dfrage

> Intel and others are years behind

Not clear anyone other than Intel is trying, but Intel's situation is much more complicated. They tried for a purely optical for its life "10nm" node more aggressive than TSMC's initial "7nm" node and catastrophically failed. There's clearly some top level management issues there, a big problem for Intel for decades, and it's extremely worrying they let them destroy a generation of their crown jewel.

But that doesn't tell us much about their "7nm" node very roughly equivalent to Samsung and TSMC's "5nm" nodes, except those companies have a lot more real world EUV experience, but it's not always good to be a pioneer. Intel could conceivably get back in the game in about the same time frame as these two companies exit "5nm" risk production, we just don't know. All we know is that they're buying EUV machines and installing them in multiple fabs.

And there wouldn't be enough smart people.

These lithography machines are completely wild: the part that fails the most spits out a tiny droplet of molten tin. That's hit with a laser, and the excited electrons emit what's called extreme ultraviolet (EUV) which is very close to the softest of X-rays. Most of the power released is wasted bouncing off a series of mirrors, and then wild stuff happens as these energetic photons hit the semi-optional mask cover (pellicle), mask, and then chip: https://en.wikipedia.org/wiki/EUV_lithography

Per Wikipedia, "the name was a condensation of "Can" from Canada and "ola" from other vegetable oils like Mazola".

> My original point is that we should not be crippling something very useful (javascript in the browser) because of flawed architectures that mostly affect one company in a way that decimates performance.

Lots of us have a different opinion on the usefulness vs. risk of running random and often deliberately hostile JavaScript in your browser, see the the popularity of NoScript, and how many of us use uMatrix with JavaScript turned off by default. Most of the time I follow a link where I don't see anything, I just delete the tab, most of those pages aren't likely worth it.

"Mostly affect one company" is something completely unproven, since AMD is not getting subjected to the same degree of scrutiny, AMD has a minuscule and for some reason declining in 19Q1 market share for servers, while desktop and laptops are modest but showing healthy market share growth: https://www.extremetech.com/computing/291032-amd-gains-marke...

While ARM is announcing architecture specific flaws beyond basic Spectre: Meltdown (CVE-2017-5754) and Rogue System Register Read (CVE-2018-3640, in the Spectre-NG batch but by definition system specific): https://developer.arm.com/support/arm-security-updates/specu...

Hits forehead for forgetting to check SemiWiki. But note that since then Global Foundries has abandoned for now offering this general node, 7nm as they name it, 10nm as Intel does.

But isn't the whole die exposed and otherwise processed as a whole piece? Very much not deeply educated here and can't justify the investment to change that, my primary mental models for defects are either something that takes out a whole die, like one lithography step being misaligned, or spot damage like a piece of dust.

But there are clearly issues in between that are statistical inside a die, I recall Semiaccurate saying one of Nvidia or AMD did a GPU tape out to a TSMC process where they duplicated vias because that process' were iffy, and they compensated with a less dense design where either one or two working was OK. If Intel is suffering that sort of problem, then the GPU is a big part of the die that can be fused out while you still have something useful. If all your CPUs or all your L3 cache banks fail, a working GPU is pointless.

That article points out two particularly suspect things Intel is uniquely trying at this node: SAQP for the metal layers, which I've seen cited before, and which they generically officially blame, and cobalt in interconnects. And at least one other thing was mentioned as suspect, and four new things total.

One ray of hope is mentioned for Intel, in that they were the most aggressive in the industry with their 14nm and 10nm nodes, and in both cases paid the price in yields, while they're being conservative for their 7nm node, no doubt because EUV is a very big step for everyone. Semiaccurate also commented and/or theorized that a compelling reason Intel is continuing to work on their 10nm at one fab is that one or more things in it are also going to be used in their 7nm, so they might as well debug them now and there, and sell some chips while they're at it.

Now to do some catching up on SemiWiki, thanks!

