Question Will Tiger Lake be Intel's Core2Duo of 2020?

[Hulk]

Edit! Big mistake in the 1st sentence! I've never owned an AMD PC, I originally wrote Intel! Sorry about that.

I should state right up front that I have never owned an AMD PC. All the way back to my IBM SX25 it's been Intel for me. But I have to say that Intel seems to have gotten themselves into quite a pickle. It feels like the P4 Netburst fiasco of the early 2000's, but I think is going to be a much tougher mountain to climb for Intel.

As we all know through the 80's and 90's AMD seemed to be more of an annoyance to Intel rather than an actual threat. It was only when Intel got caught with their pants down with Netburst did AMD present a serious challenge with the original Athlon. Luckily for Intel they had the brilliant Merom mobile core designed by the Israeli engineers and a huge process advantage.

In 2005 with the release of Core2Duo it all turned around for Intel. They had AMD handily beat on both frequency and IPC. And then they moved along quite nicely for a number of years with tick tock until the 14nm process. And AMD began to catch up.

From and "Intel guy" perspective AMD was struggling along. But they kept at it. And Intel seemed to stall. While this is anecdotal, I generally upgrade my CPU every 2 or 3 years. But I've been on Haswell for 7 years now. Until Coffee Lake there really hasn't been anything that seemed like a significant upgrade to me in the $400 price range.

Coffee Lake looks really good to me. Until I really started to research the new 3000 series Ryzen parts. I have to admit these parts are impressive. These are actual cores, not half cores, or weak in comparison to Intel.

I was considering a 9900k but you can get a basically equal Ryzen part for $200 less. AND you have an upgrade path to a 12 and then 16 core part. With the 9900k it's the end of the line.

From an architectural point of view I think it's fair to say Intel has been caught (at least). They are offering better value and better performing chips.

So can Intel pull a rabbit out of the hat? Unless Tiger Lake is to Ryzen is what Core2Duo was to the Athlon I think Intel is going to have some rough times ahead. To pull off what they did in 2005 they will have to clearly surpass Ryzen in IPC, and beat them in frequency and efficiency and core count. And on top of all that they need to do it as competitive price points.

The way Intel has been plodding along IPC-wise and struggling with process it seems like they have a big mountain to climb.

Me, I'm going to wait to see what happens with Tiger Lake before I buy a new system. But AMD is really on a roll. Gotta love it. The CPU Wars are on!

 

[Arkaign]

There is no point building a 7nm fab now. Put in an enormous order on 5nm, secure wafers now, and give TSMC the support to build their next generation.

Yeah, I agree, don't know how long we'll be at 7/7+, but it would probably take more effort to build and staff a new FAB expansion than it would be relevant if purely 7nm.

Either way, they need capacity if they're going to BE the x86 front-runner they deserve to be in measure of their product. Funding a dedicated fab with TSMC could see them not having to get in line with such a limited resource as they are now.

Let's say that currently they can get X number of chiplets per month from a limited number of wafers, and of those, the yield is 30/40/20/10 on great, good, poor, and defective. It's just not enough to displace Intel substantially outside of tiny DIY, and due to needing return on margins with lower quantity, they can't even use price as pressure without hurting themselves for no reason. Now let's further hypothesize, fairly I think, that that same 7nm Fab is shared with other fabless entities who need their stuff produced as well.

Any respins or revisions will take some extra hassles in this kind of shared environment, but above all, I simply don't think the production is there.

If they have confidence that for the foreseeable future the Ryzen series will be defacto superior x86, then that *should* allow for them to entirely lead the market, with a little lag time for marketing and OEM liaison work, as well as expanding distribution etc. But saying 'hey TSMC, can you fulfill 8x the wafers next year?' is likely a non starter, but that's just scratching the surface of OEM level volume. For all of DIY by sales price (thanks to lots of expensive/overpriced stuff), by volume it's absolutely teeny tiny compared to the overall OEM and B2B market.

To give an example more directly, speaking to the manager at Micro Center, they have been getting around two dozen TOTAL 3900/3950X SKUs when they come through, if they're lucky enough to see them. I could guess that globally they're perhaps a few thousand 3900X, and a few hundreds of the 3950X by quarter. Compare to sales to Dell, HP, Lenovo, etc, where orders can be in the millions to tens of millions of SKUs.

 

[Arkaign]

Arkaign,

You are spot on about anything 4 core from Ivy on is fine for 99% of computer users. Excel, Word, light photo editing and video editing, really not a big deal. Even though I have some need for more compute, with my current Haswell rig I'm having a hard time justifying a $700+ upgrade for CPU, motherboard, RAM, video card, and probably power supply. I don't game so really any video card will do for me. That being said, as an enthusiast I'm aching for 8 cores or more.

