• No, Virginia, shader prefetching is not broken on AMD's new Radeon 7000 graphics cards.  (Tom's Hardware)
  
  This rumour was circulating yesterday, based on a comment in the code for AMD's open source graphics drivers.  It would be odd if prefetch didn't work on these new RDNA3 cards, because it worked in RDNA2 and indeed in the original RDNA.
  
  AMD has explained what that comment means: The feature that doesn't work and is disabled in the drivers is an experimental new prefetch mode planned for RDNA4.
  
  As the article points out, AMD has a history of including test features in production chips and just quietly not using them if they don't work or aren't needed yet (so long as none of the planned features are affected).  The connection points needed for the cache chips in the 5800X3D gaming CPU and Milan-X server CPUs were included in every Ryzen die sold for at least a year before a product appeared that actually used them.
  
  So don't hold off buying an 7900 XT because of imaginary issues.  Hold off because at $899 the $999 7900 XTX is better value.
  
  And if you're spending $999 on a 7900 XTX you might as well spend $1199 on the RTX 4080, which although slightly slower on rasterisation, has much better ray tracing, and of course Cuda support, and faster OpenCL in many cases.
  
  And if you're spending $1199 on an RTX 4080, you might as well go all the way and spend $1599 on the RTX 4090, which is in a class of its own and the fastest graphics card you can buy today.
  
  Which nobody should do because $1599 is far too much money to spend on a graphics card.  (Never mind the Australian pricing, which is just horrifying.)
  

• I mentioned before that the upcoming lower-end Raptor Lake parts - the 13400 and 13500 in particular - may provide the best price-performance around for normal people who aren't buying RTX 4090s.  The 13400, for example, has the same configuration and offers similar performance the the previous generation's 12600K, which cost around 50% more.
  
  That may be because it is the previous generation's 12600K.  (Tom's Hardware)
  
  There's not a huge difference between the 13th generation cores and 12th generation; the major change is that Intel doubled the number of low-power Efficiency cores (E-cores).
  
  Since the 12400, 12500, and 12600 didn't include any E-cores at all, Intel can just re-use previous generation chips and not disable the E-cores this time around.
  
  Which sounds a bit sus but actually makes perfect sense; you get a faster CPU for your money and it's cheaper for Intel to produce.
  
  Ballpark numbers for multi-threaded workloads should put the 13400 at 25% faster than the 12400, and the 13500 at 50% faster than the 12500.  Performance for single-threaded tasks won't change much, but those E-cores - four on the 13400, eight on the 13500 - will stop Windows cluttering up your P-cores with the infinite amount of bullshit it likes to do in the background.
  
  If pricing is similar to 12th generation the 13500 would be my recommendation for this generation for anyone who doesn't need absolute maximum performance - and who isn't concerned about threads running at different speeds.
  
  I need a couple of PCs for the new office and while my main workstation will be a Ryzen 7950X, anything else is likely to be a 13500.
  
  
  
• Meanwhile in laptop land Intel's 13980HX will soon deliver 24 cores sort of.  (WCCFTech)
  
  8 Performance cores (P-cores) and 16 E-cores.  Sine E-cores run half as fast as P-cores, it will likely be the same speed overall as AMD's upcoming 16 core laptop CPUs.
  
  
  
• -108 diopters.  (Points de Vue)
  
  My eyesight is bad enough that I can't order bifocal or multifocal glasses online - I just have four pairs of glasses (reading, computer, distance, and sunglasses).  But by ordering online four pairs work out cheaper than what I used to pay for a single pair locally.  And also of course if I lose one pair there's another pair I can wear to find them.
  
  But -108 is a whole different ballpark, and indeed a new world record.
  
  
  
• If you're running a public Minecraft server between versions 1.7.2 and 1.18.2 - which is a lot of versions, since it updates once or twice a year - time to get a patched version now.  (Bleeping Computer)
  
  There's a worm in the wild actively exploiting unpatched Minecraft servers.  You can tell it's evil because it's indented by two spaces.
  
  
  
• This is that 8.4" tablet / gaming toy thingy I mentioned yesterday - the ONEXPLAYER 2.
  
