News EU Re-Imposes $400 Million Fine on Intel

[Nick_C]

Shouldn't AMD develop their own compilers instead of depending on Intel's compilers. At least AMD should pay some royalty to Intel to let them use Intel's compilers instead of crippling or blocking.

Developers use compilers that produce code that runs on compatible CPUs - and Intel and AMD CPUs are in the broad sense compatible - in the case of them using ICC, their choice of compiler would adversely affect the performance of their software on non-Intel CPUs, unbeknownst to the developer.

Other compilers are of course available, and I would hope that developers are aware that they can't rely ICC as a compiler to provide the best result for their customers, if their customers use non-Intel CPUs.

 

[thestryker]

Intel's compilers used to be way ahead of everyone else's and they stopped most of the vendorid type stuff. I'm not sure if it's still the case, but with their big oneapi push playing games doesn't benefit them currently.

 

[bit_user]

That's what the 14th gen is, instead of releasing a new arch as they should they released a gen they don't have to pay any (more) R&D for and that will be cheaper to produce due to fab improvements and that will release without any competition at all for new products.

That's one interpretation. Only Intel knows for sure why Meteor Lake-S (i.e. the desktop version) was cancelled.

 

[bit_user]

AMD paid 5,4bil for ati in 2006, a year they made negative money in...
AMD ruined themselves with that move, so much so that they were forced to sell their fabs and the only thing that allowed them to keep going for long enough to turn it around was the 1.25bil they got from intel.

Counterfactuals are hard. I think we simply don't know enough to make a good guess at what would've happened had AMD not bought ATI.

For sure, AMD still would've been outclassed by Core 2 - no question about it - that was already baked into the cake. However, AMD wouldn't have had a credible iGPU which they could later integrate into their CPUs and use to win the console market.

I think we can also say that selling its fabs is something AMD would've ultimately had to do, regardless. Maybe it would've been even worse for them to have sold them later? Hard to say.

 

[bit_user]

Shouldn't AMD develop their own compilers instead of depending on Intel's compilers.

They do.

​

https://www.amd.com/en/developer/aocc.html

​

These vendor-specific compilers are mostly just used by the HPC community and by system vendors to achieve the best scores in benchmarks like SPEC2017.

One issue, along the lines of what @Nick_C mentioned, was an optimized library (Math Kernel Library?) which Intel developed. I'm not sure if ICC does/did ever include GenuineIntel checks, vs. simple CPUID feature checks. That library is more likely to find its way into commercial software and other 3rd party code than software compiled using ICC.

At least AMD should pay some royalty to Intel to let them use Intel's compilers instead of crippling or blocking.

They can't make Intel do anything, even if AMD is willing to pay money for it. Only the courts can compel Intel to take a certain course of action.

Intel's compilers used to be way ahead of everyone else's and they stopped most of the vendorid type stuff. I'm not sure if it's still the case,

Not so much, because Intel recognizes that a tiny minority of code is compiled with ICC. So, they have mostly switched gears to contributing optimizations to GCC and LLVM (Clang).

 

[thestryker]

I think we can also say that selling its fabs is something AMD would've ultimately had to do, regardless. Maybe it would've been even worse for them to have sold them later? Hard to say.

As much as I'd love to disagree with you here I think you're 100% right. Look at the pressure Intel has been under to dump fabs by investors (FWIW I think this would be a foolish move, but money people love the idea due to how capital intensive fabs are). There's no way AMD would have been able to last even this long with fabs buying ATI or no.

 

[JarredWaltonGPU]

As much as I'd love to disagree with you here I think you're 100% right. Look at the pressure Intel has been under to dump fabs by investors (FWIW I think this would be a foolish move, but money people love the idea due to how capital intensive fabs are). There's no way AMD would have been able to last even this long with fabs buying ATI or no.

All you have to do is look at what's happened with GloFo, even with billions in apparent funding from the new owners, to see that it was a money sink. GloFo isn't dead, but AMD didn't have the ability to keep doing updated fabs.

Intel meanwhile is effectively in the same boat, which is the impetus behind IFS, Intel Foundry Services. Theoretically, if Intel executes well on foundry as well as CPU/GPU stuff, it could keep going the same way as before. In practice, opening up the foundries to other customers means that Intel isn't as dead in the water if there's a problem with a new CPU architecture. Likewise, Intel is looking at ways to maximize profits by using other foundries where that makes sense (TSMC doing GPUs for example).

The jury is still out on whether Intel can keep pace once IFS really gets rolling, though.

That's one interpretation. Only Intel knows for sure why Meteor Lake-S (i.e. the desktop version) was cancelled.

