Posts by _olli 56 publicly visible posts • joined 28 Jul 2020
I remember C compilers of early 1990's, they mostly produced inefficient code, couldn't optimize register usage, used stack variables inefficiently. Still they were used to compile programs.
95% of humans can't create a C compiler, even if they were given a year of time on it.
So yes, I am impressed of anyone who can build a working C compiler. And I don't expect his/her first working compiler version be a fully optimized one yet.
"Enterprise software suite" is a synonym for "expensive price tag and well-compensated sales team". It is not a synonym for high software quality and enhanced security, an expensive enterprise software suite can still be essentially some base software coming with a bag of hacky scripts.
Hmm, so Intel wastes substantial silicon area for AVX512, the AVX512 unit surface area being roughly half the size of a full E-core, and years afterwards discovers that users of it's most expensive flagship datacenter Xeon CPUs hardly ever need AVX512 for anything.
Was this again the genius corporation that fares magnificently, or was this the one that does really badly?
I have worked among corporate software solutions that were deployed to AWS, GCP and/or IBM clouds and am puzzled what's so scary about competing against AWS, GCP, IBM : These vendors provide scalable solutions that are technically doable but by any measure not technically superior, based much on open-source solutions, and their prices are not particularly cheap.
Much of the stuff that companies deploy into these three-letter-vendor clouds are either virtual machines or various services running inside containers plus databases plus load balancers, and all that's quite standard technology.
I don't really see a reason why European cloud vendors that can provide scalable cloud services based on open-source components with better prices that AWS couldn't win customers, in particular no the old continent.
Azure is a slightly different basket if the solution architecture is locked into Microsoft solution, but that can be avoided by not designing solutions that are locked into Microsoft solutions.
I recently installed native Linux onto my work laptop, out of frustration of how Windows has gradually grown ever more confusing and bloated environment. My actual daily work and toolchains are on top of Linux anyway, yet the default offering of the fine IT department's is to provide Windows laptops so that users can run Linux in a virtual machine for that part of the work.
The IT department doesn't promote nor support Linux, yet they seem not to disallow (or fire) people from self-installing Linux into the corporate stinkpads either, so a handful of rogues have gone that path.
Haven't regretted a minute for leaving the Windows layer out from the stack, everything is just so much cleaner and work faster that way, instead of the earlier linux-on-VM-on-Windows. The Microsoft office, teams & outlook tools nowadays work okayish (not perfect, but okay) in web browsers also, so us native linux users can still use the fine Microsoft suite whenever that is necessary.
Just the other day I was inspecting how to alleviate a performance issue in a custom software and tested using ChatGPT to help in the tesk. The prompt was (roughly):
"Give c+++ code for mutex-free thread-safe multi-producer message queue for multithread communication".
And voila, few seconds later ChatGPT gave C++ source code for a message queue. On the surface the solution looked somewhat decent, however, closer look showed that the code was not thread-safe, in other words it was useless for the purpose. Aside I set the AI and went browsing boost library modules.
AI shows promise for doing simple stuff, which is good news for people working on simple stuff, however, it's currently questionable much it can boost productivity when working on hard stuff.
From what I recall from the Windows 95 or even the Windows 3.1 era, it wasn't necessary to physically load software from floppies any more each morning: it was necessary to reinstall the operating system from floppies only once per month, with some luck sometimes even more rarely. For daily operation there were those humming and ticking scifi boxes inside the computers back then, they were called as lard disk or tard disk or something like that.
The factories wouldn't even need to be "destroyed", it's sufficient to remotely deactivate the EUV machines should China capture these fine factories. Turning off the TMSC factories would however cause massive unavailability shock for the high-end silicon chips, and a financial shock of unprecedented scale given that 8 of the 10 largest companies in the world base their businesses on high-end silicon. Which might fairly entertain China: there collapse Taiwan + US stock markets in one blow.
Sound generally like a fair advice.
However, in case your team should develop modern Web applications, creating even a rather simple React app pulls literally thousands of 3rd party open-source packages as dependencies, sub-dependencies and sub-sub-sub dependencies into your codebase from the npm (etc) online repositories.
Good luck reviewing and knowing what all those thousand+ packages contain. From security perspective that development model is badly screwed.
First IBM closed down their Watson IoT cloud platform, then Google retired their Cloud IoT, then Microsoft. Let's see if and when AWS IoT will follow.
Seems that big houses had little to offer to IoT data integration in the end.
Ramping down of IoT gateway services is a nasty surprise to companies who already have built their industrial field systems on these IoT technologies, with expectation of 10+ year service life. I know couple of such designs.
Lesson to learn: never lock your cloud systems design to specific vendor's technology, instead use open-source solutions and interfaces whose hosting you can move to another cloud vendor if the current one should go sour.
