Always strive to learn something useful. --Sophocles

A new AI company is starting up....

2023-07-07T12:19:15Z

I'm not going to be talking about the company per se. I'm talking about a generalized problem with AI and a very real problem with the worldwide growth of data centers.

This company just built a supercomputer with 22,000 H100 GPUs. That is TWENTY-TWO THOUSAND EXTREMELY HIGH-END SPECIALIZED GRAPHICS CARDS.

This computer consumes THIRTY-ONE MEGAWATTS OF POWER. That is how much power 6,000 HOMES consume in a DAY.

There are questions as to how this almost constant construction of data centers are going to be powered. For example, in Dublin, Ireland, they have a moratorium in building data centers in the town center because they can't power them without cutting power to the residents! And there are more important questions as to how all this heat will be dealt with.

A few months ago I posted about a data center in England that put a pod of computers into a mineral oil bath and is using a heat exchanger to help heat a local pool. That's pretty awesome. But there are a finite number of pools that need heating. Microsoft and others have been experimenting with putting pods of data centers underwater for cooling.

But the demand for AI has just begun. I just read an article about Y Combinator, a tech startup incubator, over 35% of new projects coming to them for assistance are AI.

We have not yet begun to truly cook this planet! Makes me glad that I'll be gone in 20-30 years or fewer.

Oh, and those H100 GPUs? Can be used for no other purpose. Like crypto mining cards, they can be used for only high-intensity computing. Now, those 22,000 cards don't represent 22,000 computer chassis. They can put like 10-20 cards in one chassis. The cards are designed with no fans and the chassis have big forced air and water-cooling systems to keep them cool. Very different from most conventional PCs.

So the next time someone says global warming isn't real, tell them about ONE supercomputer eating the power of 6,000 homes!

comments

Dell in BIIIIG trouble in Australia for not discounting the price of apparently discounted monitors

2023-06-07T21:34:20Z

Oooh, this was a sneaky one! The investigated period, from August '19 to the middle of December '21, they looked at people buying systems at the Australian Dell web site. People would see a monitor with a high price that was struck through and a lower price displayed. What people didn't know was that the lower price was in many cases actually higher than the price that they would pay if they'd bought the monitor on its own!

During the time in question, "...shoppers spent over $2 million Australian dollars ($1.33 million USD) on 5,300 add-on monitors..."

LUCY! YOU HAVE SOME 'SPLAININ' TO DO! IN THE COURT ROOM!

https://arstechnica.com/gadgets/2023/06/dell-in-hot-water-for-making-shoppers-think-overpriced-monitors-were-discounted/

comments

RIP: Gordon Moore, 94

2023-03-25T17:16:57Z

While Gordon may not be a household name, you might be familiar with a little company that he co-founded a lot of years ago: Intel.

He also coined the phrase Moore's Law, in which he states that transistor density will double about every 18 months. And it was largely proven correct. Now, the interesting thing is that he posited Moore's Law BEFORE the integrated circuit was invented!

He had an interesting life, which included him working with Robert Shockley, a person whose name should be familiar with anyone heavily interested in electronics: he won the Nobel Prize as one of the co-inventors of the TRANSISTOR.

Gordon founded a philanthropic organization that gave out over $5 billion in grants and such in the San Francisco area.

Sounds like one heck of a life.

https://www.huffpost.com/entry/gordon-moore-dead_n_641ebdbce4b0b8ee3bd25f05

https://www.intel.com/content/www/us/en/newsroom/news/gordon-moore-obituary.html

https://hardware.slashdot.org/story/23/03/25/0044230/intel-co-foundercreator-of-moores-law-gordon-moore-dies-at-age-94

comments

Ecuador: someone is mailing USB sticks with explosives in them to journalists

2023-03-23T14:32:29Z

One exploded as a journo at a TV station plugged it in to his computer, causing mild injuries to his hand and face. Another sent to a radio station did not explode - the drive was plugged into an extension cord and didn't get enough voltage to detonate.