As far as I know, it's only a rumor that MTL-S was cancelled. The latest reveals from Intel Innovation to me at least suggest those rumors are incorrect. However, it's still possible that MTL-S ends up as the modern equivalent of Broadwell desktop chips. But there are plans for a variant of Meteor Lake with half the GPU, using a different GPU tile. That would be a reasonable fit for desktop Meteor Lake. Even if MTL-S does come out, though, it might not launch until after the mobile parts.

 

[bit_user]

As far as I know, it's only a rumor that MTL-S was cancelled. The latest reveals from Intel Innovation to me at least suggest those rumors are incorrect. However, it's still possible that MTL-S ends up as the modern equivalent of Broadwell desktop chips. But there are plans for a variant of Meteor Lake with half the GPU, using a different GPU tile. That would be a reasonable fit for desktop Meteor Lake. Even if MTL-S does come out, though, it might not launch until after the mobile parts.

The thing which flies in the face of MTL-S appearing is its low apparent clock speed ceiling. According to this, it topped out at just 4.8 GHz:

​

Intel Meteor Lake Core Ultra 7 1002H Leaked in Geekbench

Results won't make owners of laptops with previous gen CPUs envious

www.tomshardware.com

That's at least going to limit it to low-power market segments, like the lower-end of the 65 W segment. Now, let's explore how much room exists for it to surpass Raptor Lake.

Model	Max Turbo Clockspeed (GHz)​
i9-13900	5.6​
i7-13700	5.2​
i5-13500	4.8​
i3-13100	4.5​

So, with Meteor Lake apparently topping out at 4.8 GHz, it would only make sense to slot in at the i5-14500 and below. In those same benchmarks, it had only a 8.9% IPC benefit, which would make it perform about the same as a 5.23 GHz Raptor Lake. That means it wouldn't be an upgrade over the i7-13700.

I'm mindful of the fact that Intel hasn't refreshed its smaller die Alder Lake-S. It could be that they're going to do exactly as you say, and use Meteor Lake for that. However, Meteor Lake was set to coincide with a new socket. That makes me very skeptical that Intel can/will "backport" it to LGA1700, because I believe we know that Meteor Lake will not support DDR4, and I think Intel won't want the headache of dealing with a CPU that only works in a limited selection of those boards.

On the flip side, if Intel indeed launches Meteor Lake-S on LGA1851, it seems like such a small market niche they'd be targeting that motherboard selection will be quite poor. Practically, limiting it to OEM-only.

 

[JarredWaltonGPU]

The thing which flies in the face of MTL-S appearing is its low apparent clock speed ceiling. According to this, it topped out at just 4.8 GHz:

​

Intel Meteor Lake Core Ultra 7 1002H Leaked in Geekbench

Results won't make owners of laptops with previous gen CPUs envious

www.tomshardware.com

That's at least going to limit it to low-power market segments, like the lower-end of the 65 W segment. Now, let's explore how much room exists for it to surpass Raptor Lake.

​

​

Model	Max Turbo Clockspeed (GHz)​
i9-13900	5.6​
i7-13700	5.2​
i5-13500	4.8​
i3-13100	4.5​

So, with Meteor Lake apparently topping out at 4.8 GHz, it would only make sense to slot in at the i5-14500 and below. In those same benchmarks, it had only a 8.9% IPC benefit, which would make it perform about the same as a 5.23 GHz Raptor Lake. That means it wouldn't be an upgrade over the i7-13700.

I'm mindful of the fact that Intel hasn't refreshed its smaller die Alder Lake-S. It could be that they're going to do exactly as you say, and use Meteor Lake for that. However, Meteor Lake was set to coincide with a new socket. That makes me very skeptical that Intel can/will "backport" it to LGA1700, because I believe we know that Meteor Lake will not support DDR4, and I think Intel won't want the headache of dealing with a CPU that only works in a limited selection of those boards.

On the flip side, if Intel indeed launches Meteor Lake-S on LGA1851, it seems like such a small market niche they'd be targeting that motherboard selection will be quite poor. Practically, limiting it to OEM-only.

Eh, early leaks on Geekbench and such don't mean much. It's almost certainly an engineering sample, and may have been intentionally clock limited. If it does top out at less than 5GHz, yeah, that's a potential issue. A better architecture could still let it hit higher levels of performance, but we're at least 3 months way from the actual launch (from the time of that leak). And a mobile part would logically get clocked lower as well.

Anyway, I'm not saying MTL-S isn't cancelled. I don't know, Intel hasn't officially said anything. It could still appear, for LGA1700 or even a different socket. Having RPL-R alongside MTL-S for desktops and having both be "14th Gen" though... it would be super weird. And weird things have happened before.