I've always tried to plan my upgrades so that compute is at least doubled. P90 to Celeron 300a/450 to P3 850 to P4 3.06 to C2D E6400 to 2500k to 4770k.

Without AMD pushing Intel we would probably still be using quad core CPU's with maybe some 6 core $1,000 parts available if you really had to have it. Remember the days of the Intel eXtreme parts? Yikes.

Does AMD still have any fabrication capacity of their own? Did they sell them off?

Haha, oh man so true. It was exhausting seeing Intel rehash from really Lynnwood (i5-750 etc) onwards with every dang gen being basically

Pentium 2C/2T
i3 2C/4T
i5 4C/4T
i7 4C/8T

With most of the efforts and die space budget put towards ever expanding IGP, even when it made little sense. Talk about sitting on their hands counting profits. How long between i3-540 to i7-870 until Coffee Lake? Like 7+ years? Bananas.

And yes, AMD sold their fabs off back around 2006-2007 IIRC, around the ATI purchase.

 

[Markfw]

Arkaign,

You are spot on about anything 4 core from Ivy on is fine for 99% of computer users. Excel, Word, light photo editing and video editing, really not a big deal. Even though I have some need for more compute, with my current Haswell rig I'm having a hard time justifying a $700+ upgrade for CPU, motherboard, RAM, video card, and probably power supply. I don't game so really any video card will do for me. That being said, as an enthusiast I'm aching for 8 cores or more.

I've always tried to plan my upgrades so that compute is at least doubled. P90 to Celeron 300a/450 to P3 850 to P4 3.06 to C2D E6400 to 2500k to 4770k.

Without AMD pushing Intel we would probably still be using quad core CPU's with maybe some 6 core $1,000 parts available if you really had to have it. Remember the days of the Intel eXtreme parts? Yikes.

Does AMD still have any fabrication capacity of their own? Did they sell them off?

They sold it off. I think thats global foundries ?

 

[Arkaign]

They sold it off. I think thats global foundries ?

You are correct good sir, IIRC Dresden (new at the time of the purchase of ATI, and sale of fabs) was and is part of the AMD to GF handoff. I know some Saudis were involved as well.

 

[RetroZombie]

So can Intel pull a rabbit out of the hat? Unless Tiger Lake is to Ryzen is what Core2Duo was to the Athlon I think Intel is going to have some rough times ahead.

Just no.
Core 2 Duo was intels 30 Watts mobile cpu converted into one 65 Watts desktop cpu. They even glued two of them and made one 130 Watts powerfull desktop cpu.

Now where is the wonderful mobile intel chip to do the same? Thats right does not exist (unless you think atom is a 'thing').
And the foundry/process advantage they had at the time, today also does not exist...

Even if they come out with something really good, there is now a new problem that didn't exist at the time for them, which is amd zen 3 with a new architecture at a new process what will allow them to got from zen2 two quad core ccx inside one chiplet to only one eight core ccx in one chiplet.

But the problem for intel will not be zen 3 it will be zen4.

Zen 4 at tsmc 5nm will allow amd to double everything again, with eight cores with two ccx in one chiplet, so 16 cores per chiplet.

Intel even if can come with some amazing gains at ipc, they will have to keep up with:
- Same amount of cores, otherwise they lose in all multi-tasking and multi-threading scenarios.
- Clocks, or....
- Power usage, or....
- ....

From what we know from their roadmap or rumored designs I don't see them addressing any of those.

 

[Hulk]

Now where is the wonderful mobile intel chip to do the same? Thats right does not exist (unless you think atom is a 'thing').
And the foundry/process advantage they had at the time, today also does not exist...

You got it here in a nutshell. No magic chip behind the desk and serious process competition. I've been following AMD and Intel quite closely all the way back to high school ('80's) so it's been hard for me to wrap my head around the fact that they have been caught and arguably passed in both architecture and process. Except for a blip in the early 2000's they've been in complete control for 40+ years. Until now.

 

[Nereus77]

Intel was caught sleeping by Zen for sure, but they're waking up now. They need something special to close the gap between themselves and AMD and most people think Golden Cove might just do it. But that's a few years away and AMD has got some incredible momentum already.

Its going to be a rollercoaster decade in the CPU space.

 

[lobz]

Haha, oh man so true. It was exhausting seeing Intel rehash from really Lynnwood (i5-750 etc) onwards with every dang gen being basically

Pentium 2C/2T
i3 2C/4T
i5 4C/4T
i7 4C/8T

With most of the efforts and die space budget put towards ever expanding IGP, even when it made little sense. Talk about sitting on their hands counting profits. How long between i3-540 to i7-870 until Coffee Lake? Like 7+ years? Bananas.

And yes, AMD sold their fabs off back around 2006-2007 IIRC, around the ATI purchase.

