Intel's 10nm Problems Might Be Worse Than First Thought

[Hellion]

2 hours ago, cj09beira said:

the people doing those things aren't the same ones so don't worry they are working on it, its hard to fix these kinds of problems 

If that's the case then why has intel continued to release a new generation without the issues being fixed? The flaws were known for over a year before coffee lake was released are still prevalent.

 

What's even worse is that consumers are still buying this garbage knowing in advance that these numerous security risks haven't been taken care of.

[Taf the Ghost]

25 minutes ago, cj09beira said:

should i mark this as informative or funny

hm.... decisions decisions 

Rain Forests. Finally remembered the late 80s/early 90s CEO Ego-boosting play.

 

Proper Environment Management is extremely important. That most of the issues revolved around products shipped to the wealthy in Western countries is conveniently missed in their pleas to help Save the Rainforests.

[Hellion]

3 hours ago, Zodiark1593 said:

Intel aside, Arm's Cortex A75 and (heavily implied by Apple themselves) Apple's own CPU cores both exhibit the "Meltdown" security flaw. However, Cortex A75 had not yet been released at the time the flaws were made public, and Apple can easily patch their iOS devices with a minimum of attention.

Well aren't apple champions?

 

Maybe that's what all that additional unnecessary cost is for when you buy one of their products before they decide it's outside their scope of interest and no longer want to support it regardless if they are at fault or not.

 

That doesn't change the fact that there's still a performance hit and the user is no longer receiving the same level they were expecting when originally purchasing the device. 

[Zodiark1593]

9 minutes ago, Hellion said:

That doesn't change the fact that there's still a performance hit and the user is no longer receiving the same level they were expecting when originally purchasing the device. 

While x86 PCs allow for measurement of performance differences due to the patches, in what way can you measure the differences in either Apple's devices or Cortex A75-based devices? Further, in the workloads common of mobile devices, would the performance impacts be of any relevance? Last I checked, we aren't running server workloads that are constantly hammering the storage on our devices, which btw is where most of the performance deficits occur.

[Hellion]

5 minutes ago, Zodiark1593 said:

While x86 PCs allow for measurement of performance differences due to the patches, in what way can you measure the differences in either Apple's devices or Cortex A75-based devices? Further, in the workloads common of mobile devices, would the performance impacts be of any relevance? Last I checked, we aren't running server workloads that are constantly hammering the storage on our devices, which btw is where most of the performance deficits occur.

If I purchase a car with 200 horse power and take it back to the dealer or a repair facility for a recall and it is returned to me with 160 horse power while being told "it's not a big deal, you won't notice a difference" regardless whether or not that's the case,  it is still morally wrong. I paid for 200 horse power and expect that level to be maintained throughout my ownership if regular maintainance is performed on the vehicle.

[pipnina]

1 minute ago, Hellion said:

If I purchase a car with 200 horse power and take it back to the dealer or a repair facility for a recall and it is returned to me with 160 horse power while being told "it's not a big deal, you won't notice a difference" regardless whether or not that's the case,  it is still morally wrong. I paid for 200 horse power and expect that level to be maintained throughout my ownership if regular maintainance is performed on the vehicle.

I agree, though if they gave it back with 160hp and said "There is a known fault with this model where the engine's fuels injection system forces the car to full throttle, the only way to resolve involves reducing the power output" and then fixed it in the next model of car and gave you a certain % of the original car's cost back in cash it would not be so bad. Still annoying, but at least you know they have done something to try and make amends...

[Daegun]

5 minutes ago, Hellion said:

If I purchase a car with 200 horse power and take it back to the dealer or a repair facility for a recall and it is returned to me with 160 horse power while being told "it's not a big deal, you won't notice a difference" regardless whether or not that's the case,  it is still morally wrong. I paid for 200 horse power and expect that level to be maintained throughout my ownership if regular maintainance is performed on the vehicle.

A more proper analogy keeping with the car theme would be this:

 

Every company that makes engines has a problem with their engines and they fix them, but the fix causes a bit of a problem. Your engine still has the same horsepower as before, but when turning left you can't go over 160mph. Not only do you or any average person not notice it, you aren't even affected by it. The only ones affected by it are Nascar drivers and, indirectly, people that watch Nascar. 