> The "don't apply the updates" argument in security is basically philosophically comparable to the antivax movement

As long as you also apply the quarantine principle when relevant. And herd immunity, both applied backwards:

Quarantine for when you run code that's under your control, like an HPC cluster; the biggest dangers from these vulnerabilities are when you're running completely random code supplied by others, JavaScript in browsers for normal users, and the multi-tenant systems of cloud providers.

Herd size matters because the bigger the herd, the larger the number of potentially vulnerable systems. IBM mainframe chips are vulnerable to Meltdown and Spectre, but there aren't hardly as many of them as x86 systems. Although the payoff of compromising them is likely to be disproportionately high.

It's a good analogy as long as Intel was the first to experience the end of Dennard scaling (https://en.wikipedia.org/wiki/Dennard_scaling) because their fab lines were ahead of the rest of the industry's. And fabs are all "semis", due to the massive amounts of capital and talent needed to move to the next node.

So much so that we're now we're down to two companies in the whole world who are successfully executing the smallest CPU nodes, unless Intel manages to make their "10nm" work, or pulls off their "7nm".

While we're hearing the very roughly equivalent TSMC "5nm" node is starting risk production (https://wccftech.com/tsmc-5nm-production-euv/ beta testing, you might say, someone outside of TSMC has to be the first, second, etc. to try to get real world dies that work on a new node). Intel isn't saying anything, but Semiaccurate has reported at least two fab lines that were slated to move to their 10nm are installing lots of EUV equipment consistent with using them for their 7nm node (and at least one fab moving back to 14nm).

KBR = Kaby Lake Refresh. Interesting, in that Intel's "10nm" node is said to be more power efficient than their "14nm" node, in this case, per AnandTech (https://www.anandtech.com/show/11738/intel-launches-8th-gene...) KBR was launched with 14+nm. Could be that early of a 10nm part wasn't yet very power optimized. And that would be a very good explanation for why the GPU was disabled vs. the yield issue, which we're pretty sure is much more fundamental than specs of dust and other isolated things that can disable a part of a die without killing it altogether.

That's interesting, and relevant in that the NexGen team was joined by DEC Alpha people to do the K7, but the K6 is not AMD's first out-of-order speculative execution x86 design, per Wikipedia the K5 was an internal project that "was based upon an internal highly parallel 29k RISC processor architecture with an x86 decoding front-end."

Way inside Tueller Drill range (https://en.wikipedia.org/wiki/Tueller_Drill), if the grizzly disabled her in the first blow her situation was indeed impossible. If not, and that's reported by some survivors, depending on what was hit and how a handgun was holstered, a handgun is a lot more ergonomic than a spray can, for example in pointing, it would have given her a better chance to strike a counter blow.

The current set of vulnerabilities target Intel specific microarchitectural features, like the first version of Foreshadow/L1TF targeted the SGX enclave, which by definition isn't in other company's CPUs.

Given that AMD is fully vulnerable to Spectre, there's absolutely no reason to believe it isn't similarly vulnerable to microarchitectural detail leakage if people were to look. And going back to what I was replying to:

> we are talking about one company that made egregious engineering decisions to maintain their market leadership which has put their customers at risk

We demonstrably aren't, seeing as how ARM, and IBM, both POWER and mainframe/Z CPUs are also vulnerable to Meltdown. That and the significant prevalence of Spectre says this is not "typical Intel" but "typical industry", a blind spot confirmed in 7 different companies, and 8 teams to the extent the IBM lines are done by different people.

The "Intel is uniquely evil" trope simply doesn't hold water.

I've heard entirely the opposite, that having a close relationship between chip designers and fabricators allows for higher performance designs. I don't know of anyone who interpreted AMD selling off its foundries as anything other than severe financial distress, and it worked supremely well for Intel while they stayed at least one step ahead of the competition. Enough so this is said to have wiped out a generation of CPU architects while Dennard scaling still worked, no matter how clever they were, Intel moving to its next process node wiped out their speed advantage.