Expanding their iGPU part made sense that time, because most of the time it was not enough even for the simplest tab switching without people who have seen a normal PC before, thinking that something's wrong with the machine. It was enough, sure. Office employees were told that this is the norm and if anyone complained, they are more than welcome to request a big personal bank loan and buy the whole company more expensive machines if they wanted fancy toys so bad. Spinning rust is the same principle but even worse than this. (Fancy toys is an actual wording I had to endure without facepalming my boss's boss once). Sure enough, if intel hadn't always expanded and improved the iGPU from generation to generation, they wouldn't always just have looked slow, they would have become unusable very fast.

Having unlocked and high-end dies designed and manufactured exclusively with iGPU nowadays is another thing. There are too many factors for me to draw a simple conclusion. I've read many pros and contras here, and my final thought is something like this: Intel as a company has wasted SOOOOOOO MUCH MONEY on asinine things, that designing 2-3 dies (i3/i5/i7+9) without GPU and either manufacture&sell it for a lot cheaper even with the same margin or putting on more cores for the same price, that could have coexisted with their normal CPUs. Not saying (coz I'm not sure) it would have been a game changer, but at least surely the opposite of asinine money wasting. We'll never know

 

[ElFenix]

anyone have a handy-dandy guide to intel's code names? really more interested in the CPU core code names than the retail product code names.

soooo many lakes

 

[cytg111]

Slightly OT but I cant fathom that AMD managed to turn this around... Bulldozer and GF disasters... for years. My money were on belly up. Glad I was wrong.

 

[Hulk]

anyone have a handy-dandy guide to intel's code names? really more interested in the CPU core code names than the retail product code names.

soooo many lakes

Resource & Documentation Center

Get the resources, documentation and tools you need for the design, development and engineering of Intel® based hardware solutions.

www.intel.com

And here is an unofficial complete list.

List of Intel codenames - Wikipedia

en.wikipedia.org

 

[ElFenix]

Resource & Documentation Center

Get the resources, documentation and tools you need for the design, development and engineering of Intel® based hardware solutions.

www.intel.com

And here is an unofficial complete list.

List of Intel codenames - Wikipedia

en.wikipedia.org

this still seems overly complicated.

from the wikipedia entry on skylake it says:
"Skylake CPUs share their microarchitecture with Kaby Lake, Coffee Lake and Cannon Lake CPUs "

so the next actual new microarchitecture (as opposed to shrinks and putting in better transistors in certain areas and better processor graphics) is sunny cove upon which ice lake and tiger lake(?) are based, correct?

the skylake and sunny cove bits are what i'm more interested in knowing (along with which retail parts those correspond to), and less interested in the kaby lake, coffee lake, ice lake, etc., part.

sorta like how i have no idea what _____ ridge means with AMD, but i know that ryzen 3000 desktop parts are zen 2.

 

[Arkaign]

this still seems overly complicated.

from the wikipedia entry on skylake it says:
"Skylake CPUs share their microarchitecture with Kaby Lake, Coffee Lake and Cannon Lake CPUs "

so the next actual new microarchitecture (as opposed to shrinks and putting in better transistors in certain areas and better processor graphics) is sunny cove upon which ice lake and tiger lake(?) are based, correct?

the skylake and sunny cove bits are what i'm more interested in knowing (along with which retail parts those correspond to), and less interested in the kaby lake, coffee lake, ice lake, etc., part.

sorta like how i have no idea what _____ ridge means with AMD, but i know that ryzen 3000 desktop parts are zen 2.

Haha, while it's certainly preferable to the alpha numeric soup of Intel product naming, even AMD managed to botch Ryzen. Zen2 can be 3000 or 4000 series, ugh.

 

[ElFenix]

Haha, while it's certainly preferable to the alpha numeric soup of Intel product naming, even AMD managed to botch Ryzen. Zen2 can be 3000 or 4000 series, ugh.

marketing: consumer confusion vs. admitting our mobile parts are 6 months behind our desktop parts

 

[IntelUser2000]

so the next actual new microarchitecture (as opposed to shrinks and putting in better transistors in certain areas and better processor graphics) is sunny cove upon which ice lake and tiger lake(?) are based, correct?

Skylake is the name of the SoC that makes up the whole thing. We don't know the code-name of the CPU core inside Skylake.

Sunny Cove is the name of the core inside Icelake.

Nearly nothing has changed in the Skylake derivatives. Not even the iGPU. Well, maybe once in Kabylake with the updated decoders but that's about it.

Tigerlake uses the Willow Cove cores.

To make it easy for you: Every lake except for Ice and Tiger uses the same core/gpu.

 

[RetroZombie]

Haha, while it's certainly preferable to the alpha numeric soup of Intel product naming, even AMD managed to botch Ryzen. Zen2 can be 3000 or 4000 series, ugh.