[Hellion]

11 minutes ago, pipnina said:

I agree, though if they gave it back with 160hp and said "There is a known fault with this model where the engine's fuels injection system forces the car to full throttle, the only way to resolve involves reducing the power output" and then fixed it in the next model of car and gave you a certain % of the original car's cost back in cash it would not be so bad. Still annoying, but at least you know they have done something to try and make amends...

That or even more fairly upgrade the vehicle at their expense to bring it back to the factory spec it was originally purchased with. A purchase should be a contract. Unfortunately that's not always the case and that's why consumer protection law exists. Sadly most tech issues are over these government regulators heads in regards to understanding so often nothing is done and the consumer gets screwed.

[Unixsystem]

This is fantastic news.

 

Now, in addition to the know-it-alls in every node thread going "actually, Intel's 10nm node is on par with everyone else's 7nm," we also get to enjoy a new group replying with "actually, Intel's 10nm is now closer to 12nm."

[Zodiark1593]

28 minutes ago, Hellion said:

If I purchase a car with 200 horse power and take it back to the dealer or a repair facility for a recall and it is returned to me with 160 horse power while being told "it's not a big deal, you won't notice a difference" regardless whether or not that's the case,  it is still morally wrong. I paid for 200 horse power and expect that level to be maintained throughout my ownership if regular maintainance is performed on the vehicle.

Remind me again where Apple specifies or advertises any comparable metric as to the performance of their device.

[Hellion]

14 minutes ago, Zodiark1593 said:

Remind me again where Apple specifies or advertises any comparable metric as to the performance of their device.

An i7 processor is still an i7 processor. A vega 64 is still a vega 64. Just because it has an apple logo on it doesn't mean it's not still a computer with the same parts as others and shouldn't perform on par within reason of manufacturing tolerance.

 

This goes back to the argument of what exactly a "pro" device is. There's an expectation there.

 

I know most apple tools are brainwashed by marketing to believe anything apple does deserves endless worship but for those without their head in the sand, any piece of technology has a spec sheet which will give a baseline on what performance level is expected.

[mr moose]

23 minutes ago, Waffles13 said:

This is fantastic news.

 

Now, in addition to the know-it-alls in every node thread going "actually, Intel's 10nm node is on par with everyone else's 7nm," we also get to enjoy a new group replying with "actually, Intel's 10nm is now closer to 12nm."

Not sure what kind of consumer enjoys seeing tech advances slow down.  It only serves to cost you more, you should be wanting Intel to hit superior densities on smaller nodes not happy that they can't.

[CommandMan7]

I was talking about this on the AdoredTV Discord server yesterday, and this is what I found:

 

The Intel "Old 10nm" was 54nm * 36nm * 399nm per transistor. However, new leaks say it's 15% behind smallest 7nm (Samsung), which is 44nm x 36nm x 270nm. That puts Intel "New 10nm" at about 62.1nm x 50.6nm x 310.5nm. For reference, the GF 12LP process, which AMD is shipping processors on today, is 78nm x 64nm x ~400nm. Intel "New 10nm" single transistor is 0.0031 µm², whereas GF 12LP is 0.0049 µm². This is a difference of 56%, but doesn't lead to a density improvement of 56%. The GF 12LP area is so high due to it being mostly identical to the GF 14LPP process, except each cell is 400nm tall instead of 480nm tall on GF 14LPP. The extra 80nm off the top allows for better cooling, easier routing, etc.

 

In addition to that past info, just today a really good article on samsung 7nm came out, plus I found some more resources.

If anyone's interested: https://fuse.wikichip.org/news/1479/vlsi-2018-samsungs-2nd-gen-7nm-euv-goes-hvm/

[DeScruff]

10 hours ago, Hellion said:

How about intel concern themselves with fixing the countless variants of security flaws in their products first?

 

It makes no difference to shrink the node size when the product itself essentially takes a 15%+ performance hit after bios and OS patches are implemented to fix their engineering neglegence because they are too busy playing the fat cat role pushing marginal 5%~ performance increases each generation.

 

What a joke intel has become. 

 

 

For the record I have 3 machines with Intel CPU's,  so in before the "fanboy" accusations start to fly.

The problem is that CPUs have a long development time. A new architecture you hear about now, probably started development 3-4 years prior.
I can't find an exact year when Spectre as we know it was discovered, only that is was made public in 2018.

Something that is 100% immune to any variation on Spectre is probably in development, but thats likely half a decade away. Everything else in between is probably just gonna be patchwork on stuff that has long since been developed, and replacing how speculative execution works with them is flat out impossible, or would be too time consuming (leaving Intel still selling vulnerable CPUs for an extended time.)

