News Deciphering Intel's Codenames: What's With All the Lakes?

[Admin]

Ever wondered why Intel CPUs have code names like Coffee Lake and Comet Lake? We spoke with Intel and a former engineer to find out.

Deciphering Intel's Codenames: What's With All the Lakes? : Read more

 

[thGe17]

One important thing to note: Intel has split up CPUs, microarchitetures and process nodes in 2019.
The original Skylake was the code name for processors (consumer and also later Skylake SP), it was synonym for the microarchiteture and bound to a specific process, here 14nm (established in 2014).
Again in 2017, Coffee Lake is the CPU and microarchiteture (a minor advancement over Skylake) and bound to the 14nm++ process.
With the new roadmap (2Q19), Intel broke up the static structure, now there are CPU designs like Ice Lake Y/U, Tiger Lake U, Ice Lake SP, there are microarchitectures like Skylake-based architetures as Coffee Lake Refresh and Comet Lake and new ones like Sunny Cove, Willow Cove, Golden Cove and there are different manufacturing nodes in use like 14nm, 10nm, and 7nm is in development an will be used as of 2HY21.
Intel now provides CPU generations with a mixture of different ingredients:

Some examples:

10th Gen Core CPUs:
Comet Lake U , Comet Lake (Skylake-based) , 14nm+++ (Gen9.5 iGPU; same for Comet Lake S)
Ice Lake U , Sunny Cove , 10nm+ (Gen11 iGPU)
11th Gen Core CPUs:
Tiger Lake U , Willow Cove , 10nm++ (Gen12 iGPU, Xe)
Future CPUs:
Lakefield , Sunny Cove & Tremont , 10nm+(+?)
Ice Lake SP , Sunny Cove , 10nm++ (Xeon, 4Q20)
Rocket Lake S , prob. Willow Cove , 14nm+++(+?),
Sapphire Rapids SP , min. Willow Cove, 10nm+++ (Xeon, PCIe5, DDR5, 2021)
And possibly something like:
Alder Lake S , Golden Cove & Gracemont , 10nm+++ (8 + 8 cores)

 

[holmes4]

Two comments based on the 15 years I spent at Intel.

1) Geographic place names can't be trademarked. Intel ran into this with one of the Itanium designs, which the ex-DEC team that worked on it called Tanglewood. Though Tanglewood is a place, it is the name of a commercial performing arts venue in western Massachusetts where the Boston Pops plays its summer concerts. The organization that runs Tanglewood complained to Intel and the processor name was changed to "Tukwilla (Design)" (keeping the TWD three-letter code used internally - Tukwilla is a city in Washington state). I had lobbied for "Townsend", a town in Massachusetts, but was ignored...

2) Sandy Bridge was originally Gesher, a name picked by the Israeli design team and an archaeological site. But this was also the name of an Israeli political party and was deemed to be a problem, so it was changed to Sandy Bridge (which I gathered was sort of a translation.) This was the start of the bridge theme.

 

[holmes4]

"So then Intel learned that your codename should be something that not many people care about. "

When the Penryn name became public, the people who lived in Penryn (UK) cared a lot, as suddenly web searches for the town's name were flooded with hits regarding the processor. Just now I did a search on the name and the first hit was for the processor, not the town. There were lots of complaints, but nothing to be done about it and the furor died down eventually. I think this experience also contributed to shifting away from town/city names to much more obscure lakes, rivers, canyons, trails, peaks, coves, etc.

 

[DZIrl]

AMD also has misleading naming, eg Ryzen 3000G based on older architecture. 99% buyers think 3000G CPU is similar to Ryzen 3000 but its not!
Is this by accident? Not sure!

 

[Integr8d]

AMD also has misleading naming, eg Ryzen 3000G based on older architecture. 99% buyers think 3000G CPU is similar to Ryzen 3000 but its not!
Is this by accident? Not sure!

Nope. And neither was 1st-gen Ryzen chipset naming... Real question is: what motivated this puff piece?

 

[cryoburner]

Giving an example, he said "If I look at Skylake and I look at Coffee Lake, there really is a huge difference in between. They're not the same architecture. They are extremely advanced versions of previous processors." However, and he argued that this applies to both official names and code names, "Because they keep the lake name [across both], they end up giving the impression of just being incremental."