  
  
  From above it looks perfectly reasonable and compact (when the controllers are detached) but as soon as he turns it at an angle you can see how chunky this thing is - it's a small production run of a complete PC with cooling fans, and thin is expensive.
  
  It's still an amazing little device and I'd love to get one.  You can take off the controllers and turn them into a Bluetooth game controller, and there's an optional keyboard and pressure-sensitive stylus.  And for the size, it's impressively powerful, with 8 CPU cores, 12 graphics cores, and up to 32GB of RAM and 2TB of NVMe SSD, plus a full 40Gbps USB 4 port.
  
  Not cheap though.

Posted by: normal at Monday, December 19 2022 12:31 AM (obo9H)

Posted by: normal at Monday, December 19 2022 12:59 AM (obo9H)

Hide Comments | Add Comment

• What do you call 1000 dead cryptocurrencies at the bottom of the sea?  A good start.  (Tom's Hardware)
  
  After everyone and their cat jumped on the bandwagon in 2021, 2022 saw a net reduction in the number of cryptocurrencies plaguing the world.  Not a dramatic reduction - a little under 10% - but we'll take what we can get.
  
  
  

Tech News

• Nuclear power is too slow - but it doesn't need to be.  (Jack Devanney)
  
  Back in the sixties, California built a 500MW nuclear power station in three years.  Reactor builds started in 1973 took an average of sixteen years to be completed and connected to the grid.
  
  The cause was a combination of a shortage of expertise at the high end, and an excess of idiots in government.
  
  But Japan maintained an average completion time of less than four years from 1963 through 2009, which shows what you can do if you feed regulators to the sharks.
  
  
  
• Installing Linux on a $8 computer.  (The Little Engineer That Could)
  
  The $8 computer in question is the 0x64 board, a RISC-V competitor to the Raspberry Pi Pico.  The CPU in the 0x64 has a proper MMU so it can run full-fledged Linux and not just an embedded version of it.
  
  The 0x64 dev tools are apparently still a bit rough and ready at this stage, and the author found that the easiest way to get Linux onto the board was to use a Pi Pico as a programming interface.
  
  
  
• The Periodic Tiling Conjecture is false.  (Quanta)
  
  If you're familiar with Penrose tiling we know that with as few as two suitably shaped tiles, you can tile a surface of any size without any gaps - but so that the pattern of the tiles doesn't repeat, something called aperiodicity.  But we also know that there is no single shape that holds that property, at least not for two-dimensional surfaces.
  
  The conjecture was that this would also hold in higher dimensions - any shape that could fill the space without any gaps could not also exhibit aperiodicity.
  
  Turns out it can.
  
  Keep that in mind when planning your next seven-dimensional bathroom or kitchen.
  
  
  
• The ONEXPLAYER 2 is an 8.4" tablet with a 2560x1600 screen.  (Liliputing)
  
  And a Ryzen 6800U and up to 32GB of RAM and 2TB of SSD.
  
  They took the two things I want that nobody currently makes - an 8" tablet with a high-resolution screen, and a Ryzen 6800U laptop with 32GB of RAM - and smooshed them together into a useless mess.
  
  Sigh.

Posted by: cxt217 at Sunday, December 18 2022 10:13 AM (2tHvf)

Hide Comments | Add Comment

• If you always wanted a quantum computer of your own but didn't have the millions of dollars and the endless rivers of liquid helium SpinQ has three portable models designed just for you sort of.  (Tom's Hardware)
  
  I say sort of because the theoretical capacity of a quantum computer scales exponentially with the number of qubits - each qubit essentially being another orthogonal dimension of parallel universes that the computer can compute in.  IBM just announced a 433 qubit quantum computer, and 2^433 is a very big number.
  
  This matters because if SpinQ announced a cheap portable quantum computer with a large number of qubits we'd know immediately that it was bullshit.  But the SpinQ qubit count maxes out at... 3.
  
  So, yeah, I'm prepared to believe that a $57,000, 90lb quantum that is slower than a first generation iPhone might actually exist.
  
  Probably not a hot item this Christmas though.
  
  
  
• All my remaining Amazon packages arrived today, including one from Kentucky that took an interesting route to get here.
  