https://arstechnica.com/gadgets/2023/03/journalist-plugs-in-unknown-usb-drive-mailed-to-him-it-exploded-in-his-face/

https://it.slashdot.org/story/23/03/22/2048210/explosives-replace-malware-as-the-scariest-thing-a-usb-stick-may-hide

comments

Data center warming public children's pool saves the pool operators an estimated $24k annually

2023-03-18T18:10:40Z

The high-performance cluster server is submerged in a pool of mineral oil - a very effective way to cool systems - and a heat-exchanger transfers the heat into the children's pool! The pool still has a boiler in case it gets too low, but overall it will save a lot of money.

The company behind the install is trying to create more installations like this to recover waste heat - this pool installation claims to recover 97%(!) of its waste heat - rather than just venting it into the air and causing more climate disruption.

https://arstechnica.com/information-technology/2023/03/free-data-center-heat-is-allegedly-saving-a-struggling-public-pool-24k-a-year/

https://hardware.slashdot.org/story/23/03/17/2037217/free-data-center-heat-is-allegedly-saving-a-struggling-public-pool-24k-a-year

comments

More on PDFs and OCR

2019-03-06T05:45:32Z

More things learned today:

Quality of OCR in Acrobat Pro decreases (fragment rate increases) if the:
-- Font is Bold
-- Page isn't flat
-- Page is skewed

I knew about the second and third, the first one surprised me. It might have been increased by non-perfect flatness and skew.

One thing that surprised me was what I learned when I did a little digging beneath the shortcut that launches the program. The executable was dated 2014! They bought the system in '15, and the software was pre-installed, so nothing has been updated since it was first set up. The head librarian is going to look for the contact info for the salesman to see what can be done for an update. Their software neither checks for updates nor has a menu function for checking for updates. I went to the manufacturer's web site and the actual software that we use isn't listed! I think they've upgraded to something with a different name, so I don't know if we'll be able to update it.

I scanned one of the hardbounds from back to front, and the behavior of the software building the PDF backwards was consistent: the PDF was in the correct sequence since I scanned it backwards.

I tested how long it took to fix fragments by repairing the first ten pages of one of the 40+ page PDFs, and it came to 3-4 minutes per page, meaning 2+ hours for the large PDFs. Speaking with the head librarian, she didn't think it was a good use of my limited time right now, so we're not going to do it. I think it's a good call, we can focus on the core job of getting all of the reports scanned which is the main goal of my internship. The OCR is good enough that anyone who wants to do Finds on these documents will have reasonable hit rates. Still, spending half an hour fixing 10 pages was a good use of my time to determine that 3-4 minutes per page number.

I learned of the bold problem seeming to increase fragment rates while I was doing the edit. The second page of these reports is a table of contents with an index entry on the left and then periods filling to the page number on the right, and the entire page is bold. On many lines the program identified these repeating periods as fragments, I'd have to tell it that these are not words. If the page had been laid out in an actual publishing program with proper kerning and such, maybe it would have scanned better and the OCR would have performed better, I don't know. It's definite that just bolding the entire page made it harder to read.

And this highlights a programming weakness in Acrobat Pro. The fragment interface has an option to highlight all fragments in the document, but if you skip over one, you have no way of going back to the one that you skipped. Bad interface design! But it's also Acrobat v11, which is long past support date, so I guess that I shouldn't be surprised. It's possible that current AcroPro versions have improved functionality, I wouldn't know: my version on my Mac is version 10, a generation older!

comments

Learned some interesting things about scanning books and OCR processing today

2019-03-05T07:19:19Z

I'm doing an internship in our local university library through April, and my main task is scanning their annual 'Reports To The President', a précis of college activity sent up to main campus and bound in a book, usually hard-bound. The oldest book was 1965-66, the newest that I've seen thus far is '98-99. I believe there are newer already in PDF format online on the local network. Apparently by scanning them and then coding an RDA record for each file, we can get them hosted by the state academic library organization, or somebody, for free.