A huge difference? Per-core performance hasn't really improved that much outside of minor bumps to clock rates each generation, and efficiency hasn't improved to any notable degree either, due to them being manufactured on a similar process node. The overall IPC is similar, they use the same generation of RAM and PCIe, and integrated graphics performance hasn't improved much from where it was five years ago. It's hard to give people the impression that a product is new and exciting when each generation is more or less just increasing clock rates by a few percent from the previous year, and maybe patching some security flaws.

About the only thing that's significantly changed to the end user is that higher core counts have finally transitioned to lower price points and made their way to mainstream platforms after stagnating for a decade. And that's great, though it doesn't strike me being a major change in terms of the core architecture.

I had lobbied for "Townsend", a town in Massachusetts, but was ignored...

Townsend sounds like a really boring name compared to Tanglewood or Tukwilla. : D

 

[Giroro]

"The goal for them is to get something people can understand what it is internally, while outside, people are confused about what it is."

That's a terrible goal considering these codenames get used on public-facing roadmaps that get started with the press and investors.

Also the Tom's Hardware mobile site is difficult to scroll without changing pages, and keeps crashing my browser

 

[bit_user]

This:

Lake-themed code names would become more and more common at the company up until 2015's Skylake, after which they would become the dominant naming scheme for the company's consumer-facing projects into 2020.

and this:

"In general," he elaborated "on the client side of the company, our SOC name and platform names are now the same and they are lakes."

Don't explain Cascade Lake (Intel's current 14 nm Xeon generation).

This does:

"The goal for them is to get something people can understand what it is internally, while outside, people are confused about what it is."

For example, if Intel now names everything after lakes, it's difficult for those outside the company to tell products apart. There's also no immediate reason why "Rocket Lake" might be better than "Comet Lake," going by just the name.

But that basically defeats the main point of the article, practically dismissing any underlying logic behind the names.

Continuing on this theme:

Tripp did contradict Piednoel's assertion that the company's confusing code names are intentional. "We definitely work hard to avoid any confusion for our fan community,"

Wow, well they certainly haven't succeeded! How else do you justify calling the 9000-series Coffee Lake Refresh, when it appeared to differ more from the 8000-series than Kaby Lake (the 7000-series) did from Skylake (the 6000-series)?

And why do Lakes span all the way from low-end laptop SoCs to desktop, HEDT, workstation, and even "Platinum"-level Xeon Scalable CPUS? And with Ice Lake, they even expanded the lakes to cover 10 nm. This does not seem like they're trying to pierce through the veil of confusion!

 

[bit_user]

"Maybe people don't understand this, but the biggest competitor for Intel is a four years old PC from Intel,"

Yes, that was still true in 2017, when he left Intel. But, it's gotten less true with each passing year.

the idea that simply updating the generation number on a product gives the impression that it is simply a slightly improved version of a last-gen processor, rather than a unique piece of hardware with specialized abilities of its own.

This does a good job of explaining why Nvidia jumped to the RTX 2000 series, and even why AMD switched naming schemes with RDNA/Navi.

 

[bit_user]

"If I look at Skylake and I look at Coffee Lake, there really is a huge difference in between. They're not the same architecture.

Okay, now I'm done believing this guy, too.

All of the empirical data says they're the same micro-architecture, even if he asserts otherwise.

For code names, Piednoel suggested that fixing this issue would be as easy as just dropping themes. "I think they should stop doing this and go back to names like Katmai, Willamette, Prescott. They don't sound the same."

I liked when "Bridge", "Well", or "Lake" would at least tell you which CPU socket and motherboard generation you needed. At least it meant something useful.

"When Sean Maloney decided to call it 'core,' a lot of people were confused

Yes. This was the dumbest thing Intel's marketing ever did. All the confusion around Core 2 Solo vs. Core Duo was just completely unnecessary. And now the "core" part of the processor names is just an adjunct that's often dropped, since it really adds no value.