Intel portfolio's single-thread performance saw 2,2-fold increase from i7-4790K (released 2014Q2) to i9-13900KS (2023Q1) during the past decade (or 9 years ~ a novade?). Admittingly, multi-core performance improvement of these same CPUs were 5-fold i.e. more than double that of the single-core performance improvement.
For the time being the trend seems to hold if not even accelerate, Google's Cloud TPU already reaches 10 exaflops in training of large language models, although it's not enlisted onto the official TOP500 list, and while that score may not be directly comparable to linpack it's yet upon doing something useful. And if working on AI, a 400-teraflop cloud tensorflow accelerator is available at roughly $1.2 per hour spot price, so it's 0,3¢ per teraflop-hour. Insane times.
I would gladly take a macbook any day if I just could install my OS of choice into it. Which I can't, thus am locked out of the Apple ecosystem.
I will also gladly take this new qualcomm wonder if it's really twice as fast as i7, which I suspect it won't be. I by the way measure the performance with AC power connected, because I have never needed to run any seriously heavy workload while being outside the power grid. Sure I have cleaned up notes and polished PowerPoint slides on airplane, but any laptop can sustain that kind of light work much much longer than decently necessary.
I don't even care about Windows support, I would gladly purchase an ARM-based laptop that could match concurrent (say Zen4) x86 price & performance and allow running Linux natively, just for sake of being geek enough to want to tinker with one.
However, having so far not seen one to emerge to market makes one a bit skeptical if that shall appear now. I am not holding my breath.
The best ARM PC-alikeish offering as of currently is Apple M, yet even that is more expensive and slower than Zen4 x86 and does not really well support native Linux installation. Then there's been a bagful of Chromebook-and-similar laptop-like products based mostly on Qualcomm chipsets, yet those have either been too slow or too expensive to be an appealing alternative for x86.
There aren't RISC-V soc chips available in volumes that would equal to A76 in performance yet. Also Linux distribution support for RISC-V SBC computers is work in progress, there're development releases but not yet stable, matured releases.
RISC-V will get there but it's not there quite yet. Rpi's decision to stay in Arm architecture for the pi5 is very understandable.
Very very small fraction of people that develop AI actually develop CUDA code directly, instead they use libraries such as TensorFlow in between their frontend and the GPU
So how about getting that fine TensorFlow to EASILY work on those good intel GPU chips, and same to you AMD?
I have a suite of Amd, Intel and Nvidia GPUs in household and would be agnostic of which of them to use, yet Nvidia is the only brand that is easy to get working with common open-source AI frameworks. Amd has this rocm layer yet good luck getting it working unless using a very spesific OS distribution and/or kernel and/or GPU version.
Something that can let a script running in one browser tab see the password entered into a bank's website in another tab would definitely not be of medium but of screaming-high severity.
The description of the issue suggests that leak appears in the upper bits of vector registers when invoking special cpu instruction, which sounds like something that can not be trivially exploited with javascript code. Then again, webassembly might turn out to be an unintended lockpick for hackers.
An old acquitance working in logistics industry shared an insight that he's never seen a business case where replacing an existing ERP with another ERP system had demonstrated a positive return of investment, because typically whatever existing ERP system just works, ERP transition programs are expensive and new ERP always causes all sorts of unexpected hassle across the supply chain.
Transition may of course be inevitable if an ERP system is tens of years old and badly outdated, but swapping from SAP to Oracle (or vice versa) sounds bit like swapping from cholera to plague.
Yes. Also memory-overwrite bug being considered as the most dangerous type of bug does not mean it were the most common and/or high-scoring type of bug.
Majority of the high-CVE bugs on these known-bug lists are nowadays not related to low-level memory access but rather to patchy web services that leave networked systems vulnerable to remote attacks. As such services are typically already written in somewhat memory-safe high-level language such as java, javascript, python etc, it wouldn't really make a difference even if these patchy web services had been originally written in rust.
Using rust in low-level stack is good idea though.
I think it well can be up to something. Just other day followed a teenager having a discussion with the public ChatGPT chatbot about horsecare, and the bot appeared well-mannered and more knowledgeable about the topic than most of people, including myself. The teenager got upset when I eventually just closed the browser without wishing it proper good-night first.
I am generally impressed with anything that can impress teenagers.
Oh, and that public ChatGPT bot is still based on the previous-gen GPT-3, not the GPT-4 that is rumoured to be much more advanced and have got people sign these petitions.
More often things stored into DBs event aren't inherently really that complex. There's just this school of system architects who think that "complexity is good" and oh boy can they coat everything with additional layers of complexity. Presto, mess is ready.
Those global networks have by the way so many much more useful uses besides recreational uses such as streaming movies and showing endless video advertisements o' Netflix & Google.
