Cannon Lake: Intel’s Forgotten Generation | Hacker News Reader

Cannon Lake: Intel’s Forgotten Generation

chipsandcheese.com

62 pointsrbanffy33 comments

I worked at Intel’s main R&D site from Sandy bridge to Meteor Lake bring-up on a team looking at electrical data. As I recall, the first Cannonlake wafers produced data relevant to me around New Years Eve 2015. This was not the first 10nm wafer by a long shot —- the process had been in development for years at that point and with many test chips run. Still, the initial data was so unbelievably bad, due to the lowest metal layers (supposed to be 10nm’s secret sauce), that even just looking at the basic line resistor test structures it looked like the measurement machines had malfunctioned or had the wrong hardware configuration.

Worst bring-up in my time. I feel bad for the architecture team. I know Intel is getting dunked on my AMD right now but internally the worst of it was years ago. I retain confidence in the direction things are heading now

LOL I remember working on Cannonlake A0. The dozen (not plural) parts we got passed sort but all bricked. Turns out the PLL was randomly losing lock due to variation alone. Also the analog designers refused to run timing checks on digital control logic (if we just match the layout it would work amirite?)

I left a year after that... tbh don't see things getting any better from talking to friends. We'll see.

Honestly, the past 5-ish years of Intel generations have blended together for me. I'm not even sure if they've already released a new one this year or what it might be called. Everything is some kind of Lake, they're all i3, i5, i7, i9, and they have some weird x000 naming scheme (then there are the utterly indecipherable xx00nn ones).

Am I the only one who prefers the simplicity of Zen 1, 2, 3, 4?

On the desktop side Intel's naming is not so bad. We are at 13000 now, 12000 was before, 14000 follows. They call it 13th gen accordingly, which is incorrect but memorable.

Zen isn't all that better. Sure, Zen 1, 2, ... seems clearer, but can you tell me on the top of your head which processor has which architecture? Including nice outliers like the Ryzen 5 1600 AF? Now do the same for the mobile processors, the naming there is a total clusterfuck.

Also, maybe a bit more on topic: The current processor generation is actually nice and gives a real performance increase. Enough so that all three (i5-13600K, i7-13700K and i9-13900K) take the top three positions in my meta benchmark (for games). AMD follows right behind that (and needs less energy to do so), so it's really a good race now, just like the article concludes.

> but can you tell me on the top of your head which processor has which architecture?

Not sure how that's relevant. By name, I have no idea if Cannon Lake is before or after Skylake. I only know because I've heard of Skylake longer than Cannon Lake. I'd have no idea how to place any of the post-Skylake architectures amongst each other chronologically or performance-wise. I know that Zen 1 is older then Zen 3, because that's how the naming scheme works. The same way I know that the PS5 came after the PS4.

It's like the Xbox bullshit. The only reason why I know Xbox One is newer than 360 is because I'm a gamer who pays way too much attention to this shit.

> I know that Zen 1 is older then Zen 3, because that's how the naming scheme works.

I think GP's point is that "zen" is the architecture, same as some random lake. Sure, 2 comes before three, whereas it's not clear whether the sky comes before the tiger.

Whereas the AMD model number is 5600, say.

But then you have interesting things, like 5500U mobile parts which are actually "zen 2+" and others, like the 5560U which are "zen 3".

This was an issue for me last year when attempting to jump on the "zen 3 bandwagon", but having to look up "enterprise laptop" models, which wouldn't specify which kind of zen the cpu was, only some xyz number. Most were 4xxx, so I knew those were "mobile zen 2", but I almost picked one of the "zen 2+" 5xxx models.

I haven't followed things that closely, but I'm pretty confident that an intel Xyyy is "gen X". And, usually, marketing copy says i5-Xyyy.

It's not always. For example "8th gen" Intel has both Kaby Lake and Coffee Lake CPUs, some gens have a mix of 14nm CPUs for desktop and 10nm for Mobile. It's honestly pretty confusing as to which Lake follows what at this point as they are slight refinements of slight refinements.

AMD is at least making it so the second digit will indicate architecture (the first will be year released).

"8th gen" is indeed an extreme case, with a very large number of slightly different microarchitectures being covered by this name, due to a chaos at Intel, caused both by the failed 10 nm process and by the Meltdown and Spectre bugs.

Nevertheless, all the "Kaby Lake" proper are "7th gen", not "8th gen".

Among the "8th gen" there are only "Kaby Lake R" (R from Refresh) and "Kaby Lake G" (G from GPU, i.e. discrete AMD GPU on package).

There are also 2 kinds of CPUs that have been weirdly branded as "8th gen", but which, both based on the date of their launch and on their features should have been branded as "9th gen", together with Coffee Lake Refresh: Whiskey Lake U and Amber Lake Y.

The remaining CPUs branded as "8th gen" were Coffee Lake and Cannon Lake U.