"If you drive a BMW 750 from the '80s, you're not saying 'I'm driving an E32 [the code name for that model]'," he said. "Nobody does.

Um, anyone still driving an E32 would certainly know its chassis number. You can't drive a car that old without frequenting the online forums and having lots of conversations with your car buddies and/or a local repair shop specializing in them.

I get the point, but it was a bad example. He should've picked on a current generation, but he probably didn't know the chassis number, off hand.

 

[NightHawkRMX]

Many people do know their chassis/generation codenames for cars, though it is mainly enthusiasts, the same with CPUs.

Anybody that drives an M3 knows if they have an E30, E36, or whatever. Anybody that drives a Mustang knows if they have a S550, S197, Newedge, whatever.

 

[bit_user]

Intel CPU Code Names Through the Years

Without being separated by market segment or product type, this adds very little value.

It's more informative to go back and look at some leaked roadmap slides. Then, you can start to pick up on more patterns and themes.

 

[bit_user]

Real question is: what motivated this puff piece?

I actually enjoyed it, in spite of my issues with many of the interviewees' statements.

I mean, maybe you already knew this, but I sure didn't:

in the case of Conroe, that Piednoel's team was the only group at Intel to know what it was. More impressively, they also sent it off to OEMs and ODMs for evaluation by disabling its execution unit and telling them it was a better Yonah — an earlier Intel processor. So while this made their work more difficult, it also meant they were able to show up at trade shows with a completely unleaked processor that was suddenly "40% faster than the previous processor."

That's a pretty cool bit of Intel lore. I know Engineering samples are often somewhat crippled (either by design, or just early hardware errata), but it sounds like it certainly wasn't a common practice to do it intentionally, back then.

They should interview that guy some more, although maybe he hasn't been very accessible until now. I'll bet he has all kinds of great nuggets, like that.

 

[bit_user]

With the new roadmap (2Q19), Intel broke up the static structure, now there are CPU designs like Ice Lake Y/U, Tiger Lake U, Ice Lake SP,

I'm pretty sure that the Y/U/S and Xeon D/E/W/SP market segmentation goes back a lot further than that.

new ones like Sunny Cove, Willow Cove, Golden Cove

AFAIK, this is what's new - separating out the uArch codenames of the cores from the codenames of products which utilize them. Previously, Intel would have a product, like Wiskey Lake, and the fact that it shared the basic uArch with Skylake was just something informally known.

Some examples:
...
Lakefield , Sunny Cove & Tremont , 10nm+(+?)
...

Somebody should really be slapped for that one. It'd almost be okay, if the dies it contained spanned the "lakes" generation and a hypothetical, new "field" generation. Otherwise, it's almost like it exists just as an affront to anyone trying to divine a pattern in intel's maelstrom of codenames.

 

[bit_user]

AMD also has misleading naming, eg Ryzen 3000G based on older architecture. 99% buyers think 3000G CPU is similar to Ryzen 3000 but its not!

Intel has long done this with their high-end desktop processors. You'd have like the i7-6850K, which was actually built around cores from the 5000-series. However, since it had more cores and the launch of their high-end segment was routinely delayed by almost a generation, they liked to position it above the current-generation mainstream CPUs, like the i7-6700K that launched before it.

It's just an example of marketing people "adding value". If the naming structure was too rigid and technically accurate, then it could self-perpetuate and you could just get rid of most of your marketing department. So, they like to do nonsensical things like above, and like using "i7" branding on dual-core laptop CPUs, back when every other example of an i7 processor had 4 cores.

I think it comes down to establishing a rule for just long enough that the market learns it, and then you break it, in order to juice up sales of a few key models or product segments. This is why they get paid the big bucks. That, plus all the business school jargon, they like to throw around to obscure just how simple their job really is.

 

[vinay2070]

They should have stopped the lake naming with 10nm. Would have been less confusing.

 

[Exploding PSU]

Back then I always thought the Bridge in Sandy Bridge or Ivy Bridge refers to a component somewhere inside the CPU, hey I'm no engineer, how would I know better.

Also, to quote

"We obviously want names that give off an image of strength, speed, power, etc."

