Posts by Bitsminer

$1/hr/GPU

Let's do some math here:

There are 8000 hours in a year (give or take). Assumption #1: a 3-year payback is needed as the tech goes obsolete pretty quick. That's $24k to play with based on customers with long-term commitments.

Your biz needs at least 50% margin(*) else nobody will invest, so you have $12k to play with now.

Your electric, plant, other utilities is about maybe 30%.

Your actual capital cost (of the RAM + CPU + GPU + network + + +) is about maybe 40% assuming (assumption #4) you are paying some kind of premium for the tech.

You have to pay somebody, perhaps yourself, for hardware and software maintenance and support and that takes up the last 30%. This may seem high but you have to have lots of spares and separate labour pool to run it, so there is an interest cost, capital cost, and labour cost all separate from the capital pool used to buy the gadgets. Source: me and many years of data centre costing experience.

The implication is that a GPU+related CPU is only, based on my assumptions, about 40% of $12k or $4800. Sounds too cheap (**).

I'd like to meet their bankers. I have some ocean-front property in Arizona that needs refinancing....

(*) margin is defined as percent of revenue that exceeds your direct costs, so therefore omits all corporate stuff like CEO, management, finance costs, taxes, 2-inch carpeting, company cars, sales droids, marketing flunkies, senior corporate account executives, vice presidents of government influence, and all that. In the end you might have 20% profit at a corporate level. That also sounds too low.

(**) This is at the margins, where here marginal cost applies provided you are buying thousands of units. (And yes "margin" is a word that has too many overloaded meanings.)

...machines can 'learn' to adapt to new mediums...

And here I thought astrology and Tarot card readers were out of fashion....

LSI used to supply large drive arrays with fail-soft capabilities...

The user interface was some java-based app.

Whenever a (SATA) drive had some little hiccup, it was switched to "offline" mode.

To bring it back online, no kidding, you pressed the "Resurrect" button.

I wonder what their Buddhist customers thought?

people who invested their lives in the company...

Obv didn't read the prospectus.

What's old is new again

All those little boxes with fat, fat cables connecting them...

Reminds me of SGI Altix 350 (the smaller Altix with up to 16 individual dual-cpu nodes with a big switch in the middle.) They had two very fat cables for each node.

Fun fact: a node was called a "brick" in an unfortunate coincidence of terminology.

Bigger Altix like the 3700BX2 used the same cables to interconnect multi-CPU blades in a larger chassis, with multiple chassis per rack.

The 64-processor BX2 ran about 20kW in one rack. if I recall correctly. Somewhere I have a screenshot of vi occupying 13GB on a 128GB machine...

We spent more time on the phone with SGI Minnesota negotiating power and grounding needs than we did with the salesman to buy the thing.

Software supply chains

They said "beware of bugs and exploits in your software supply chains."

They forgot to say "beware of exploits of your software supplier businesses."

VMware is a single-point-of-failure for many businesses and organizations. It's only now that they're beginning to understand that.

(bad) language models are good for education....

When you have to fact-check the results of a LLM query, then you are now searching the Internet for relevant facts to support or refute the meanderings of the machine.

And, as a resut, you learn.

(Assuming you know you have to fact-check it to begin with. And can avoid the deeper rabbit-holes you are bound to find.)

I had to educate management on the Oracle software...

Complementary to Nate, we had very assertive yet wrong management who insisted, decades before ChatGPT, that they knew what was what according to Oracle.

We were delivering turnkey systems to customers and needed EE due to Spatial. Oracle 8 (and 9 and 10) included spatial, trusting that you would pay Oracle for EE prices if you used it. (The features and software quality of spatial was, however, something else altogether...)

I had several complaints to our management about (a) why Oracle (b) again, why Oracle, and (c) when do we move to Postgres plus PostGIS for the spatial features.

Turns out they had only recognized the name, never mind the cost. PostGIS got serious, afaik, around the time Oracle 9 came out so we gradually moved ourselves and our customers to that.