I did not know that about Intel. Guess I'm lucky I wasn't in the market for a new PC at that time... Do they at least all support Windows 11, since I think "gen 8" is the first one officially supported?

> AMD is at least making it so the second digit will indicate architecture (the first will be year released).

Unfortunately, that's not always the case. See, for example, the Ryzen 5 5500U which is a zen 2 part and the 5560U which is a zen 3 part. The architecture is quite different, too: the first being a 2x3 configuration, while the second is 1x6.

The Zen 3 one was also released as a 54xx "PRO" version. See [1] note 1.

They were both released in January 2021, but then you have the Ryzen 5 4680U which was released later, in April 2021, while being a zen 2.

As a matter of fact, I'm quite sure the second digit is the model's place in the range. See 3700 desktop parts, which are zen 2, and 5700 ones which are zen 3.

Moreover, to your point about differences between desktop and laptop parts, I've always considered that a given. Zen 2 desktop parts already had pci-e 4, whereas the zen 3 laptop ones only had pci-e 3, even the higher end ones. The first mobile parts to have pci-e 4 seem to be the 6xxx models.

---

[0] https://en.wikipedia.org/wiki/Zen_2#Lucienne_(5000_series)

[1] https://en.wikipedia.org/wiki/Zen_3#Cezanne_2

The AMD changes haven't taken place yet. They start 2023. Also I was slightly wrong with the 3rd digit actually being architecture.

https://www.theverge.com/2022/9/7/23340934/amd-ryzen-model-n...

The Windows 11 supported CPU list is pretty arbitrary and doesn't have much to do with the capabilities of the CPU. The main real requirement is TPM 2.0 and even that can be worked around. And there are some officially supported 7th gen CPUs because MS realized how dumb it was for the Surface Studio they were still selling to not be supported.

I remember Reddit discussions where they tried to figure out if the CPU support list lined up with any particular CPU features. It didn't make any much sense at all. Basically, Windows 11 supported 3-years old CPUs generations and (and later added) CPUs in Microsoft products that were actively being sold.

The confusion is intentional. It’s easier to sell an old thing if the customer can’t tell it’s an old thing, and a lot easier if they also don’t get any useful information about how the product compares to its competitors.

I’d be really surprised if this wasn’t the exact reason. When you go to Costco, Walmart, etc. you’ll see “Intel i7” in huge letters, but the actual generation might not even be listed somewhere you can see without opening the box.

Names like Skylake and Cannon Lake are just codenames for the architecture. AMD does the same thing with Matisse, Rome, Vermeer, Milan, etc. I actually find AMD naming mildly confusing Zen 3 Ryzen 5 5600x (Vermeer).

The actual break down is quite simple and described on their website: https://www.intel.com/content/www/us/en/processors/processor...

tldr; Intel processors are numbered like 13900K (13th gen) the first 1-2 digits are the generation. So 9800 is 9th gen 10850K is 10th gen.

Have you tried a power limited version of Intel to compare?

I have my 7700x limited to 65w with very little performance loss -- it's very easy with eco mode. I've seen lot of benchmarks with 7000 series chips on eco mode but not so much on Intel side.

These chips really should have been power limited with a little less multicore performance instead of being power drains.

Hey, kind of. I'm just collecting benchmark results to create a global ranking (global as in spanning release generations), not benchmarking myself. I saw that the eco mode of the AMD processors saw some attention, and then in the 13000 reviews as well. https://www.computerbase.de/2022-10/intel-core-i9-13900k-i7-... would be one example.

The Intel "core" series is just stale branding. Really stale branding. That name started being used in 2006 for an iteration of the Intel P6 microarchitecture (first implemented in 1995 with the Pentium Pro). Intel's NetBurst (Pentium 4) wasn't great, so they brought something old back, made improvements, made it dual core, and we had the Intel Core Duo. They then did the usual licensing back and forth with AMD for AMD64, and they released the Core 2 (Penryn) just a little later in the same year. The Core iN series started with the Nehalem microarchitecture in late 2008. This was a major improvement over, but still an iteration on, P6. However modest these initial changes were, we are now so far removed from those earlier designs that carrying the "core" branding is just bad. Intel really needs refreshed branding. They should also do away with the largely irrelevant numbers and letters that they shove on the end of their CPU family names.nI would even be okay with just "Intel 8core 3.6gHz 2020" and so on, but Intel tends to prefer cringe apparently.

The last canonical example of Intel refreshing their naming convention was with their Xeon line up which is now utterly incomprehensible, at a glance you used to be able to understand what was going on;

EX-YZZZ vA

E3 - Rebranded Desktop Class parts with ECC support

E5 - High-Desktop & Regular Multi-Processor Server parts

E7 - High-End Multi-Processor Server

Y - Denotes Socket Support. ZZZ - Denotes a model number, indicating from low to high increases in features, cores and/or clock speed. vA - Denotes Generation.