So that's cool.

I'm using a fairly spiffy Fujitsu specialized document scanner that can scan two pages of a bound book in one pass, but I don't think their software is as good as they claim it to be. It can handle a pretty significant amount of curvature in the books - for example, I was scanning three pages starting at page 385 of 427, so LOTS of curve when you're that far in. I was holding up the left side to get the right page reasonably flat, then holding down both sides with one finger.

And yes, the fingers were captured by the scan.

After you've done your scan, you get into the next phase, where you drag this wire frame to line up one line down the spine between the two pages, then you align four corners to the outside corners of the pages. The program does a good job of detecting the edges and snapping to it, but sometimes you have to do some dragging to improve alignment. Once you've aligned all the scanned pages correctly, you click an Apply button and it re-cuts the scans into individual pages and flattens them, programmatically removing the curve. It does a very good job, though not perfect.

THEN you have to go back through every page and remove the fingertips! It has a special tool just for it and works a lot like Photoshop's patch tool, but it auto-selects the fingertip. Click Apply, and the fingertip vanishes.

Once you've removed the fingertips, you can save it to PDF. Theoretically the program performs OCR (optical character recognition), but I can't see that it has any effect. I end up loading the PDF into Acrobat Pro and running OCR there.

And this is where I learned something tonight. While you can't do a spell-check on a scanned document because you're dealing with a scanned image, not words, there's something that's similar: a fragment check. Fragments are words that Adobe Acrobat recognizes as 'I think this is a word or something, but I'm not sure, therefor I didn't map it into the OCR side of the document. Fix it.' Acrobat can't provide a dictionary of suggestions like Word, so when it sees something that it thinks was a word but it couldn't map, you have to type the correction. Or page number. Or budget number. Or tell it to ignore it.

It took me a good half an hour to fix a three page document. I don't know how many times I typed the San of San Juan. Just the San, apparently Juan was recognizable.

And that was a three page document from '67-68. The latest document from '98-99? That was 40some pages, I'm going to run a fragment check on it tomorrow afternoon and we shall see how long it takes to fix.

One very odd thing about scanning two pages at once in bound books - the page sequence is reversed! This is easily fixed in Acrobat Pro when you're dealing with a handful or two of pages, you just slide page thumbnails around. But dealing with 30 or 40 pages? Next week I'll try scanning a book starting with the last pages and working my way forward and seeing how that works.

So important tip when creating PDFs from scanned docs for public consumption: running OCR is only half the job. If you need the document to be searchable, you MUST spend the time to run a fragment check on it and fix all of the problems! Otherwise you're going to frustrate anyone needing to do anything serious with the document.

One thing that makes me really wish I had a working Mac laptop: I'd like to take an unfixed doc and run it through text to speech and see how it works. Then run the fixed doc through TTS. Might be interesting.

comments

Apparently the latest Windows update continues to break things!

2018-11-27T21:26:07Z

Specifically, Windows Media Player and 32-bit apps!

WHEEEE!

This may be a carryover from the problems with the October update, hard to say. There's also apparently issues with iCloud installations: they may be blacklisted from receiving the 1809 update, which is fine by me as I have iCloud installed and I definitely don't want a buggy update installed!

https://arstechnica.com/gadgets/2018/11/latest-windows-10-update-breaks-windows-media-player-win32-apps-in-general/

comments

There's a MacHeist Black Friday sale going on that has some attractive stuff for $25

2018-11-21T04:02:57Z

There's some pretty good stuff in this deal for $25, particularly the VPN and dupe finder. mSecure is a program that I've been using on my iPhone for pretty much as long as I've had one, and I've found it invaluable: it'll be nice to have a desktop version. And the PDF/OCR program sounds like it'll be quite useful with all of the medical records that I maintain. I'm not familiar with Acorn, it says it will open Photoshop PSD files, so it'll be worth some time for me to check it out.

