You can add "perpetually delayed 10nm" and "Kafka-esque badge engineering" to the long list of intel blunders. Apparently things are so bad now that Anandtech is calling out their press team for announcing products with zero information.

https://www.anandtech.com/show/12878/intel-discuss-whiskey-l...

This is what a decade of dominance does to a company. Success kills.

This is good. Old wood should die off, so new growth can replace it.

Arrogance is the cause, not success itself.

Complacency kills.

They'll do like Facebook and Microsoft, apologize, make new good deeds, send their PR team, then everybody will love them again.

It's not a problem anymore. The majority of the public never hold any company accountable on the long run. They now know that, and they adjusted accordingly.

s/company/politician

s/company/celebrity

Facebook never made any good deeds.

Microsoft, as much as I hate them, has actually made a lot of of them.

And in general, actions speaks louder then words. And people do remember what they have done. Forgiving was never going to be a easy thing, especially when you audience are developers.

The Network Effect makes people stick to the company/product. Opposition/alternative can snowball slowly and surely but needs a catalysator/enabler. This can be an event/deed such as eternal rain causing floods on Earth or poisoning of the aquaduct.

Companies and countries and personae know this. Its why ruling with authority happens. Its why those in power in Arabia feared the domino effect of the Arabic Spring. It can even be a reason why people swear.

Trying to avoid it (the domino effect) from happening is called damage control.

Can't edit my post anymore but PG's essay "What You Can't Say" also touches on the subject of the power struggle.

[1] http://paulgraham.com/say.html

Sub ambient cooling with all the insulating material to prevent condensation. The unit itself is like 1k7W.

However, the best part is that now Intel cannot argue that overclocking is unintended as they bloody set 2.7Ghz chip to run at 5Ghz, albeit for like 5seconds. (I doubt the cooling would even survive a minute)

Just a small clarification,

The chiller is rated at

1.7kW thermal while consuming

1kW electrical

That's about 1/2 tonne of AC. Insane ...

(and before anyone complains, yes, thermal pumps can have greater than 100% "efficiency" - that is: heat transported per energy spent)

What's 1k7W?

Shorthand for 1.7kW or 1,700 Watts. The shorthand is common in some fields, I've mostly seen it in electronics where you might have 4k7 to represent 4.7 kOhm or 4,700 Ohms.

This is a new one to this old dog. 2k18 is 2018. 1k7 looks to me like 1007. Much shorter to use 1.7kW which most know how to parse to 1.7 x 10³ Watts.

I can vouch for the fact that 1k7=1700 is a common convention. It comes from the face that . can be hard to read sometimes. You often also see something like an “0R5” resistor which is 500 milliohms.

More common in EE contexts than anywhere else.

Sorry to break the news, but 2k18 is 2180. You're a few years off.

It's not. "2k" is pronounced "two thousand". "2k18" is "two thousand eighteen".

For example: https://en.wikipedia.org/wiki/NHL_2K

You, but in the case where the shorthand is required, the suffix (W in this case) is also dropped, or at least has been every time I've seen or used it.

Never saw that once in EE.

It's a fairly common notation on printed circuit board resistors where a dot could get lost. Better readability. They usually drop the suffix though, not sure if it's common for non-resistance items to use, but I got what he was trying to say.

A similar shorthand exists for calendar years - e.g. "2k8" means "2008".

which is clearly different to 2k8 meaning 2800. (or 1k7 meaning either 1700 or 1007...)

It's mainly common in EE schematics because 1k7W doesn't have any weird characters like '.' that may not parse correctly in CAD software.

1,700w CPU cooler

>"the meltdown/spectre "working as intended" joke"

I was really surprised by this. Why would you treat your customers as if they are idiots like this and make them lose trust in you? Its not like the main people making purchasing decisions are "the public" either, they should be much more informed than the typical person buying a laptop:

>"Intel now has eight major customers that are responsible for 75% of its revenues. In 2016, Intel's three largest customers were responsible for 31% of the firm's accounts receivable. Intel might obey Moore’s Law, but the Pareto Principle (a.k.a the 80/20 rule) is a different story." https://www.investopedia.com/articles/markets/100214/inside-...

Wasn't their public reaction as it was due to liability concerns? i.e. if they admitted that something was wrong with their chips, they may have gotten into significant legal / financial trouble?

I didn't think of that, interesting point. Stuck between a rock and a hard place...

> This really went wrong for intel

We do really miss you, Andy Grove. I can bet he wouldn't have let this insanity happen.

Not only that, but the CPU that Intel used was most likely an Xeon 8176, which retails for $8000 US. I'm thinking that the Threadripper with 32 cores will cost less than $2000 US.

For reference, the Epyc 7551P (32C, 2.55 GHz turbo) is $2100. The new Threadripper should be in the same ballpark.

It's a PR hit but certainly not a disaster. Everybody and their grandmother knew that a chip demoed to run at 5GHz is an overclocked hand picked engineering sample, Intel has been pulling the supercooled CPU stunt for decades.

At 250W thermal, the 4 die, 7nm Threadripper AMD announced will have similar cooling issues, a dead give away was that Lisa Su completely omitted Cinebench results - despite those being prominently featured before.

Bottom line is that AMD and Intel have roughly similar products and can easily cover for any performance differences using the large financial margins high end CPUs enjoy. The game then becomes a marketing one, who can keep more of that margin by convincing it's fan base they have decisively won the match.

At 250W you won’t see anywhere near the thermal output of that 2.7->5Ghz overclock on 28c/56T. There have been gpus with higher thermal output than 250W of the Amd’s cooled on air.

That CPU draws over 250W when running stock.

They had a separate 20+ phase power delivery board that could pump up to 2000W into that cpu...

Nonsense. We know the new Threadripper was running on air cooling. We know whatever Intel showed needed more than one horsepower of phase change cooling. They're both monstrous chips, but not even in the same ballpark.

250W << 2 horsepower

Before calling what you read 'nonsense', please follow the argument. We have a massively overclocked and unrealistically chilled chip from Intel, with massive results. And we have a cool chip sitting in Lisa Su's hand, that claims no performance numbers. Both use comparable technology, have comparable number of cores and per core performance and, we expect, comparable real life frequencies and dissipation when air cooled.

My conclusion is that the chips are roughly comparable in performance and we are witnessing a marketing war, not a performance showdown. Nothing you said seems to disprove that or even address that conclusion.

> This really went wrong for intel.

Why? Anyone with a brain knew this was carefully prepared and staged, and a nice muscle-flexing publicity stunt before AMD shows their new 32-core Threadripper.

I get it, it's slow season so journalists have to find news where there aren't any. Regardless of how did they achieve it, running 5GHz on all 28 cores is pretty neat! I am wondering if some 3rd party tries to push new TR to see how far will they go, so that a few "rocket scientists" among us can build similarly overclocked rigs.