But it's a brittle model, if a company screws up a node and is too messed up to handle the failure gracefully, as Intel is doing with their "10nm", no doubt with pride as a factor. And it's not uncommon for institutions to permanently lose abilities, I'm not counting on Intel succeeding with their "7nm" node.

On the third hand, we're now down to 2-3 high end CPU fab companies, Samsung, TSMC, and maybe Intel. That also can be a brittle thing.

These vulnerabilities can jump process address space boundaries. It's a lot harder but can be done, look at the original Spectre paper: https://spectreattack.com/spectre.pdf

Not tntn, but I've been following this and it seems to be both Intel's now decades long history of very bad high level engineering and personal management catching up with their crown jewel, and being more aggressive than TSMC's initial 7nm node. Perhaps Intel depending on a particular lithography? technique that TSMC isn't, or isn't yet heavily, but we don't really know, no one authoritative is talking, and Intel is still claiming 10nm is going to make it.

Amplifying the other current replies, what you're bemoaning and what slagged the Netburst "marchitecture" is the end of MOSFET Dennard Scaling: https://en.wikipedia.org/wiki/Dennard_scaling

Moore's Law is "the number of lowest cost transistors doubles at X interval", and 193nm UV immersion lithography limits have been hitting it hard lately (see https://en.wikipedia.org/wiki/Multiple_patterning). But chip manufacturing equipment makers haven't run out of tricks quite yet.

Sounds like it. None of the brand names you mention were known in my part of the US at the time, and I don't even recognize any of them today except for Gundig.

For TVs, Japan very conveniently for the US shares the same NSTC standard, their only issue is that half of Honshuu, the big island, is on 50 cycle power, the other on 60 cycles (both at 110-120V for consumers like the US with our 60 cycle power). So with their two biggest markets being satisfied with the same gear, I can well imagine the fragmented European market suffering.

Note also SECAM for France, countries it influenced, and apparently Eastern Europe per Wikipedia either initially so its subjects under Soviet domination couldn't receive PAL signals from across the Iron Curtain, or because it's insensitive to amplitude and phase variations from long cable or microwave links.

What about radios? While Sony pioneered the small and inexpensive lo-fi radio with their own transistors, this was very big in the 1950s through the 1960s, but by the time I was paying attention in the 1970s no Japanese company had a commanding lead in radios in the US, they were pretty much a low end commodity by then, except for shortwave sets, which weren't big in the US by then.

The first Walkman was released in Japan in 1979, the brand and general format was very big in the US by the early 1980s: https://en.wikipedia.org/wiki/Walkman

Sony and Phillips also closely collaborated on the Compact Disc, its capacity was supposedly dictated by one Sony guy's insistence that Beethoven's Ninth Symphony as conducted by Herbert von Karajan would fit on one (you have to split a movement between sides for a single vinyl LP release, that was very annoying).

They also scored a big early hit two years later in late 1984 with a very small CD player, the head of the project estimated how small was possible with current technology, made a block of wood that size, and challenged his engineers to make one no bigger. Ah, the "Discman", it was per Wikipedia and my memory very important in the early adoption of CDs: https://en.wikipedia.org/wiki/Discman

> Also, to keep this all in perspective, we are talking about seven companies that made egregious engineering decisions to maintain their market leadership which has put their customers at risk.

Fixed it for you, the current list I have is AMD, ARM, Freescale/NXP, IBM both POWER and mainframe/Z, Intel, MIPS, Oracle SPARC, and maybe old Fujitsu/HAL SPARC designs for Spectre, with at least four of those CPU lines also subject to Meltdown.

If it had been practical to keep with the PPC line Mac users today would have the same problems, seeing as how POWER CPUs suffer from both Meltdown and Spectre: https://en.wikipedia.org/wiki/Meltdown_(security_vulnerabili... This theoretical analysis says G4 and up ones look worryingly vulnerable to Spectre: http://tenfourfox.blogspot.com/2018/01/more-about-spectre-an... Current Freescale/NXP PowerPC designs are vulnerable to Spectre: https://lkml.org/lkml/2018/12/27/98

I've read that the RS-25 level of "reuse" was highly qualified, they required a significant rebuild between flights because they pushed the margins so hard. Wikipedia says "refurbished" https://en.wikipedia.org/wiki/Space_Shuttle_main_engine

But you're generally right about the insanity of the Senate Launch System.