Finetuning your metafora, current situation is more like Ford demanding that "Esso" should provide unlimited free gasoline to it's cars, because Ford already built factories and ships cars over ocean (or whatever they'd claim).
Good news is that large monopolistic conglomerates eventually fail. Always. See the Dow Jones stock index of large US corporations as an example: There aren't a single of the original big companies left in the index: Every single one of the original DJ corporations have lost their innovation, became surpassed by more innovative companies, and shrunk in their influence. The same will also happen in the future.
Accordingly there won't be "late stage capitalism where competition does not exist", unless of course open market capitalism is replaced with some inferior closed market system such as communism.
Intel sells embedded x86 processors and FPGAs to mobile network manufacturers. Oh, and they also tried to sell their 10nm chip fab services to one major 5G mobile network manufacturer just few years ago, but because of their infamous 10nm trouble they couldn't really deliver working chip, which almost sunk that poor customer.
I use an old MS Office version and it works fine for me. It's at least 10 years ago, legally acquired and does everything+more that I have ever needed to do with Office.
For basic user the Office kit was perfected already 15 years ago. Thus I don't see what value I'd get by paying $80/year for Office 365 subscription.
Everyone, literally everyone in IT industry should climb up from their pits of comfort and change job at least once in a decade, or else they'll stagnate their personal development and eventually may end up becoming unemployable due to obsoleted set of essential skills. These news of big techs letting tens of thousands go may sound terrible, but there's currently high demand for skilled professionals and in the end a change will do good for all the involved parties.
Sure I would love to get a new 24- or 32-core CPU, but I just still struggle to utilize even half of what my current 8-core, 16-thread zen2 delivers. It's possible that the CPU load may have hit 50% CPU workload once or twice, but even then only for a brief while.
Windows admittingly eases finding some utilization for your precious cores as it's OS so fond of running all kind of unnecessary junk services in the background.
There are no cultural or economical barriers to entry for software development either.
The most essential barrier to entry for software development is between ears: It requires certain level of curiosity and mental capacity to get started & comprehend abstract logic. AI won't lower that threshold.
Adobe paying $5000 per every current user is indeed a bubble price for a tool that is ok for drawing vector images and UI mockups, but c'mon, those are still just vector images and UI mockups.
Even I am one of those 4M figma users but I don't see how they possibly could extract even a fraction of that $5k from users like me just to break even: hike the price, I'll move elsewhere.
UI designs age fast nowadays and thus UI mockups are sort of throw-away stuff anyway after the actual product has been implemented, so no-ones gonna lock in to the same vendor eternally just because their original year-2022 UI mockups happened to be there, given that they are not the only doable design tool around.
I have a colleague in his mid-40's whose retirement plan is to become a Java-grandpa like these COBOL-grandpas, so that his retirement days in 2040's will every once in a while get interrupted by invitations to work against a three-figure hourly wage when some big corporation needs to change anything in their enterprise Java software systems that these modern javascript kiddies can't just comprehend.
Given that North Korea is one of the most isolated and backward countries, where free thought is verboten, access to Internet very limited, access to university education very limited and curriculums are all about studying juche ideology, this picture that paints such country as source of masses of skilled IT workers is just not credible.
The false assumption behind "low code" is that the hard thing in software development would be writing text, and thus low-code tools attempt to resolve difficulty of software development by taking away that writing-of-text part.
Which of course is not the difficult part at all. Source code lexicons are almost a trivial part of the challenge, so that a sharp 11 year old with textbook or tutorial site can teach him/herself the rules required for writing syntactically correct source code.
Intel is a prime example of how a CEO can indeed contribute remarkably to development of the stock price. Couple of previous CEOs were so extraordinary in their execution of the strategy that the company value still hasn't stopped sliding.
Indeed this announcement shall raise eyebrows in corporations that have jumped into IBM Watson bandwagon with intention to use their IoT cloud platform to host industrial products with intended life cycle counted in decades.
What makes things worse is the IBM Watson IoT bluemix way of integrating services to IBM database solutions and concealing the underlaying server and container system from the user, so it won't be quite trivial to transfer the current services to run elsewhere without remarkable refactoring.
Lesson learned: Never lock your cloud software solutions to single vendor so that you couldn't easily relocate them to another cloud vendor. Same applies to AWS, Google.
That is fascinating: One sitting on his sofa and thinking of two heavenly rocks circulating each other would not very easily get into conclusion that tidal waves can conduct energy in such efficient way that it makes massive rocks gain more gravitational potential energy and gradually make them drift farther and farther away of each other. Vice versa, one might intuitively suppose the opposite, that potential energy get wasted in friction of tidal waves and the moon would gradually fall closer to the earth. But evidence reveals the surprising direction of this process.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
