Monday, November 26th 2018

14nm 6th Time Over: Intel Readies 10-core "Comet Lake" Die to Preempt "Zen 2" AM4

If Intel's now-defunct "tick-tock" product development cadence held its ground, the 14 nm silicon fabrication node should have seen just two micro-architectures, "Broadwell" and "Skylake," with "Broadwell" being an incrementally improved optical shrink of 22 nm "Haswell," and "Skylake" being a newer micro-architecture built on a then more matured 14 nm node. Intel's silicon fabrication node advancement went off the rails in 2015-16, and 14 nm would go on to be the base for three more "generations," including the 7th generation "Kaby Lake," the 8th generation "Coffee Lake," and 9th generation "Coffee Lake Refresh." The latter two saw Intel increase core-counts after AMD broke its slumber. It turns out that Intel won't let the 8-core "Coffee Lake Refresh" die pull the weight of Intel's competitiveness and prestige through 2019, and is planning yet another stopgap, codenamed "Comet Lake."

Intel's next silicon fabrication node, 10 nm, takes off only toward the end of 2019, and AMD is expected to launch its 7 nm "Zen 2" architecture much sooner than that (debuts in December 2018). Intel probably fears AMD could launch client-segment "Zen 2" processors before Intel's first 10 nm client-segment products, to cash in on its competitive edge. Intel is looking to blunt that with "Comet Lake." Designed for the LGA115x mainstream-desktop platform, "Comet Lake" is a 10-core processor die built on 14 nm, and could be the foundation of the 10th generation Core processor family. It's unlikely that the underlying core design is changed from "Skylake" (circa 2016). It could retain the same cache hierarchy, with 256 KB per core L2 cache, and 20 MB shared L3 cache. All is not rosy in the AMD camp. The first AMD 7 nm processors will target the enterprise segment and not client, and CEO Lisa Su in her quarterly financial results calls has been evasive about when the first 7 nm client-segment products could come out. There was some chatter in September of a "Zen+" based 10-core socket AM4 product leading up to them.
Source: HotHardware
Add your own comment

123 Comments on 14nm 6th Time Over: Intel Readies 10-core "Comet Lake" Die to Preempt "Zen 2" AM4

#101
RichF
bug said:
The thing I have against IGPs is the die size they take. If it was a tiny thing that offers a fallback in case your real GPU craps out, I'd be fine with it. As it is, IGPs are monsters that take 33-50% of the die space, depending on configuration. And that translates into $$$ (just look at Turing).
iGPUs also are bad because they exacerbate the heat concentration problem. As transistors shrink, cooling becomes more difficult because of surface area bottlenecking.

The one thing they're good for is reducing latency.

Since GPUs are so parallel, though, and because they require so many transistors — they're a poor fit when it comes to integrated high-performance 3D graphics.

Additionally, they eat into the surface area transistor budget for things like dummy material to reduce hot spots and duplicated resources for yields/harvesting.
Posted on Reply
#102
Vya Domus
GoldenX said:
o they are octa cores in an ALU sense, but quad cores in an FPU sense.
It just doesn't work like that, cores != ALU/FPU. A core is something that fetches , decodes , executes instructions and writes back the answer. Whatever those instructions may be and however that is accomplished, it doesn't matter. That's it, you can't mangle that definition in a million ways. A CPU can't act like an 8 core now and a 4 core in other situations, unless they are physically turned off.

A Zen core has 4 ALUs, does that mean a 1700 has 32 cores when you do instructions with integer operands ? Obviously, no.
Posted on Reply
#103
GoldenX
Vya Domus said:
It just doesn't work like that, cores != ALU/FPU. A core is something that fetches , decodes , executes instructions and writes back the answer. Whatever those instructions may be and however that is accomplished, it doesn't matter. That's it, you can't mangle that definition in a million ways. A CPU can't act like an 8 core now and a 4 core in other situations, unless they are physically turned off.