Your mileage may vary.

And if you're not a Mac person, why are you bothering with this post? ;-)

https://www.macheist.com/sales/the-award-winning-black-november-mac-bundle-ft-acorn-6

comments

There's nothing that succeeds like a budgie with no teeth!

2017-09-05T22:46:14Z

Or a computer geek with something to prove!

I succeeded in sucking the files off my wife's computer, though it took some time to configure it. I couldn't do it with a Win 10 computer, so I dug out the OOOOOOLD box -- XP. Only two generations newer than the Compaq. They both spoke the concept of Workgroups, which 10 does not. Once I set up a common workgroup and confirmed they could see each other through the router with a ping, I created a share on my XP box, linked the Compaq to it and had no problem copying stuff to it. I even went and xcopy'd her bookmark*.html to it so if she had anything interesting there, it'll be preserved.

I even found her PhD thesis from 20 years ago in LaTex files!

It was pretty small, only 100 meg. Copied it on to a flash drive, then up to Dropbox.

Mission accomplished. I'm leaving everything set up so she can review the Compaq before I break it down, then we have to figure out what we'll do with the old beastie.

comments

Heh. Sometimes I happily outclever myself with computers.

2017-09-05T19:56:28Z

I've been on a mission of late to purge a storage locker to eliminate (A) a bunch of crap that I don't want or need, and (ii) a monthly debt that we definitely don't need as I'm in my 14th month of being unemployed. To further get rid of clutter, and to get some files off of my wife's ancient Compaq desktop (Windows 98 anyone?), I ordered a Vantec IDE/SATA to USB 3 adapter from NewEgg for $20 with free shipping. That allowed me to check A LOT of hard drives that have been sitting around for files that I might be interested in. Only one, clearly labeled BAD, was not readable. Some had already been wiped. Unfortunately pretty much all of them are unusable: the smallest is an 800 MEGABYTE laptop drive (HUGE amount of space in DOS days), and aside from an old 4 TB drive from my previous iMac, the largest was only 8 gig. So I'll end up opening them up, stripping some of them for their magnets and platters because it's fun, and tossing the rest.

My wife's Compaq posed a different sort of problem. First, let it be proclaimed far and wide, that as a rule I HATE COMPAQ COMPUTERS! HP also falls under this broad proclamation. This goes back to the '80s when I first had to open them up and work in their guts. They're notorious for being fickle in their configuration and requiring that you buy components only from their makers. Well, this one is not much different. It's running Win 98, and I didn't want to try and put it on a network since the concept of trying to find network drivers for such an old OS, not to mention an OS CD!, was pretty much unthinkable. And I had just bought this spiffy Vantec adapter, so I figured I'd just remove the hard drive and suck it in to my iMac and copy the contents on to a USB stick for my wife.

Fat chance. First off, they secured the hard drive with screws on both sides. The motherboard is on the far side, so those screws are inaccessible. So remove the entire hard/flopy drive cage! They even secured THAT with screws on both sides! So without removing the motherboard, I can't get the damn hard drive out. I tried undoing the ribbon cable header and plugging in the Vantec adapter, but for reasons unknown that did not work: I used the Compaq itself to power the drive during that experiment. Now I'm going to grab my Windows 7 box and my router, plug both computers in to the router side-by-side, and get them talking to a local workgroup so I can just suck the files directly to the Win 7 computer. THEN I can copy the files to a USB stick. I couldn't even get the 98 machine to recognize a 256 meg USB stick without it wanting additional drivers!

Hopefully this will work. If not, I have an even older XP machine sitting in the corner that should boot.

But here's the clever bit.