Well actually is quite easy, just follow the G.
(Assuming they wont add any zen1 to 4000 series)

 

[IntelUser2000]

Well actually is quite easy, just follow the G.
(Assuming they wont add any zen1 to 4000 series)

Shhh!!

You'll give marketers ideas!

 

[IntelUser2000]

marketing: consumer confusion vs. admitting our mobile parts are 6 months behind our desktop parts

*Shameless self promotion*

Info - List of modern Intel/AMD CPUs by code-name

Information is likely not complete. Please feel free to add in the comments section! Maybe I'll update, maybe I won't. :D AMD - Client Summit Ridge Ryzen 1xxx Zen 14nm N/A Desktop N/A Raven Ridge Ryzen 2xxx Zen 14nm Vega Desktop Mobility Whitehaven Threadripper 1xxx Zen 14nm N/A HEDT...

forums.anandtech.com

 

[aigomorla]

I dont think you realize how badly C2D slaughtered Athlon XP, and Bulldozer.

Not even Ryzen can be called a C2D in AMD terms, because C2D was just that far ahead of its time, especially with overclocking.

So im going with No....

Were not going to see another C2D, until probably we get self thinking chips, or Intel decides to go an entirely different route, like what happened with C2D after they said "lets go with "Dolthan and make it Yonah" vs continuing with "Prescott"

To be honestly, ever since it turned into a Lake, its been going downhill at a increasing price rate...
The last greatest chip i will say intel had is probably Haswell.

 

[IntelUser2000]

The last greatest chip i will say intel had is probably Haswell.

Haswell was needed for U laptops, but meh for desktops. Same with Ivy Bridge. 22nm was not a big deal at all, except for Atoms, which got quickly surpassed anyway. So all that work for no benefit. OK GPU(I expected more still), mediocre CPU. Skylake is also a disappointment. Nehalem is for server and another passable one on PCs.

I'd say the peak was Sandy Bridge. Still have my 2600K. It was a big advancement in all areas.

I don't want another Core 2. I want more steady advancements. That's what caused the ARM cores to catch up. Core 2/Sandy Bridge then what? Bunch of anemic advancements until panic?

Same with AMD.

 

[JoeRambo]

I'd say the peak was Sandy Bridge. Still have my 2600K. It was a big advancement in all areas.

Agreed, Sandy Bridge was a beast. Nehalem was basically Core2 with integrated memory controller bolted on ( while at same time reducing L2 by order of magnitude and having shared L3). SB was Intel's proper answer to AMD Athlon64 uArch, new architecture from ground up, taking best from Pentium3 and Pentium4 lineages, adding brutal vector FPU and AVX instructions. It kinda flew under radar cause Intel was in the lead and AMD's faithful were busy at Bulldozer altar.

EDIT: oh, and almost 5Ghz clocks with OC were epic too.

 

[guachi]

Here's Anandtech's article about the Core 2 Duo: The Empire Strikes Back.
The $316 2.40 GHz E6600 was consistently beating the fastest Athlon AMD had. And there were still two Core 2 chips faster. Tiger Lake won't pull this off.

Even Zen didn't. But I think Zen 3 might be the culmination of a slow-motion four generation Core 2-like wipeout.

 

[A///]

Meanwhile, Intel is still a mess and constantly delaying products. How can they beat AMD when they can’t even properly deliver Cooper/Ice Lake server/Comet Lake-S on time?

I'm curious how and if they'll deliver their super computer contracts. Mainly Aurora and not including unannounced ones....

To be honestly, ever since it turned into a Lake, its been going downhill at a increasing price rate...

Well a lot of historic lakes have dried up due to natural evolution in nature or human made errors...

 

[coercitiv]

Even Zen didn't. But I think Zen 3 might be the culmination of a slow-motion four generation Core 2-like wipeout.

AMD's Core 2 moment isn't some particular Zen arch, it's their execution. For the first time in history Intel is chasing a fast moving target.

 

[Hulk]

Let's say that currently they can get X number of chiplets per month from a limited number of wafers, and of those, the yield is 30/40/20/10 on great, good, poor, and defective. It's just not enough to displace Intel substantially outside of tiny DIY, and due to needing return on margins with lower quantity, they can't even use price as pressure without hurting themselves for no reason. Now let's further hypothesize, fairly I think, that that same 7nm Fab is shared with other fabless entities who need their stuff produced as well.

This is an insightful analysis. Intel has been the leader for a long time and have been able to deliver the superior product in the required numbers to many large buyers such as Dell, HP, Acer, Dell, Lenovo, etc... Beyond the objective qualifications for their partnerships (quality product and ability to deliver) there are also the subjective aspects such as buyers and sellers who have been working together for a long time.

The big question assuming AMD is able to hold the performance lead is exactly how many chiplets are or can be produced per month or quarter under the current arrangement?

Do these fabs operate 24 hours/day with workers in shifts? I would assume considering the capitol required to build them that they would.