As much as I personally hate it one thing you will hear from programmers a lot, and it remains true with hardware design: Developer Time > CPU time.
You could probably optimize something to run 2% faster, but the amount of time it take to do that is not a good trade off.
 

20 hours ago, Ben Quigley said:

AFAIK IBM has managed to make a 5nm chip already. The current belief is that 3nm is the theoretical limit now, if you can somehow overcome quantum effects. There are already a number of players started to test 3nm production technologies. 

Huh! Granted there is a difference between a theoretical limit, a set of prototype chips, and something that is feasible for mass manufacture. but this is quite interesting.
Time to see who gets to the finish line first: Feasible 3nm, or Feasible other materiel.

[cj09beira]

11 hours ago, Hellion said:

If that's the case then why has intel continued to release a new generation without the issues being fixed? The flaws were known for over a year before coffee lake was released are still prevalent.

 

What's even worse is that consumers are still buying this garbage knowing in advance that these numerous security risks haven't been taken care of.

it takes longer than a year to take a final design to market so coffelake couldn't be fixed, same thing with zen+, we dont even know what type of fix zen 2 will have, 

[Stefan Payne]

11 hours ago, Sypran said:

The problem is that CPUs have a long development time. A new architecture you hear about now, probably started development 3-4 years prior.
I can't find an exact year when Spectre as we know it was discovered, only that is was made public in 2018.

Its more like 5 Years+.

So Development of Ryzen began when Bulldozer just came out.

 

As for Spectre/Meltdown: It was roughly 6 Months prior. The one who found the issue gave all major Partys 6 Months time for that.

 

11 hours ago, Sypran said:

Something that is 100% immune to any variation on Spectre is probably in development, but thats likely half a decade away.

No, that is imposible!

The only CPUs that are immune to Spectre are the EPIC based ones -> Itanium, the XBox 360 CPU and a couple of Atom CPUs. Because they are in order without speculative excecution.

 

So to be immune to Spectre/Meltdown you have to remove speculative excecution.

 

 

5 hours ago, cj09beira said:

it takes longer than a year to take a final design to market so coffelake couldn't be fixed, same thing with zen+, we dont even know what type of fix zen 2 will have, 

According to one of the Guys who found Spectre/Melttdown Intel knew for about 4 Years that there was something going on because they were very interested on one of the patches that they were developing.

That was the thing that got the Spectre/Meltdown thingy rolling in the first place!

Because the guy who did that, was very sceptical and looked further into this issue, wich caused him to find the Spectre/Meltdown issue.

 

Sadly can't find the (German) Interview where one of the guys finding this issues talked about it right now

[DeScruff]

23 minutes ago, Stefan Payne said:

Its more like 5 Years+.

So Development of Ryzen began when Bulldozer just came out.

 

As for Spectre/Meltdown: It was roughly 6 Months prior. The one who found the issue gave all major Partys 6 Months time for that.

I mean moreso the announcement of said archetecture, then release dates.

I couldn't find any exact year dates in a quick search other then Zen starting design in 2011 or 12 i forget which, and AMD announcing Zen in 2015. Which is... 3-4 years. But yeah I don't doubt it can be much longer

 

[Zodiark1593]

33 minutes ago, Stefan Payne said:

Its more like 5 Years+.

So Development of Ryzen began when Bulldozer just came out.

 

As for Spectre/Meltdown: It was roughly 6 Months prior. The one who found the issue gave all major Partys 6 Months time for that.

 

No, that is imposible!

The only CPUs that are immune to Spectre are the EPIC based ones -> Itanium, the XBox 360 CPU and a couple of Atom CPUs. Because they are in order without speculative excecution.

 

So to be immune to Spectre/Meltdown you have to remove speculative excecution.

 

 

According to one of the Guys who found Spectre/Melttdown Intel knew for about 4 Years that there was something going on because they were very interested on one of the patches that they were developing.

That was the thing that got the Spectre/Meltdown thingy rolling in the first place!

Because the guy who did that, was very sceptical and looked further into this issue, wich caused him to find the Spectre/Meltdown issue.

 

Sadly can't find the (German) Interview where one of the guys finding this issues talked about it right now

Cortex A53 and probably A55 are both in-order cores and are immune to Spectre as well. There was an article on the Raspberry Pi website regarding Spectre, actually.