I had an old 15" CRT VGA monitor sitting around, so a monitor for the Compaq was set. I was able to find a PS/2 keyboard in its original box, so that's set. Couldn't find a mouse. Not a big deal, I can navigate old Windows with a keyboard just fine, it's just easier with a mouse. But while cleaning out my storage shed, i found a full box of Belkin mice! I'm talking an actual factory box with something like 10 mice in it, all of them PS/2 or serial! I brought one home and plugged it in!

And it only moved the cursor sideways.

Being the geek that I am, and more than a little handy with a screwdriver, I took it apart.

The vertical sensor axle had popped out of its far side. Simply popped it back in to place, snapped the housing back together, popped the ball back in (yes, pre-optical mice), and it worked just fine. THEN I put the screw back in.

So for the first time in almost 35 years of working with IBM PCs, I fixed a mouse.

Now let's hope my little networking misadventure works. My previous wireless access point had the radio die, but the router side continues to work just fine, so I've kept it in service for that use as my Apple wireless router has only one user ethernet port.

(Tech lesson: the way that the original mice worked when they had the little rubber balls was that they had two sensors, X and Y axes, connected to rods that were in physical contact with the ball and rotated as you moved the mouse. Electronics in the mouse and software translated the rotation of the rods for the computer to move the mouse pointer on the screen correspondingly to the mouse movements. Very clever design. Then they went to optical mice, 'rubber eraser' pointers, and trackpads. Trackballs were just upside-down mice with billiard balls, nothing particularly innovative there. The rubber eraser pointers used strain transducers to sense where the pointer was being bent and how hard it was being pushed to provide mouse pointing information, trackpads use a similar strain transducer tech.)

comments

For today I am a man

2017-07-09T05:02:36Z

I successfully replaced the battery in my wife's previous MacBook Pro AND transplanted my hard drive from my MBP in to hers. Mine, aside from also needing a new battery, has developed a fault with the MagSafe port and I don't have the bucks to replace it. Fortunately, they're both 2011 model year computers, which made everything pretty easy.

In other Rights of Manhood performed over the last couple of weeks, I assembled my first Lego kit: a Batman/Phantom Zone thingy from the Lego Batman movie! It was a giveaway when I saw the movie, it's been sitting on my desk for ages. AND I baked a cherry pie which was extremely successful! More on that, along with pix and recipe, later.

comments

U.S. Supreme Court protects rights to buy refilled toner cartridges!

2017-05-31T14:40:16Z

Finally it has been decided. A long time ago in this galaxy, Lexmark filed a suit against a company called Impression who not only refilled Lexmark-brand toner cartridges, but Impression also jiggered with a chip that Lexmark built in to the cartridge. Lexmark claimed that this was a DMCA violation. Impression said that Lexmark lost its patent rights once the cartridge was sold as part of first sale doctrine, and finally the highest court in the land agreed.

The basic issue has been that all printer manufacturers have been selling printers at cut-rate prices, expecting to make huge profits on ink cartridges. To ensure this, they followed Lexmark's and HP's leads by putting microchips in the ink cartridges that told the printer that these were "Genuine" cartridges - accept no substitutes. Or if a substitute were to be found, bitch endlessly that a substitute was present and that a complete meltdown was imminent and that it was all the printer owner's fault for not using Genuine Ink or Toner Cartridges! And it was illegal, or at least a violation of the Digital Millennium Copyright Act, to break the code in the chip and spoof that the third-party refilled cartridges were original. Sometimes the printer would lie and say the third-party cartridge would exhaust quicker.

So it's all over, barring printer manufacturers buying more congressmen to change the laws to make it illegal again. We can not only legally get ink cartridges refilled, we can legally get the chips reset.

https://hothardware.com/news/us-supreme-court-protects-consumers-right-to-refill-ink-cartridges

https://hardware.slashdot.org/story/17/05/30/171253/us-supreme-court-protects-consumers-right-to-refill-ink-cartridges-in-precedent-setting-lexmark-vs-impression-case

comments