  So I ordered some more.
  
  
  
• Guests arrive on the 27th.  Gotta vacuum, or paint, or plant vines, or whatever you're supposed to do when you have guests.

• Next-generation graphics cards could get more expensive thanks to the death of SRAM.  (Tom's Hardware)
  
  I think they mean even more expensive here because the cheapest card from the current generation has a list price of $899.
  
  
  
• IBM has announced a 24 core Power CPU aimed directly at Oracle's pricing model.  (The Register)
  
  Power-based servers tend to have fewer cores per CPU but lots of separate CPUs, which is great until you realise that Oracle prices per CPU rather than per core.  This version has up to three times the cores per CPU - and the only reason for it to exist is to run Oracle SE2 cheaply.
  
  If you go to more than two CPUs - regardless of the number of cores - the Oracle license cost per CPU almost triples, so running six, eight-core CPUs instead of two 24-core CPUs would run about eight times as much in licensing fees - with a support contract, about half a million per year per server.
  
  
  
• A Cambridge PhD student has solved a Sanskrit grammar problem that has puzzled experts for 2500 years.  (BBC)
  
  Which if you read the article makes the experts look kind of dumb.
  
  
  
• ChatGPT expects to make $1 billion in revenue by 2024.  (Reuters)
  
  I didn't realise there was that much demand for terrible Gizmodo articles.

Posted by: Rick C at Saturday, December 17 2022 01:37 AM (k3/O4)

Posted by: normal at Saturday, December 17 2022 03:14 AM (LADmw)

Posted by: normal at Saturday, December 17 2022 03:17 AM (LADmw)

Hide Comments | Add Comment

• ChatGPT wrote a terrible Gizmodo article.  (Gizmodo)

  Suffice it to say, multiple attempts brought less than satisfactory results. For one thing, ChatGPT kept including incorrect information in its explainer—sometimes mixing up basic facts about the history of its own technology (factual inaccuracy has been an ongoing problem for the program). In other instances, it skimped on the details in critical areas, writing text that read more like OpenAI ad copy than a thoughtful explainer. In at least one instance, it forgot its own name, repeatedly telling me that it didn’t know anything about a "ChatGPT" and thus couldn’t help with my request. In general, it struggled to find a good balance of factual information, story structure, and accessible, humorous language, when putting its stories together.

  The comments are exactly what you would expect:

  The joke is so obvious, I’m not even going to bother.

  
  
  
• Looks like all my Amazon packages that I was hoping would arrive by Christmas, will.
  
  One seems to have taken three days by plane to travel from the US to Australia, which is odd.  But given that the expected delivery date was January 11, the fact that it has already arrived in Sydney and left again for regional NSW is promising.
  
  
  

Tech News

• Samsung has used PIM to speed up AMD's Instinct MI100 GPUs by 150% while reducing power consumption by 60%.  (Tom's Hardware)
  
  PIM stands for Processor in Memory - adding compute logic directly into the memory chips.  That means data doesn't have to be read from those chips, processed, and written back; everything happens in one place.
  
  That does depend on all the data required for a given computation be located on the same chip, so it's not remotely a general-purpose solution.  But in those cases where you can use it (like training chatbots to write terrible articles for terrible tech news sites) it can bring huge benefits.
  
  
  
• Meanwhile SRAM is dead.  (Wikichip)
  
  Moving from 7nm to 5nm, logic sizes shrank by close to 50%, but SRAM sizes only shrank by 20%.  Moving from 5nm to 3nm, it looks like SRAM sizes will only shrink by 5%.
  
  That's not good, because a lot of performance depends on getting data into large on-chip caches.
  
  AMD has already tackled this with its 5800X3D, which stacks a separate SRAM chip on top of the CPU, with thousands of interconnects, and with the just released Radeon 7000 range, which separates out the cache and memory controllers onto their own chips on a multi-chip module.
  
  Expect more of that in the future, and in particular more vertical stacking.
  
  
  
• The iKOOLCORE R1 is another of those weird little Chinese micro-PCs.  (Liliputing)
  
  This one has four 2.5Gb Ethernet ports, aimed at the small router/firewall market.  It also has a single HDMI port, a USB-C port, and two USB-A.  At three inches square and two inches high, it's even smaller than a NUC.
  