A Zen core has 4 ALUs, does that mean a 1700 has 32 cores when you do instructions with integer operands ? Obviously, no.
Hey, not my fault AMD decided to create a Frankenstein's monster with the FX. Check the performance for yourself.
What happened to Zero-Two?
Posted on Reply
#104
RichF
GoldenX said:
The 8 and 9 series FX have 4 cores (modules, as they were called), so they are octa cores in an ALU sense, but quad cores in an FPU sense.
This sentence is self-contradictory. Here are the premises it contains:

1) The 8 and 9 series FX have 4 cores (modules, as they were called)
2) they are octa cores in an ALU sense
3) quad cores in an FPU sense

Firstly, a module is not the same thing as a core. Secondly, it's impossible, as Vya Domus said (and I was about to), for a CPU to be both 8 and 4 cores simultaneously.

bridgmanAMD
there is a single FPU per module but it has two independent 128-bit FMAC pipes to allow executing two instructions (one from each thread) in parallel. So arguably each module has two FPUs when running 128-bit instructions and one FPU when running AVX-256 instructions (or MMX instructions).
AMD_James
The FPU is able to process two 128-bit FP threads simultaneously. It combines into a single unit to process 256-bit operations. Either core in the module can dispatch instructions to the FP unit said:
the "only has 4 fpus" situation only really applies on 256 bit operations. In 128 bit and lower operations it is a split unit and can do two separate instructions at the same time. Modules also have a single L2 cache, and that is part of AMD betting on CMT instead of SMT
Kromaatikse
It's not hyperthreading. That's Intel's trademarked brand name for SMT (Simultaneous Multithreading). It's not generic SMT said:
Since Bulldozer shipped, AMD has made modest improvements to the CPU’s overall efficiency and performance. Kaveri cut the penalty for multi-threading in half, from ~20% to 10% compared with typical core scaling. If AMD hadn’t been forced to lower clock speeds to compensate for its 28nm manufacturing process, Kaveri would’ve outperformed Richland across the board. Bulldozer is absolutely capable of executing eight threads simultaneously, and executing eight threads on an eight-core FX-8150 is faster than running that same chip in a four-thread, four-module mode. Bulldozer can decode 16 instructions per clock (not eight) and it can keep far more than eight instructions in flight simultaneously.
GoldenX said:
Hey, not my fault AMD decided to create a Frankenstein's monster with the FX. Check the performance for yourself.
If you want a much more Frankensteinian design, take a look at Broadwell-C.

GPU integrated
separate eDRAM chip
SMT

Bulldozer/Piledriver are much simpler designs. Elegance and simplicity don't speak to performance. The Frankensteinian L4 on Broadwell-C certainly didn't hurt its performance, nor did the inclusion of an IGP or SMT. Broadwell-C was quite impressive considering its low clocks and power consumption. In fact, Peter Bright of Ars complained that Intel was robbing people of performance by refusing to put that L4 on Skylake.

Bulldozer/Piledriver had poor IPC due to a combination of being forced to last much longer in the market than their Intel counterparts as well as design inefficiencies that may have been able to be overcome with strong further development. One of those was AMD's bet to make a pipeline that was very deep. Deep and narrow failed with the P4 but AMD tried that strategy, although adjusted, again. Had AMD devoted the kind of resources to improving their CMT design that Intel devoted to creating Sandy the performance would have been better. Better cache performance (e.g. Piledriver's L3 isn't so much better than fast DDR3, as far as I recall). Better instruction caching and prediction. Possibly a shallower and wider design. Possibly the addition of SMT to supplement the CMT. Possibly better Windows scheduling and optimization. Also, had Intel not created Sandy and had, instead, created a weaker CPU — then perhaps AMD would have decided to create a successor to Piledriver and it could have been quite improved.

How many years have we seen people brag in forums about still using their Sandy chip because it was such a great value? There are three reasons for that. First is that AMD didn't compete well. Second is that Intel slowed down IPC gain. Third is that Intel got rid of solder. However, how much code optimization has to do with performance remains an open question. The game Deserts of Kharak showed Piledriver hanging in there quite nicely, despite being well out of date. Many other games showed terrible FX performance.