 

[bit_user]

If it does top out at less than 5GHz, yeah, that's a potential issue. A better architecture could still let it hit higher levels of performance,

If you look at the comments of that article, you'll see I already did the math - it works out to an 8.9% IPC improvement, which is how I projected parity with a Raptor Lake running at roughly 5.2 GHz.

Anyway, I'm not saying MTL-S isn't cancelled. I don't know, Intel hasn't officially said anything. It could still appear, for LGA1700 or even a different socket. Having RPL-R alongside MTL-S for desktops and having both be "14th Gen" though... it would be super weird. And weird things have happened before.

When Ice Lake shipped for laptops, Whiskey Lake was marketed along side it, in the HX-tier laptops. Whiskey Lake was a 14nm Skylake-derived CPU, akin to Coffee Lake-R, I believe.

 

[thestryker]

When Ice Lake shipped for laptops, Whiskey Lake was marketed along side it, in the HX-tier laptops. Whiskey Lake was a 14nm Skylake-derived CPU, akin to Coffee Lake-R, I believe.

Comet Lake was the 10th gen line across mobile (there were no Ice Lake i9 mobile CPUs just i3/5/7) and desktop. 8th gen was mixed between Coffee, Kaby and Whiskey which has always amused me given they're all Skylake derivatives.

 

[bit_user]

Anyway, I'm not saying MTL-S isn't cancelled. I don't know, Intel hasn't officially said anything. It could still appear, for LGA1700 or even a different socket. Having RPL-R alongside MTL-S for desktops and having both be "14th Gen" though... it would be super weird. And weird things have happened before.

Well, according to this, it's going to land in early 2024:

​

View: https://www.youtube.com/watch?v=KQW5lfQtLTs​

​

It'll be interesting to see if they're doing a full-range launch, or if it'll be focused mostly on the lower-end of the product stack. The latter is the only thing that makes much sense to me, since the low-end has yet to be properly refereshed since Alder Lake. Then, maybe Arrow Lake will just replace the higher-end models that are based on Raptor Refresh and the lower-end Meteor Lake ones will just get a re-badge and minor specs tweaking (as they have since Alder Lake).

I doubt Intel would do 3 full-range desktop launches within only a 12 or 13 month period. So, they must be partial-range launches.

 

[mikewinddale]

Oh, right. Because nothing anyone ever concluded in 1958 could possibly be wrong.

Let me tell you some things economics has learned since then:

• Humans aren't perfectly rational market participants
• Participants' access to information affects the efficiency of markets
• Systems thinking & modelling has grown by leaps and bounds

That's enough to make me very skeptical of some old orthodoxies about things like pricing.

Everything you just listed was already known in 1958. That humans aren't perfectly rational in a Neoclassical sense, and that access to information affects the efficiency of markets, was known a long time ago. E.g. Hayek, Friedrich A. "Economics and knowledge." Economica 4.13 (1937): 33-54.

And none of it is relevant to antitrust. Whether humans are perfectly rational or not does not affect whether predatory pricing is a rational strategy. As it turns out, predatory pricing is an irrational strategy, so even if someone was irrational enough to try it, they would end up hurting themselves more than they hurt their intended victims (their competition).

So if anything, the less rational people are, the less antitrust has to worry about predator pricing. Because the less rational people are, the more they will use predatory pricing strategy, which means they will actually hurt themselves, not their competition.

 

[bit_user]

As it turns out, predatory pricing is an irrational strategy, so even if someone was irrational enough to try it, they would end up hurting themselves more than they hurt their intended victims (their competition).

A lot of these generalities don't apply in special circumstances. For instance, if someone has a lock on a market through vendor lock-in or aggressive use of patents, then it can be very profitable to engage in predatory pricing. This has been demonstrated by drug companies, time and again.

 

[helper800]

Everything you just listed was already known in 1958. That humans aren't perfectly rational in a Neoclassical sense, and that access to information affects the efficiency of markets, was known a long time ago. E.g. Hayek, Friedrich A. "Economics and knowledge." Economica 4.13 (1937): 33-54.

And none of it is relevant to antitrust. Whether humans are perfectly rational or not does not affect whether predatory pricing is a rational strategy. As it turns out, predatory pricing is an irrational strategy, so even if someone was irrational enough to try it, they would end up hurting themselves more than they hurt their intended victims (their competition).

So if anything, the less rational people are, the less antitrust has to worry about predator pricing. Because the less rational people are, the more they will use predatory pricing strategy, which means they will actually hurt themselves, not their competition.

The problem with saying predatory pricing does not work is that all we need is one counterexample to prove your statements incorrect or force you to caveat what you are saying heavily. I would say that your premise is on shaky ground because predatory pricing is an illegal practice shown to work, and as such, is a criminal offense to prevent its use in different markets.