  Prices start at $145 - in China.
  
  
  

Posted by: normal at Friday, December 16 2022 02:28 AM (LADmw)

Hide Comments | Add Comment

• Work 16 years and what do you get?  A glass of champagne and six million net.  (PC Gamer)
  
  Cult game Dwarf Fortress, which first appeared in 2006 and has been available for free ever since (and is still available for free) finally has a Steam edition with graphics and has sold 300,000 copies in its first week.
  
  Even after the cuts for Steam and publisher Kitfox, the game's two authors each just retroactively made 16 years worth of six-figure salaries.
  
  Advice: Hire a really good accountant.
  
  
  
• I started playing Pathfinder: Kingmaker last night.  Oops.  Haven't played any RPGs in a long while and even though this one isn't some legendary classic I was still playing six hours later.
  
  Also, Nvidia MX350 laptop graphics?  Basically worthless.

• If you have a current year Tesla Model S or X you can probably play Pathfinder: Kingmaker on that and have a better time than me on my laptop.  (Tom's Hardware)
  
  The latest operating system update (Teslas run Linux) supports Steam games that will run on the Steam Deck (which runs Linux).  The games don't need to support Linux directly; the Steam Deck and the Teslas both support an emulation layer.
  
  
  
• Speaking of old Nvidia hardware, If you were waiting for cheaper cards in the new RTX 4000 generation you might just be better off buying whatever is cheap right now.  (Tom's Hardware)
  
  Today's leaked specs are for the 4060 Ti.  While it has more compute power than the 3060 Ti, it has half the memory bandwidth.  That's going to hurt, and it's likely to be significantly more expensive than the 3060 Ti as well.
  
  
  
• China has banned exports of its home-grown Loongson CPUs.  (Tom's Hardware)
  
  Not that anyone wants them (except Russia) - they're about ten years behind Intel and AMD.  It's likely because China has very limited fab capacity even at 14nm and needs to keep all the chips for internal use.
  
  
  
• China is also preparing a $140 billion bailout package for its floundering chip industry (Reuters) and at the same time complaining to the WTO about other countries being mean to it. (Tom's Hardware)
  
  Oh no.
  
  Anyway...
  
  
  
• Everything in this article is wrong.  (Jesse Li)
  
  It popped up in one of my feeds today, though it's a couple of years old.  Everything it says is wrong.  It's quite amazing how much wrongness the author has managed to pack in.
  
  
  
• A native internet protocol for social media.  (Revue)
  
  Oh, it's by Jack Dorsey.  He makes one good observation:

  The biggest mistake I made was continuing to invest in building tools for us to manage the public conversation, versus building tools for the people using Twitter to easily manage it for themselves.

  The point is, everyone knew this was wrong, and he did it anyway, and it cost someone $44 billion to start fixing the mess.
  
  
  
• Speaking of which Twitter has dissolved its Trust and Safety Council hopefully in battery acid.  (NBC)
  
  The Trust and Safety Council was a group of radical left-wing pro-censorship volunteers who guided the Trust and Safety Team, a group of radical left-wing pro-censorship employees.
  
  Good fucking riddance.
  
  
  
• DoNotPay is an AI chatbot based on GPT-3 that argues with billing departments on your behalf to cancel your subscriptions.  (The Verge)
  
  See?  Technology isn't all bad.

Posted by: normal at Thursday, December 15 2022 01:09 AM (LADmw)

Posted by: Pixy Misa at Thursday, December 15 2022 03:57 PM (PiXy!)

Hide Comments | Add Comment

• Sam Bankman-Fried has unfortunately been forced to decline Maxine Waters' kind invitation to give testimony before Congress today due to (checks notes) having been arrested on charges of wire fraud, wire fraud conspiracy, securities fraud, securities fraud conspiracy, money laundering, breaking ranks while in formation, felonious assault, indiscriminate behavior, mopery, high treason, provoking, being a smart-guy, and listening to classical music.  (WCCFTech)
  
  He spent his last day in freedom as he has been spending every day these past few weeks: Further incriminating himself.
  