What is the special sauce in Deserts of Kharak that enable the FX to hang in there with Haswell? We never found out because it was basically an outlier. It would be interesting for someone to interview the developers to discover why their code ran so well on FX. The most basic assumption is that it, unlike other games of the time, did a much better job of leveraging the 8 cores. However, other factors are most likely also in play, like a lack of a DRAM speed bottleneck.

Personally, I think it's much more interesting to know what is possible with CMT than it is to complain about AMD's failure with it.
Posted on Reply
#105
GoldenX
I still think it was ahead of it's time.
Adding to the problems AMD had, being stuck on 32/28nm didn't help with the development of the FX arch, and Windows, as always, was very slow to adapt, a problem Linux didn't have.
Posted on Reply
#106
Prima.Vera
Manu_PT said:
bla bla bla
Kid, what are you, 12? Calm down. Gues what? I can play CS at 250FPS on my 100Hz just fine, I don't need a 240Hz monitor for that. The bending it's NOT that distracting or serious as you might believe, relax.
Posted on Reply
#107
Midland Dog
Vya Domus said:
The FPUs are shared between what again ? :roll:



No, they don't. Each core has 2 ALUs, look this stuff up before you even try to have a shot at arguing what is a fake core and what isn't.
tru, i thought each core shared an fpu, i remember reading something like that somewhere, probably getting the names of stuff muddled up my bad
Posted on Reply
#108
Chloe Price
GlacierNine said:
As I mentioned earlier, when Pentium 4 was the thing, the fastest pentium 4s were essentially the HEDT CPUs of their day, and they were as expensive as they were, because Intel didn't have a separate HEDT platform with different chips on it.

If you compare the prices of what most gamers were actually buying back then, to the prices of the top end, and then you compare the modern equivalents to those, you'll see that "Mainstream" has jumped in price by ~63% in the last 7 years and "HEDT" has jumped by ~88%, while the $ itself has only risen by ~12%
Excluding P4EE, they were the same chip with just a higher multiplier. Some of the cheapest models lacked HT, but a cheaper model was easy to overclock to similar (or even better since the OC'd FSB) performance as the flagship ones.

What I meant, when thinking about the prices of hardware back then, you couldn't get the high-end version at 600eur/usd.

Manu_PT said:
Reason why Im still with Intel is because I use 240hz monitor and cant even look back to 144hz. Ryzen could not sustain those 160-200 fps on most games. But if Zen2 can do it I will jump to 8 core finally.
I don't see any difference between monitors, so I rather buy beer than use that money for something which doesn't give me any benefits. Drinking beer gives. :toast:
Posted on Reply
#109
GlacierNine
Chloe Price said:
Excluding P4EE, they were the same chip with just a higher multiplier. Some of the cheapest models lacked HT, but a cheaper model was easy to overclock to similar (or even better since the OC'd FSB) performance as the flagship ones.

What I meant, when thinking about the prices of hardware back then, you couldn't get the high-end version at 600eur/usd.
Yes, and what I'm saying is that because Intel had yet to segment their proucts into "Mainstream" and "HEDT", P4EE was not the equivalent of a 6700K, 7700K, 8700K, 9900K etc. It was the equivalent *within intel's product stack* of a 7980XE, 9980XE, etc. That is to say, it had more raw power than any other Intel Chip and they priced it at the top of their stack because of it.

To compare a P4EE to a 9900K in terms of where in the product stack they are is simply wrong - the 9900K is not Intel's single most capable consumer product. the 9980XE is, and is therefore the comparable CPU in 2018.
Posted on Reply
#110
Adam Krazispeed
what???? 240hz panel cant run in ryzen??? u on drugs. i only have a Ryzen Threadripper 8 core 16 thread 1900X on a Gigavyte Arous Gaming 7 X399 MB and run at DDR4 3400Mhz Wuad Channel an dthe 1900X All cores @ 4.0 GHZ and i have absolutely no problems runnin 240Hz????

now on a GTX 1070, no, i cant... on a1080..... No either.... but a 1080Ti, yes, i can,,, my 1080 can only do max of may be 144-200Hz or 144- 200 Frames per secopnd @ 144-200Hz, ( i set the RR a it lower but on a weaker GPU... the cpu i have handles it fine, but my AMD TR system is Heavily TUNED



Manu_PT said:
Maybe another cpu that gets 90 degrees on a 100€ AIO and costs 600€. Is starting to get boring.