Xeon E5-2980v2 - is a high-core count server processor for use in dual socket servers, belonging to the Ivy Bridge Generation of parts.

The new naming "convention" is as follows - from a glance can you identify any information from this;

Xeon Bronze 3204

Xeon Silver 4214R

Xeon Gold 6212U

Xeon Gold 6342

Xeon Platinum 8280

Xeon Platinum 8380

Xeon W-2295

Bronze, Silver, Gold, Platinum - are so unclear as to be meaningless - there are dual socket Silver processors more capable than some Gold processors. W prefixed parts are Workstation parts. The First Digit bears no connection to anything (except in the platinum series where it denotes 8 socket capability), perhaps PCI-E Lanes? The Second digit apparently denotes generation. The Suffix denotes that it's only capable of running in a single processor configuration - except when it doesn't.

Ohh, and the 'E3-**** v*' convention still stands for desktop rebranded parts.

It's a complete mess.

> The First Digit bears no connection to anything

One pattern is: If first digit is 3,4,5 then it has one AVX512 FMA unit, if it is 6,7,8 then it has two.

...except the 5122 (2), 5219Y (2), 5222 (2) and 6122 (1)

good shout - I'm not so up to speed with what AVX512 can accelerate, but is it such a distinctive feature that it warrants a full delineating model number?

an extra avx512 fma unit means it can do twice as many z=a*b+c instructions per clock. this only really matters for matrix multiplication (for anything else you are memory bandwidth limited), but 2x matmul performance is a really big deal for hpc

> Intel's NetBurst (Pentium 4) wasn't great, so they brought something old back, made improvements, made it dual core, and we had the Intel Core Duo.

Core was based on the mobile-focused Pentium-M design from their Israeli team. That lineage traces back to the Pentium III but both the Pentium-M and Core designs had significant modifications so I wouldn’t describe it as bringing something old back — more like admitting that the NetBurst big gamble had failed and they switched to a design more like what everyone else was making. The P4 was kind of interesting as Intel’s second bust around the turn of the century based on a design which tried to hit higher performance assuming everyone had access to highly-sophisticated compilers and would use them carefully. As with Itanium, I do wonder whether anything would have been different if they’d open sourced icc – the profits weren’t significant and while that could have benefited AMD, I doubt that would have outweighed two major flops.

With the way Intel segments CPUs you need the stupid numbers. "Intel 8-core 3.6GHz 2020 no integrated GPU, no ECC with unlocked multiplier" is a bit of a mouthful.

That's a really fair point. I support you could do something like:

i82020GEM or i82020NNN

so you have intel 8 core, 2020, GPU, ECC, multiplier or without... then we're back to really bad clusters of numbers and letters, but at least in that scenario it has meaning.

But anybody who can parse that without looking it up obviously cares enough to look at the current product numbering scheme and hop on over to ark.intel.com all the same.

My biggest issue with that is I couldn't find it documented anywhere and it feels like they come up with new suffixes every year. I was on board with F and K, but now there's a bunch of others that I just have no clue what they mean.

https://www.intel.com/content/www/us/en/support/articles/000...

To be fair, Intel also uses "13th generation Core" instead of Raptor Lake, "12th generation Core" instead of Alder Lake, "6th Generation Core" instead of Skylake, and so on, while AMD also has meaningless code names that are used for products before launch, e.g. Raphael for Zen 4, Vermeer for Zen 3, Matisse for Zen 2, Pinnacle Ridge for Zen 1+, Summit Ridge for Zen 1.

In both cases, the shorter and easier to remember names have won in the conscience of most of the public, i.e. it is easier to remember Zen 2 than Matisse, but it is easier to remember which was Kaby Lake instead of which was the 7th Generation Core.

Because both companies launch products typically every year, I would prefer something like Intel Core 2017 instead of Coffee Lake and AMD Zen 2022 instead of Zen 4.

This is what happened when a company has marketing departments making such decisions unchecked. Sun had the same problem with their idiot marketing dept naming the Sun Niagra systems bizarre names like CoolThreads 1000 and systems with numbers like 5420. Larry came in and said it was stupid and to name them precisely what they were so they became T3-1, T4-2, etc. to indicate the chip generation and the number of chips on the system board.

It's a good thing they fixed that problem and became the powerhouse that they are today. </sarcasm>

AMD will be selling laptop chips of various older generations under one name soon in order to deceive consumers who aren't paying attention: https://youtu.be/MipuN4RU4m8

A 7000 series laptop chip could be Zen 4, 3, or 2...

Agree that Intel naming schemes are very confusing and very few people understand them or care, even.

However, AMD isn't any better ... is it a Vermeer or is it a Cezanne, a Chagall or a Milan?

Ryzen 5xxxG(E) or Ryzen 5 5500 -> Cezanne

Ryzen 5xxx -> Vermeer

Threadripper 5xxx -> Chagall

EPYC xxx3 -> Milan

Reply on news.ycombinator.com