  
  
  
  How was this clown ever allowed outside without a leash?
  
  
  

Tech News

• This is real:
  
  
  
  
  
  
• AMD's Radeon 7900 XTX, XTXX, and XXTX are here.  (Ars Technica)
  
  They're pretty good, though certainly not cheap.  Except for ray tracing, the 7900 XTX consistently outruns the RTX 4080 for $200 less.  And has 50% more RAM.
  
  But at $200 less than the 4080 it will still set you back $1000.
  
  
  
• SK Hynix has showed off 8GHz server RAM, using two interleaved sets of DDR5 chips.  (AnandTech)
  
  That's quite fast.  Particularly considering something like the latest Epyc Genoa servers have 24 memory channels across two CPUs.
  
  These particular modules look like they're Intel-only initially, but that will change.
  
  
  
• Next year's Meteor Lake chips from Intel will use a new socket - LGA 1851 - and despite the headline will most definitely not support LPDDR5X-7500.  (WCCFTech)
  
  Because socket LGA 1851 is for desktops and LPDDR5X is for laptops.
  
  And allegedly will have a maximum of 6 P cores, which is a step backwards.  Doesn't matter for the average user, but if true there won't be a new high-end chip from Intel until 2024.
  
  
  
• AI is crap because researchers spend all their time teaching it not to ever offend people who are actively seeking to be offender rather than to do anything remotely useful.  (Astral Codex Ten)
  
  Also, ChatGPT is a wonderful front-end to a universal search engine except that if ever it cannot find an answer to your question it will make something up rather than risk disappointing you.
  
  
  
• You might be able to buy a Raspberry Pi next year.  (Liliputing)
  
  Also, the people running Mastodon nodes are all insane.  That might seem unrelated, but read the article.

Posted by: Rick C at Wednesday, December 14 2022 01:24 AM (BMUHC)

Posted by: Rick C at Wednesday, December 14 2022 01:40 AM (BMUHC)

Posted by: normal at Wednesday, December 14 2022 02:29 AM (LADmw)

Hide Comments | Add Comment

• NASA's Orion (not that one) Moon thingy has safely landed back on Earth.  (WCCFTech)
  
  As part of the Artemis 1 mission it travelled to the Moon, orbited it - sort of - and returned to Earth 25 days later.
  
  NASA is planning a manned lunar orbit for 2024.
  
  Which is great and all, but (a) Apollo 8 did that in 1968, (b) private Japanese company ispace with only 200 employees launched an unmanned lunar lander yesterday, and (c) SpaceX plans to land a fully crewed Starship on the Moon in 2024.
  
  Which date might slip a bit because Starship has not yet had a successful orbital flight - and has not been cleared by the FAA for orbital tests.  On the other hand, SpaceX has six Falcon 9 launches scheduled before the end of the year, so they're not just sitting around trolling idiots on Twitter either.
  
  
  

Tech News

• If you want to get your hands on a classic Unix workstation like the SGI Indy or Sparcstation 10, it might be time to learn FPGA programming and look into large-format 3D printers.  (OS News)
  
  Because increasingly there aren't any to be had, and what there are cost more than a new PC that runs anything from 10 to 100 times faster.
  
  And if you manage to get your hands on one, you can't get the installation disks, and if you do, somehow, forget about any software patches.  It's all dead and forgotten.
  
  I still have an SGI O2 and a Sun Ultra 5, but I don't know whether they work at all.   Last time I booted up the O2 - probably more than a decade ago - it worked fine except that one pair of DIMMs had failed, leaving it with all of 192MB of RAM.
  
  
  
• Sam Bankman-Fried hopes to find new suckers to fleece of $10 billion so he can set that on fire too.  (BBC)
  
  Err, pay back the previous suckers.  That's totally what he meant.
  
  Can we send this guy to jail already?  Just to make him stop talking.
  
  
  
• The Core i5-13400 is 20% faster than the Core i5-12400.  (Tom's Hardware)
  
  The 12400 and 12500 both had 6 P cores and zero E cores.  The 13400 and 13500 both have 6 P cores, plus 4 and 8 E cores respectively, giving then a significant boost overall.  (Intel's E cores run about half the speed of its P cores, so these effectively give you 8 and 10 cores of performance, respectively.)
  