I just hope AMD can finally improve IPC/clocks and gaming performance (yes I know someone will jump and say that is only a 5% difference when is not but whatever) and then Intel is officially done unless they start selling products with decent pricing. 500€ for a 8 core no HT cpu in 2018 is unacceptable.

Reason why Im still with Intel is because I use 240hz monitor and cant even look back to 144hz. Ryzen could not sustain those 160-200 fps on most games. But if Zen2 can do it I will jump to 8 core finally.

This might seem dumb for you, but as a curiosity, when I tell my mom that my "dream" cpu costs 500€ (she knows I love computers), she doesnt believe it. Is completly outrageous. I think some ppl dont even think about what 500€ are worth. Same goes for those dudes pumping 1300€ for a video card to play videogames. Mental. Im on this hardware stuff for almos 2 decades and I never seen high end pc components being so premium. I have money to buy those products but mentally I would never accept paying that kind of cash. Makes no sense.

And even if you want a solid gpu like 1070 or 1070ti you spend 400€-500€ and already need to lower a lot of details at 1080p on recent games like shadow of the tomb raider or black ops4.
i just dont think people that are used to using INTEL SETUPS, Dont understand how Ryzen works....

Iv used AMD for years, most of my life actually ( back to the AMD 386.... even the K6-2 550 Super socket 7 CPUs, any one remember those lol, but besides iv been working with intel.and AMD back in the Pentium MMX days (JESUS I been doing this too damn long... but iv had intel back to the 8088/ then 8086, the first x86 CPU the 8086.... 2Mhz. wow that was sloww.. 286 8mhz , then 12m 16, to 20mhz then 386 + to pentium. pentium pro MMX Pentium II 3 and so on... i liked it better when AMD/Intel used the same DAMN CPU SOCKET???WTF ever happend to that..


but my point is,, if you been with intel for the last 10+ yrs, You wont know AMD's cpus as much as they ARE very different, AL LEAST MY AMD CPUs ARNT AFFECTED BY MELTDOWN, and im protected form Spectre now with no performance penalty

i even have an
Intel Core i7-6950X Extreme Edition (old friends PC when he upgraded to ryzen threadripper 16core) (he gave me the system, because it stopped working but i fixed it) MB failed... :(


that i cant even get to run @ 240Hz on those panels??? but my 1900x AMD Ryzen threadripper can, but i messed with he ryzen bios scence march 2017 launch day
Posted on Reply
#111
hat
Enthusiast
E-curbi said:
Can't wait to overclock a new Intel mainstream-enthusiast 10-core on a ROG Stryx board. lol :roll:

No wait, how about a TUF board 10-cores 20-threads. :roll:

The ROG forum should be fun to watch.

"Waddaya mean it's only a 2-phase VRM?" :D
Looks like board manufacturers are gonna have to step up their game again. Thought it's disappointing any HEDT board was frying VRMs...
Posted on Reply
#112
gmn 17
10000K series here we come
Posted on Reply
#113
GoldenX
gmn 17 said:
10000K series here we come
That's the series name or the absolute temperature they reach?
Posted on Reply
#114
hat
Enthusiast
GoldenX said:
That's the series name or the absolute temperature they reach?
Only on Pluto.
Posted on Reply
#115
bajs11
m4dn355 said:
How come intel's lake never dry up?!


even if it does
they'll just name it Dry Lake
thatll be the name of the 14nm refresh∞
Posted on Reply
#116
Chloe Price
GlacierNine said:
Yes, and what I'm saying is that because Intel had yet to segment their proucts into "Mainstream" and "HEDT", P4EE was not the equivalent of a 6700K, 7700K, 8700K, 9900K etc. It was the equivalent *within intel's product stack* of a 7980XE, 9980XE, etc. That is to say, it had more raw power than any other Intel Chip and they priced it at the top of their stack because of it.

