7y agoWell, his time at Tesla was basically (publicly, IIRC) determined to be a bad fit between him and the company, so can't really fault him there. So since his long tenure at Apple he had a misstep at Tesla and a shortish, but certainly reasonable, length at Google. Would definitely not call that job hopping.
- pier25 7y ago> determined to be a bad fit between him and the company How so? Edit: Jesus it was a legitimate question, why the downvotes?
- coldnose 7y agohttps://twitter.com/clattner_llvm/status/877341760812232704 https://twitter.com/clattner_llvm/status/877341760812232704
- turdnagel 7y agoThat doesn’t really answer the question. “Not a good fit” is a bit of a euphemism in the industry - it typically is a polite way of saying “I didn’t get along with the CEO/CTO/leadership because X, Y, and Z” - and that’s what GP (and myself) are looking for. I doubt we’ll get an answer anytime soon - he’s a classy guy and probably still under NDA with Tesla.
- 0x8BADF00D 7y agoIt doesn’t really matter - he’s no longer there and I’m sure much happier because of it. There are a lot of irrational things management can come up with. Sometimes the manager in particular is a sociopath or narcissist. Or maybe you unknowingly offended them in some way. At that point there’s no reason to stay, regardless of the reason.
- syntaxing 7y agoI was super tempted to buy the new learning development board they just released (which is actually tough to get in the US) but I actually haven't been able to figure out the benefits of the new SiFive processors compared to other traditional arms boards like a M0. Anyone here familiar with their boards that can provide some insight?
- Erlich_Bachman 7y agoDo you mean besides that fact that it is a RISC-V? You know it is an open source architecture right?
- syntaxing 7y agoYeah, I've been following SiFive for a while now, especially since their announced tool for custom "ASICs" and I know they're a huge driver for RISC-V. I just haven't seen their MCUs used much and curious what advantage they have over other competitors.
- nickik 7y agoTheir value proposition is basically that you can get to market with a costum chip cheaper then with anybody. They argue that their chips are lower energy then use less space. They are more configurable, and the RISC-V tooling is build around this as well. They do seem to quite a few costumers but their current growth is VC funded. There is nice stuff coming down the pipe as well, the RISC-V Vector extension and hopefully finally some linux boards. RISC-V SoC with FPGA is going to be a product. But hardware product cycles are just long and a takes a while.
- creato 7y agoWhy should a random tinkerer trying to learn or get a basic job done with a microcontroller/CPU care whether the architecture is open source? It's still overwhelmingly difficult to do anything interesting with the fact that the architecture is open source.
- 7y ago
- dang 7y agoThe related press release, which doesn't say much either, is https://www.businesswire.com/news/home/20200127005141/en/Google-Tesla-Engineer-Chris-Lattner-Lead-SiFive https://www.businesswire.com/news/home/20200127005141/en/Goo... (via https://news.ycombinator.com/item?id=22160226 https://news.ycombinator.com/item?id=22160226)
- baybal2 7y agoI have hard time connecting "machine learning" with what RISC-V is about. Above all, the original plan for RISC-V was to make a barebone MCU ISA first, and everything else second. This was largely to ARM being very militant with terms on RTL access for M* series cores for commercial use. If you throw enough extension, and workarounds even on top of 8051, you should be able to make a CPU grade core with it. But you being able to do it, doesn't mean you should.
- brucehoult 7y agoRISC-V was originally developed because some vector-processor / ML people at Berkeley needed an extensible control processor for their specialized hardware. They'd previously been using ancient 32 bit MIPS but they needed 64 bit, a good amount of spare opcode space for custom instructions, and reasonable licensing and nothing suitable existed so they rolled their own. RISC-V with the almost-done Vector extension is likely to be a big force in ML hardware.
- pjmlp 7y agoML hardware has more to gain from FPGAs.
- brucehoult 7y agoBased on what? Whatever number of ALUs / DSP slices you can put in an FPGA and soft-wire together, you can put just as many hard-wired in a custom SoC with lower area and cost, and faster performance. An FPGA is good for prototyping this until you figure out the best arrangement, sure, but three months later you can have real chips.
- pranith 7y agoLooking forward to seeing a high performance dev-board from SiFive.
- mark_l_watson 7y agoI wonder if this will curtail the effort to implement TensorFlow in Swift, turtles all the way down? That would be a shame. Python ecosystem with TensorFlow, PyTorch, mxnet, etc. has been good for rapid progress but I think we need something better to break out of just using deep learning. This needs a hackable infrastructure. I personally don't have the skill to hack the C++ TensorFlow core. I think a new ecosystem based on Swift, TensorFlow, and future tools and platforms makes some good sense. An alternative would be a similar hackable infrastructure based around the Julia language, which is also very good.
- byt143 7y agoEven in swift it isn't "turtles all the way down". The AD stuff is hardcored into the C++ guts of the compiler, whereas Julia's source to source autodiff accesses a compiler pass from a fully Julia user package. Aside from making it easier to hack and improve the AD system as just a Julia user, this capability enables other package program transforms like that in https://github.com/MikeInnes/Poirot.jl https://github.com/MikeInnes/Poirot.jl for prob programming. So Julia is already further ahead in that regard and it's more hackable.
- mark_l_watson 7y agore: "So Julia is already further ahead in that regard and it's more hackable." I agree. Flux is very concise, very nice to work with. I just had some trouble with my small playing-around code snippets when going from one minor release to the next, but that probably means I should revert to the LTS 1.* version. I have tried Julia with non-mathematical stuff like using it with sqlite, fetching and using RDF data, and general text processing - nice for those use cases also.
- lsllc 7y agoThis totally negates @bradfitz leaving Google/Go!
- karnajitw 7y agohttps://www.sifive.com/blog/with-sifive-we-can-change-the-world https://www.sifive.com/blog/with-sifive-we-can-change-the-wo...