Oracle's "development" license was, in fact, only a testing license. Whenever we asked what that really meant, their interpretation was that development activities were "production" and therefore license fees, plus annual maintenance fees at 15% or 20% of list price, were due for factory integration and test systems. Eventually management noticed this and accelerated the switch.

This was all for deliverable systems. Corporate IT had their own shitshow with Oracle audits. Think VMware.....oh my!

Big fans

While I will resist the obvious joke that 1kW servers should enroll in onlyfans....

Call me when the airflow reaches Mach 1...

Endurance

Google has little to fear, except perhaps falling asleep during the investigation.

Or Pichai may retire first.

The Competition Bureau typically takes many years to investigate and prosecute frauds, anti-competitive actions or price fixing:

Canada bread price fixing (2001 through 2018): undetected until a whistleblower tipped them in 2015. Companies convicted and fines levied 2023.

"Business directories" fraud from 2012 through 2019. Charges laid 2021, conviction and fines 2023.

Price fixing and bid-rigging by firms shipping automobiles from Japan to Canada. First complaint in 2008, other complaints for 2011 and 2012. Conviction and fines levied 2023.

Problematics

As noted by others, there are a few issues with picking on Sandvine.

The Citizen Lab report was published in 2018, and mentioned Francisco Partners as a Sandvine investor. They still are. Francisco is based in San Francisco, California, USA. They are not, afaik, under US sanctions.

Just about any packet processor can pick out an ip address and redirect http(s) traffic to an alternative site. Or a range of IP addresses and alternative sites. The Citizen Lab report explains this in detail. (Much of the malware downloaded by Egyptian-in-the-middle redirections was because the downloads sites used http not https, leaving doors wide open for abuse.)

Competitors mentioned in their report include BlueCoat, bought up by Symantec which was in turn bought out by ... wait for it ... Broadcom.

Another competitor is NSO group which has attracted it's own legal issues, mostly from the US.

Sandvine's products appear to be relabeled whitebox computers. It's pretty hard to see how a US sanction is going to affect them.

The question becomes: is any high-performance packet processor vendor going to be subject to similar sanctions depending on how their customers, perhaps secretly, use their products? Excepting Cisco of course...

...and a large number of engines that would probably not be very happy eating moon dust

These are rocket engines, not jet engines.

But dust is indeed a problem. On landing the rocket engines will be blowing a significant amount of dust into lunar orbit.

You can count on the fingers of no hands how many people will like that.

I expect the next major private mission will be a dust-proof parking lot. Parking: $1300 per hour. Half price on weekends.

$3.8 billion purchase price on $1 billion revenue?

One, or both, of those numbers is very wrong.

The purchase to revenue ratio is normally in the 10 to 15 range, and 20 or higher only sometimes.

For example, IBM paid $34 billion for RedHat whose 2019 revenue was $3.4 billion.

Shades of SUN

In the early days, like the 1980s, Sun built servers that required 208VAC. Only the SF area supported that, the rest of the world was 220 or 240VAC. I think that's called parochial but correct me if I'm wrong.

When $WORK finally realized that, and put in a step-up transformer, the loud fan noise and irregular reboots stopped happening.

For Oxide to build a (customized) rack that is more than 2m high is....parochial. There are elevators that won't take such items even tilted and empty. The tallest Dell rack (48U) is 2273mm high. These guys have not done their homework. They haven't had to actually install such equipment in any variety of customer locations worldwide.

2.74m is just....dumb.

And putting just 15kW of CPU power into an oversized rack is....under whelming. There are higher power (and therefore faster) CPU and storage options that aren't physically unmanageable.

The Air Canada Slogan...

"We're not happy until you're not happy."

Faraday "cage" is (almost) all you need

The car thieves use retransmission gadgets to widen the range of a key fob inside a house. Wrapping it in a metallic container like a biscuit tin helps a lot.

But there are also the CAN-bus hackers that take off a headlight to access the car electronics and unlock the car and start the engine.