To compare a P4EE to a 9900K in terms of where in the product stack they are is simply wrong - the 9900K is not Intel's single most capable consumer product. the 9980XE is, and is therefore the comparable CPU in 2018.
I know what you mean, and it's hard to point out what I'm meaning since English isn't my native language, and I can't put my thoughts into words. :D
Posted on Reply
#117
Midland Dog
Midland Dog said:
tru, i thought each core shared an fpu, i remember reading something like that somewhere, probably getting the names of stuff muddled up my bad
i retract my statement, you are completely wrong, 2 alus to a module and 1 fpu to a module, look up a bulldozer block diagram
Posted on Reply
#118
Vlada011
For enthusiast with DDR4 no reason to think about upgrade before DDR5.
Posted on Reply
#119
bug
Vlada011 said:
For enthusiast with DDR4 no reason to think about upgrade before DDR5.
Why not? If I could get 8-12 cores for $200-250, I'd try it just out of curiosity.
Posted on Reply
#120
GlacierNine
bug said:
Why not? If I could get 8-12 cores for $200-250, I'd try it just out of curiosity.
Assuming he means no reason to upgrade to Intel, I'd say it's because assuming someone built a Skylake or Zen 1 system that fit their use case, and as long as that use case hasn't changed, there's really no cost-effective upgrade they can make right now to do the same workload better.

For people with Ryzen 1, a 2700X is a chunk of change for not very much uplift in heavy multitasking or parallel workloads. Sure, build a new system with one, but upgrading from Ryzen 1 to that? Not really worth the time and cost.

For people with a Skylake or Kaby Lake chip, those are still extremely capable as gaming chips, and will handle 144Hz 1080p or 1440p competitive titles just fine even at some very respectable settings. Upgrading to an 8700K or 9900K isn't really a big improvement either, and that's even without factoring in a new motherboard and inflated pricing for Intel chips.

If DDR5 comes along and people are upgrading to that platform, then yes it's even more expensive once again, but it'd be a good time for people to throw out old platforms and upgrade, because they'd be getting all of this in one move:

- An upgrade path (They don't get this on Z390)
- Likely a new node of manufacture for their CPU (7nm for AMD, and we presume Intel will have 10nm by the time DDR5 comes)
- A new RAM technology (Which is a great time to sell old sticks and buy new ones that will last you through potentially a couple systems, saving money)
- Almost certainly significantly improved architectures, being either Zen 3 or a *real* Intel response to Zen architecture, rather than just more rehashes of Skylake.

For people who bought high-end rigs and who's need is staying the same, there's simply not really much reason to upgrade from Skylake or Ryzen 1. All the things those systems were good at, they're still extremely good at. And if they didn't buy high end rigs, well, they have an upgrade path within their own platform - they can buy an 1800X or a 6700K to replace their 1500X or 6400 and that will be much better price/performance uplift than investing in an entirely new platform.
Posted on Reply
#121
Vlada011
bug said:
Why not? If I could get 8-12 cores for $200-250, I'd try it just out of curiosity.
Great, I hope you will use them 24/7 together with HT Link, than you could enjoy nicely.
You can buy and Threadripper with 16 cores for 500$ and you got nice 1600X for most of the time, excellent.

I talk about normal improvements, about next gen core and 30-40% difference core vs core, 50% faster memory, etc...
And I could say WOW I want i9-9820X, 10C less than 1000$... but real difference between them and my CPU is how much is better on up to 4-6 cores, when difference is good in that situation we can talk than about even bigger improvement in core numbers. But only core number is nothing.
Posted on Reply
#122
bug
Vlada011 said:
Great, I hope you will use them 24/7 together with HT Link, than you could enjoy nicely.
You can buy and Threadripper with 16 cores for 500$ and you got nice 1600X for most of the time, excellent.

I talk about normal improvements, about next gen core and 30-40% difference core vs core, 50% faster memory, etc...
And I could say WOW I want i9-9820X, 10C less than 1000$... but real difference between them and my CPU is how much is better on up to 4-6 cores, when difference is good in that situation we can talk than about even bigger improvement in core numbers. But only core number is nothing.
If you'll notice, I currently being served perfectly fine by a quad core. So when I said I would get more cores out of curiosity, I meant just that.
Posted on Reply
Add your own comment