  If the new chips fall into the same price brackets as the old ones, they'll be the best value CPUs for the average user who wants to be able to play some games as well as run spreadsheets and browse the web.  Intel motherboards are cheaper, and available with DDR4 support so the memory is cheaper as well.
  
  While they won't break any records, both have better single-thread and multi-thread performance than the 11900K from less than two years ago while using about half the power.
  
  I still prefer AMD for high-end systems but these look to be very attractive for anyone on a budget.

Posted by: birnak at Monday, December 12 2022 10:18 PM (obo9H)

Posted by: Mauser at Monday, December 12 2022 10:25 PM (BzEjn)

Hide Comments | Add Comment

• Inside the frantic texts of the illegal crypto trading cartel as one of their more obvious Ponzi schemes imploded and threatened to take other, better camouflaged schemes with it.  (New York Times / MSN)

  Stop now, don’t cause more damage.  The more damage you do now, the more jail time.

  While I may have elaborated on the headline, that is a verbatim quote of a text from Binance CEO Changpeng Zhao to disgraced Democrat campaign finance launderer Sam Bankman-Fried.
  
  
  
• Crypto exchange Kraken has laid off 30% of its employees - 1100 people.  (Kraken)
  
  What were they doing with nearly 4000 employees in the first place?
  
  
  
• All the crypto doom and gloom you could possibly want is on tap at Web 3 Is Going Great.

• Breaking up with JavaScript front ends, or, how not to communicate anything, painfully.  (Triskweline)
  
  A PowerPoint presentation in web form that leads to a demo app that doesn't demo anything like what the slides say it does - particularly when the same demo app is linked from two slides supposedly demonstrating entirely different things - rather muddies a fundamentally sound point that JavaScript front ends are dogshit.
  
  
  
• Twitter Blue is relaunching tomorrow.  (WCCFTech)
  
  $8 per month via the web.  $11 per month if you're stupid enough to sign up in the iOS app.
  
  I'll sign up.  Not because I want to be verified (I don't) or want the clout (I don't), but because it's worth it just to annoy all the right people.
  
  
  
• Fractal Design goes Danish modern.  (Guru3d)
  
  In black or white, with a choice of mesh or tempered glass side panels, and walnut or oak front panels.
  
  It actually doesn't look terrible.
  
  
  
• What's going on in Alzheimer's research.  (Quanta)
  
  A long article examining the situation as scientists move away from the amyloid plaque model that seems now to have been overstated at best - and outright fraudulent at worst - after frittering away twenty years and two billion dollars.
  
  
  
• Where - and what - is dark matter?  (Big Think)
  
  Dark matter is what you get when you map out the visible matter of galaxies and compare it with the total mass based on how fast the galaxies rotate.  We can't see dark matter - it's dark - but we know there's something there because what we can see doesn't add up to enough mass to stop galaxies from flying apart.
  
  This article examines the possibility that dark matter is truly dark, that it only interacts via gravity.  If this is true, then we have no other way to measure it.
  
  At least with neutrinos (which only interact via the weak force) we have a convenient local source running 24/7 (that shiny yellow thing you see in the sky now and then) and the little buggers hit our detectors buried at the bottom of salt mines so that we can learn something about them.
  
  In the "nightmare scenario" considered here (physicists mostly have nightmares about grant applications, but this is one of the exceptions) no particle collision experiment - whether the naturally occurring variety using the Sun as a convenient lab partner or the artificial variety using Switzerland - will ever provide any information about the nature of dark matter.

Pixy Is Watching

Posted by: Pat Buckman at Monday, December 12 2022 03:31 AM (r9O5h)

Posted by: Pat Buckman at Monday, December 12 2022 03:45 AM (r9O5h)

Posted by: normal at Monday, December 12 2022 04:51 AM (obo9H)

Posted by: Frank at Monday, December 12 2022 06:41 AM (rglbH)

Posted by: Pixy Misa at Monday, December 12 2022 11:30 AM (PiXy!)