They did well with Devil's Canyon. I know it's name of a place but still, "yeah I'm running the new Devil Canyon, what do you have" sounds great

 

[crysis2]

Nice article!

 

[nutraman]

Ever wondered why Intel CPUs have code names like Coffee Lake and Comet Lake? We spoke with Intel and a former engineer to find out.

Deciphering Intel's Codenames: What's With All the Lakes? : Read more

The new one is named COW LAKE because, you know, Intel is really MILKING IT!

 

[everettfsargent]

Ever wondered why Intel CPUs have code names like Coffee Lake and Comet Lake?

Hell no!

But go ahead and confuse the matter, even more so.

They should name their final 14nm CPU Vast Lake, and their 1st anything above 45W and on anything less than 14nm Last Lake, which I would expect sometime in 2022.

Other good names: Hot Lake, Boiling Lake, Vapor Lake, Plasma Lake, Sun Lake, Star Lake, Black Hole Lake, Big Bang Lake, Terawatt Lake ... you know, something so hot and fast, that just buying the thing voids the warranty

 

[rpcy]

Back in the mists of time, say, the 1980's, Intel named its microprocessor developments P3, P4, P5, P6. As we were finishing P6 around 1995, I was asked to speak about it at an evening Microprocessor Reports dinner. I was asked what I'd be working on next, and the real answer at the time was "P67", the idea being that we'd extend P6 into other markets before undertaking the new flagship that came to be called Willamette. I ducked the question, because I realized at that moment that the name of the project was much too revealing about Intel's product development plans. The logical answer would have been P7, and had I given that answer, the implication would have been that the P6 was beginning the next project, which would be another x86. But had I answered with the real name, P67, it would have been a tipoff that the next project would not in fact be the next x86 flagship, but something else instead. Either way, it wasn't information we wanted to make public. I relayed this problem to our executive VP, who immediately understood the concern, and he asked marketing to come up with a naming scheme that (a) would not tip our hand in the future, and (b) would be unlikely to get us sued by anybody. That is how the Pentium 4 became "Willamette", which is a river in Oregon. (I had asked for the name to be a different Oregon river, but this same VP didn't like it. Otherwise we would have been the Rogue Project.) The tradition has continued ever since.

 

[bit_user]

As we were finishing P6 around 1995, I was asked to speak about it at an evening Microprocessor Reports dinner.

Great story, and I really do appreciate any firsthand accounts we get from insiders.

However...

I was asked what I'd be working on next, and the real answer at the time was "P67", the idea being that we'd extend P6 into other markets before undertaking the new flagship that came to be called Willamette. I ducked the question, because I realized at that moment that the name of the project was much too revealing about Intel's product development plans.

The correct answer is "no comment". Always. Without even a second thought.

Since it's the expected answer, the press can't even read into it. They shouldn't ask, but you can't really blame them for trying.

I relayed this problem to our executive VP, who immediately understood the concern,

And he should've confirmed that "no comment" was the right answer, whether or not the codename tipped off anything about the company's plans.

However, even if workers don't go blabbing about their projects, codenames do tend to leak. So, perhaps he appreciated that they needed codenames which wouldn't mean anything to anyone, when they eventually did leak.

I had asked for the name to be a different Oregon river, but this same VP didn't like it. Otherwise we would have been the Rogue Project.

I can't blame him. I think "Rogue" is not really in line with the image Intel was trying to project to its customers, at the time.

BTW, was the P67 what would eventually become the Pentium 2?

 

[rpcy]

Er, you are in a different reality than I was/am. You have to call your Intel development project something, and you have to reveal it to your potential customers, to make sure you're communicating. It's not really leaking. We visited customers like Dell, IBM, and Microsoft all the time, because they had a vested interest in our expected feature set, schedule, performance, and so on. You have to tell them the name of your project. It's similar to what the DoD does with its efforts -- they have a big book with random code names, and when they need to name something, they pull out a couple and paste them together. That way no unintended information is leaked.

Intel violates this a little bit with the names they let the teams choose. The Oregon teams tend to pick Oregon names, California teams pick CA names, and the Israelis like names like Banias and Dothan.

No, P67 turned into Willamette. Pentium II codename was Klamath.