I absolutely don't know anyone who was saying that about Japanese electronics in the 1970s in my part of US flyover country. Sony was considered to be one of the very best in the world, and my family was greatly disappointed in the 1980s when Sony moved a lot of manufacturing to Mexico and they had to fix many cold solder joints on TV sets. If you wanted a good reel to reel tape deck, Teac was your first choice.

One thing left out of this formulation is that it was realized at some point, certainly by the 1970s for electronics, that the Japanese added "and improve", which no one says about the Chinese. Who are said to have a pattern more than a century old of steadily decreasing quality until the customer complains. Different culture, vastly different societal trust levels, Japan high, China low, different results.

Your theory proved correct in the 1970s, things got so messed up "everything" rose, both unemployment and inflation driving a stake into the heart of the Phillips Curve, and what people demanded to buy private or Federal debt. The latter got so bad "Carter bonds" were created, denominated in West German Marks or Swiss Francs: https://en.wikipedia.org/wiki/Carter_bonds

Killing the bear isn't the objective, stopping it from killing you is,the same goes for legitimate self-defense against humans. And it turns out that's frequently possible, even with human self-defense handgun calibers. You might discourage the bear with a wound, pain is a great motivator and works half the time with humans, but for bears disrupting a shoulder for example works well. Here's one study of 63 cases with 95% success rate: https://www.ammoland.com/2019/03/pistols-or-handguns-used-to...

And in serious bear country, serious people carry serious long guns, a 12 gauge shotgun with slugs, or a centerfire rifle with a significant payload. And it's easier to shoot those well under stress.

In the most recent discussion of this it was reported that is no longer an easy option with Firefox: https://news.ycombinator.com/item?id=20012750 There's also a general impression that would be too slow for many web pages, and that comment noted it took minutes for logins using JavaScript cryptography to execute.

The other area they can't complete is on variable or disaster recovery scenarios, the sort where you'll reserve CPU instances, but only spin them up for your heavy load parts of the day or when you have an unexpected migration. And when you're not using them, AWS puts them on their spot market.

They're probably a bad match for when you have a single batch of work to do, like when the New York Times converted their archives, or every time Netflix has to reencode their video library.

Red Hat released info about z/OS mainframes, they're also vulnerable to Meltdown as well as Spectre. ARM also has one new design that's vulnerable to Meltdown, and everyone has Spectre vulnerable designs.

Oops, didn't read your comment closely enough, strike my above comment.

Followed the Wikipedia link to the 1995 Fujitsu/HAL out-of-order design (https://en.wikipedia.org/wiki/HAL_SPARC64) and it says it's superscalar, but then that the first version can execute as many as 4 instructions at once and out-of-order. And from the very first version had branch prediction, which equals speculative execution as far as I know.

I'm saying Intel's design decisions are found in so many other vendors, Spectre in all of them I've looked for, that it represents an industry blind spot. Which calls for different kinds of "accountability", unless you're a rent seeker.

What's expensive, and some time ago became a significant limiting factor in densely packing machine rooms, is removing heat from the room. As long as you can supply cool air to the racks they can take care of removing heat from motherboards and other components.

If cooling systems for machine rooms haven't caught up with modern systems that dissipate more heat, the direction you're thinking has merit. But I doubt there are serious limiting factors here, besides cost.

I was just discussing this with a friend. Obviously there's a certain degree of multi-tenancy sharing that's most cost effective. For electrical power use and removing heat there's a floor where, if not being used, the best you could do would be to turn a system off, which has its own issues. Go too far and you won't be able to also run a spot market in CPU instances like AWS does.

Or did no one bother to check the older designs??