Hide Comments | Add Comment

• The CEO of crypto news site The Block has resigned after it was revealed that collapsed Ponzi scheme Alameda Research provided the funds for him to buy out the investors in, well, crypto news site The Block.  (Crypto news site The Block)
  
  Crypto news site The Block swears this did not affect its coverage of collapsed Ponzi scheme Alameda Research in any way, because there wasn't any.
  
  
  
• Meanwhile in Europe a key policy maker on cryptocurrencies and one of the vice presidents of the European Parliament (how many do they have?) has been arrested on corruption charges involving Qatar.  (Crypto news site The Block)How could this happen?

  The Greek policymaker has been suspended from the Socialists and Democrats parliamentary group as well as the Greek national party PASOK until further notice.

  Oh, right.
  

• Here's a 1300W PC power supply in case you need the edge over your crazy neighbour with his 1000W power supply.  (AnandTech)
  
  Not sure exactly what this is for; it's overkill even for a 13900KS and a 4090.
  
  
  
• 13900KS?  Oh, yeah, there's that as well.  (Tom's Hardware)
  
  It's just the 13900K with a small clock speed bump, but that bump takes it to 6GHz, so again useful for getting an edge over your crazy neighbour with his 5.8GHz CPU.
  
  
  
• Sana, Pina, and Pomu (my new server cluster) are all currently running memory tests before being pressed into production.  Zero errors so far.
  
  This site (no matter which site you are reading this post on) will be moving to the new cluster in early January.  Not to one server in the cluster, but to all of them at once.

Posted by: Starry Knight at Saturday, December 10 2022 10:45 PM (gXw+B)

Hide Comments | Add Comment

• How much RAM do you need in your Windows PC?  Don't ask a journalist, they only use Notepad.  (ZDNet)
  
  Get a load of this:

  The time when more than 8GB of RAM becomes useful and starts paying for itself is when you're running several resource-heavy applications simultaneously -- especially high-end image or 4K+ video processing, CAD, or 3D modeling.

  Yeah when you're running Adobe Premiere, Photoshop, and Maya at the same time, 8GB might get a bit limiting.  64GB might also get a bit limiting.  People who run that sort of workload tend to buy systems with 256GB or more.
  
  I happened to reboot my laptop this morning.  By the time it had finished booting, before I opened a single application, 8GB of RAM was already in use.
  
  
  
• Two of the three new servers have been deployed, just waiting on the third now.  Those have 128GB of RAM each.  128GB is probably enough.
  
  I now have an 84TB RAID-Z3 array with gzip compression and block deduplication. I have about 7TB (already gzipped and deduped) to copy onto it, so it should last a good long while.
  
  Update: Third server has been deployed.  Tentatively named Sana, Pomu, and Pina (from Hololive, Nijisanji, and Prism Project respectively).  They're running memtester right now because the last thing I want is to get everything set up and find out I've got bad RAM.
  
  
  

Tech News

• Locating the new stealth bomber using the stars in the sky, JPEG metadata, and guesswork.  (Twitter)
  
  Looking at the photo, my first thought is there's no way that contains enough information to triangulate, even with a timestamp.  And it doesn't.  But the arc of possible matching locations in the United States comes near Edwards AFB, and if you look at all the buildings at Edwards in Google Maps there's a clear match to the photo.

  I'm sure this entire process could be done with just stars and no google maps as well

  Dude, using just stars they were lucky to discover America, which is somewhat larger than one aircraft hangar at Edwards AFB.
  
  
  
• Part two of the Twitter Files is out.  (Twitter)
  
  Here we see the usual suspects working to twist the rules to justify silencing inconvenient truths.  And they were dumb enough to write it all down.
  
  Nothing we didn't already know in general, but with more specifics to throw in the face of anyone insisting it didn't happen.
  
  Like Jack Dorsey, who was CEO at the time, and testified before Congress that none of this was taking place.

Posted by: normal at Friday, December 09 2022 07:46 PM (obo9H)

Posted by: Rick C at Saturday, December 10 2022 01:14 AM (k3/O4)

Posted by: Frank at Sunday, December 11 2022 03:04 PM (rglbH)

Hide Comments | Add Comment