The statistics are grim. About half of 45,000 cars stolen in 2023 in Ontario and Quebec were never recovered. Only about 1700 were intercepted at the ports of Montreal and Halifax.

There is a CBC TV report where they tour the used-car lots of Accra, find a stolen car complete with Canadian registration, phone up the owner back in Canada and tell him what he left behind in the glove box. Funny and sad at the same time.

https://www.youtube.com/watch?v=R0u1OWYdJrs

Coprolitic

A new word every day.

weak password

Despite all the flim-flam about "password spraying" it was basically a successful guess of the test account and password spelling.

So, what was the password?

From stochastic parrot...

...to gullible intern on mushrooms.

The analogies just keep getting better.

...will allow customers to extract more value...

s/customers/Broadcom/

Banks and Money

Bank robber Willie Sutton, who, when asked by a reporter about why he stole from banks, answered: “Because that's where the money is.”

W11 video playback is screwed up

On my machine at least. The left third of the screen runs at a different "contrast" setting when full screen. But if I move the mouse it goes back to normal. For a second. But if I turn on subtitles the screen flickers with every new subtitle and stays normal for a second. If I don't run fullscreen it works-ish sort of.

ffplay is (still) flawless.

...surpassing a $1 billion run rate on an annualized basis...

No actual annual revenue was reported here, just an extrapolation. By the company, of course, not El Reg.

Doesn't anybody have any facts anymore?

Jesus Christ on a motorbike!

when is a channel not a channel?

When it is geographically bound.

I've been through that at $FORMERWORK when ESRI (geo-information software vendor) insisted very much on supporting their geography-based channel scheme. We system integrators had to buy their software in the destination country, often at inflated prices, and of course without support because, you know, the channel partners don't support outside their own country.

Or HP(E), which did the same on a large procurement (estimated $2M pa for several years). They lost that one.

I could name some others but IBM is too hard to spell.

If Broadcom fails to support their multi-/inter-/cross-national markets (VARs and system integrators and multinational corps) then they will lose marketshare bigtime and with Musk-like RUD.

Yes but no

(There are some interesting comments on HN and rebuttals by Liam. But here is my take.)

usually completely ignoring all the lessons learned in the previous generation

Umm, not really. RSX-11 on the PDP-11 was a decidedly calculated subset of OS/360 on the competition's machine. It had file IO added (and the whole almost-flat filesystem) but the rest is fairly simple process management etc. IBM's designs were good enough to copy. And added to the 16-bit cheap-ish CPU that was a follow-on to numerous other designs by a very experienced software and hardware design crew at DEC.

The notion that lessons were not learned I think is incorrect. The lessons were very much learned: Choose the best two-thirds of your competition's product, add your own third. Sell and sell hard. (And, as always, beating IBM on price is child's play.)

The Symbolics stuff was legendary (at the time) for complexity and cost. And uselessness -- were there any actual software products sold requiring a Symbolics machine? Nobody buys $100k machines to heat up a room with the wasted power. You have an expensive user because they have skills (orbital calculations, microprocessor design, chemical plant optimization) so you buy them an expensive tool. Who bought a Symbolics workstation for an end-user? Anybody? Bueller?

Another example: the IBM 360 series had the optional model 2250 display. It was early days and used a light-pen to detect the flash on the CRT but it sold well because there was design software to go with it. That was the beginning of the workstation era -- build or design things quicker with interactive computing. Only affordable by mega-corps but the market was then "proven". And competition quickly appeared.

I don't claim the notion that a skilled thinker could manage to produce useful products with LISP-like languages is wrong. Arguably the equivalent modern example is the construction of WhatsApp by a handful of people using Erlang. Now that is an exotic language.

These are cases where the product filled needs of the market. Sometimes the market didn't even know it needed them.

Arguing over New Jersey/Stanford or MIT "approaches" is irrelevant. Designers do what designers do and adding post-facto explanations by technology historians is not adding anything useful.

There were only these two schools of thought you say? Oh.