 

If Graphene processors (the ones that can supposedly hit tens of GHz) ever become a thing, this could pose an opportunity to do away with any vulnerabilities regarding Speculative Execution altogether without stepping backwards in absolute performance. Though this would rely on CPU designers not starting up a performance arms race.

[Unixsystem]

17 hours ago, mr moose said:

Not sure what kind of consumer enjoys seeing tech advances slow down.  It only serves to cost you more, you should be wanting Intel to hit superior densities on smaller nodes not happy that they can't.

I'm not making any qualitative comment on the tech itself. But if you've spent a lot of time here or on other tech forums, every thread about this topic for the last 6-9 months has that one guy who needs to inform everyone that node naming conventions aren't 1:1 comparable, even though at this point at that should be relatively common knowledge at this point. Now if it turns out that Intel 10nm is still denser per nm but not quite as dense overall as other competing 7nm processes, it's going to add another fun layer to these threads.

[mr moose]

49 minutes ago, Waffles13 said:

I'm not making any qualitative comment on the tech itself. But if you've spent a lot of time here or on other tech forums, every thread about this topic for the last 6-9 months has that one guy who needs to inform everyone that node naming conventions aren't 1:1 comparable, even though at this point at that should be relatively common knowledge at this point. Now if it turns out that Intel 10nm is still denser per nm but not quite as dense overall as other competing 7nm processes, it's going to add another fun layer to these threads.

Sounds to me like the people you are listening to know nothing about density. Yes we always get one or two that pop up and try to use dies size and transistor count to devise a number they think is relevant, but so what? maybe this article is wrong in their appraisal too.   So again why is it fantastic news that Intel's 10nm is not going to be as good as expected?  Is it just because you want some people who don't know what they are talking about to be visibly wrong?

[jagdtigger]

On 8/4/2018 at 1:41 AM, Stefan Payne said:

Clockrate could be no better than it is on Ryzen right now.

Seeing how intel hit a pretty hard wall in that regard what AMD is doing could be the solution IMO. Top it off no-one will care about Intel's higher clock rate if its hot headed like in case of Haswell...

[Hellion]

18 hours ago, Sypran said:

The problem is that CPUs have a long development time. A new architecture you hear about now, probably started development 3-4 years prior.
I can't find an exact year when Spectre as we know it was discovered, only that is was made public in 2018.

Something that is 100% immune to any variation on Spectre is probably in development, but thats likely half a decade away. Everything else in between is probably just gonna be patchwork on stuff that has long since been developed, and replacing how speculative execution works with them is flat out impossible, or would be too time consuming (leaving Intel still selling vulnerable CPUs for an extended time.)

As much as I personally hate it one thing you will hear from programmers a lot, and it remains true with hardware design: Developer Time > CPU time.
You could probably optimize something to run 2% faster, but the amount of time it take to do that is not a good trade off.

Which is both unacceptable and clear neglegence.

 

They know about the problem,  so stop selling flawed CPU's until the issue is fixed properly.

 

Really hoping intel is fined billions in the current court cases. They deserve it.

[Zodiark1593]

7 hours ago, Hellion said:

Which is both unacceptable and clear neglegence.

 

They know about the problem,  so stop selling flawed CPU's until the issue is fixed properly.

 

Really hoping intel is fined billions in the current court cases. They deserve it.

You expect Intel to altogether stop selling CPU's for a span of literally years? They're already losing marketshare to AMD. If forced to stop selling, there will probably be nothing left for Intel to sell to, and we'll wind up with an AMD monopoly (every bit as unappealing a thought as an Intel monopoly).

 

Please clarify the name of this fantasy land you reside in?

[mr moose]

On 8/5/2018 at 1:37 PM, Sypran said:

I can't find an exact year when Spectre as we know it was discovered, only that is was made public in 2018.

 

It has been speculated to exist for a very long time (I've heard some people claim 10 years but I suspect that's talking about cache timing attacks Which I don't know anything about but sounds impressive anyway), but they have actually only proved it could be exploited in 2017, Intel was made aware in july 2017. 

 

There is no way Intel could have processors ready for sale now that is hardware immune to spectre like flaws, And lets not forget AMD are updating and working on mitigation for this as well which clearly indicates they haven't got 100% immune hardware either.

[SupremeGOAT]

they're fiiiine~