Intel 10th Gen Core "Comet Lake" Lineup and Specs Revealed

[btarunr]

Ahead of a possible reveal in the sidelines of CES, followed by an early-Q2 2020 product-launch, company slides detailing Intel's 10th generation Core desktop processors in the LGA1200 package, codenamed "Comet Lake-S," leaked to the web courtesy Informatica Cero. They confirm that HyperThreading will play a key role, with Intel enabling it across the lineup. The range-topping Core i9 series will be 10-core/20-thread along with 20 MB of L3 cache. The Core i7 series will be 8-core/16-thread, along with 16 MB L3 cache. The all-important Core i5 series will be 6-core/12-thread, equipped with 12 MB of L3 cache. The Core i3 series will have two sub-tiers: i3-103xx series with 4-core/8-thread and 8 MB L3 cache; and i3-101xx series 4-core/8-thread with 6 MB L3 cache.

The Core i7 and Core i9 "Comet Lake" chips will feature native support for dual-channel DDR4-2933, while the Core i5 and Core i3 will make do with native DDR4-2667 support (memory overclocking possible). Besides core/thread counts, and cache size increases, Intel will dial up clock speeds across the board by as much as 300 MHz per SKU (vs. their 9th gen predecessor), and introduce Turbo Boost Max 3.0, which has been exclusive to its HEDT processors. The introduction of Turbo Boost Max 3.0 could also bring about modern favored-core capability (benefiting Windows 10 1909 and later). The classic Turbo Boost is also available. There's also a mysterious new feature called "Thermal Velocity Boost," with its own set of clock-speeds depending on core/thread load. The chips could also feature Modern Standby C10 power-state support (first to the desktop platform). Intel is said to have also added several new core and memory overclocking features on the K-SKUs.



Built on a 14 nm-class silicon fabrication node, and featuring the same IPC as "Skylake," the 10th generation Core "Comet Lake" series will rely on aggressive power-management to sustain 65 W TDP rating for most SKUs, but Intel's virtual barrier for 95 W as the TDP number for unlocked K SKUs ends with the 10 generation (the i9-9900KS already breaks that). The 10th generation Core K SKUs have a scorching 125 W TDP rating not just for the 10-core i9-10900K and 8-core i7-10700K, but also the 6-core i5-10600K. The table above details the various SKUs we could make out from the low image-quality slide screenshots.



View at TechPowerUp Main Site

 

[btarunr]

Imagine paying upward of $400 for a locked processor in 2020.

 

[davideneco]

Z490 only MB with oc

nice shintel

 

[Ferrum Master]

Imagine paying upward of $400 for a locked processor in 2020.

I still don't understand the need of having an iGPU in leading SKU's. Lack of pcie4. More or less... it is more of the same again, again and again.

 

[TesterAnon]

And they are still selling locked CPUs, they just never learn.

 

[Frick]

I still don't understand the need of having an iGPU in leading SKU's. Lack of pcie4. More or less... it is more of the same again, again and again.

You need a spare GPU otherwise.

 

[londiste]

And they are still selling locked CPUs, they just never learn.

In this case, locked CPUs might be a good thing given fixed PLLs. 6 or 8 Skylake cores will generally go just fine to 5.0 GHz or more with power consumption that comes with it. Doing that on any cheap motherboard will bring VRM quality to the light.

AMD does not quite have that problem right now. Temperature or reasonable clock ceiling is fairly close to stock clocks.

 

[Crackong]

lols the Thermal velocity Boost

Are we getting a new kind of boost every year?

2021 - Vector Unidirectional Boost
2022 - Space Continuum Boost
2023 - Quantum Entanglement Boost

 

[londiste]

IIRC Thermal Velocity Boost is a thing since Coffee Lake but primarily (or only) on mobile CPUs.
Tl;dr - it is an automatic 100-200MHz boost to clock speeds if CPU temperature is under a certain threshold which I think was < 70°C. Provided it fits into power budget of course.

When Ryzen 3000 came out and its boost behaviour was explored and detailed, there was a comment in Reddit that basically said Intel could technically do the same with their CPUs - feed 1.5V into a core and let it boost to 5.2-5.3 and this should be doable in pretty much any Skylake-derived core. Regardless of the marketing names on the features the table above seems to show they are starting to do exactly that.

 

[Melvis]

Clock speeds look impressive but how in the world are you going to cool these 10Core CPUs? Going to be a furnace! (Must be cooled with a 360m RAD)

 

[john_]

Intel 2019: Hahaha, AMD's 3000 CPUs have Turbo speed issues.

Intel 2020: We can warranty NOTHING. "Up to" even for the base frequencies.

 

[londiste]

i9-10900K with 10 cores at 3.7GHz might actually fit into 125W. Kind of curious about how that turns out for AVX2 load though.
65W TDP version i9-10900 and 10 cores at 2.8GHz in 65W is a bit more doubtful but still realistic enough.

I mean, this meshes well with what we know about Intel current CPUs. i9-9900K can do 3.6GHz on 8 cores at 95W, i9-9900GHz can do 3.1 at 65W.

 

[Zubasa]

Intel 2019: Hahaha, AMD's 3000 CPUs have Turbo speed issues.
Intel 2020: We can warranty NOTHING. "Up to" even for the base frequencies.

Maybe M$ will finally fix their Windows scheduler and thread migration.

 

[john_]

Maybe M$ will finally fix their Windows cheduler and thread migration.

I don't think this is a M$ problem. When I see "Up to" for the base frequencies, in my mind it means that even there the chip will be on it's limits. Have a hot day, a case without good air flow, or a good enough, but not great cooling solution and probably the base frequency will be lower than 3.7GHz.

 

[yeeeeman]

They should've unlock all the models, that is one way I could see people still going for these CPUs in 2020. Given the limitations, some of these are DOA unfortunately.

 

[Zubasa]

I don't think this is a M$ problem. When I see "Up to" for the base frequencies, in my mind it means that even there the chip will be on it's limits. Have a hot day, a case without good air flow, or a good enough, but not great cooling solution and probably that 3.7GHz frequency will be lower.

Windows does have the habit of shoveling the work load around the threads faster than the boost algorithm can keep up.
It is true even on 1909, the difference is Windows now juggles the workload within a CCX / CCD for AMD.
As for Intel I am not sure, but most people that got the 9900K just manual OC all the cores so no one really cared about how the boost worked.
For X299 most people got it for multi-threaded work loads so again no one really cared about the favorite core boost.

 

[john_]

They should've unlock all the models, that is one way I could see people still going for these CPUs in 2020. Given the limitations, some of these are DOA unfortunately.

Unlocking the processors will only make a difference in retail market and I don't think Intel cares about the retail market. It cares about the OEM market and there that 5.3GHz will sell much more CPUs than an unlocked multiplier.

Windows does have the habit of shoveling the work load around the threads faster than the boost algorithm can keep up.
It is true even on 1909, the difference is Windows now juggles the workload within a CCX / CCD for AMD.
As for Intel I am not sure, but most people that got the 9900K just manual OC all the cores so no one really cared about how the boost worked.
For X299 most people got it for multi-threaded work loads so again no one really cared about the favorite core boost.

I am not looking at the boost speed. I am looking at that "Up to" for the base frequencies.
If I am not mistaken, this is something completely new, not being able to warranty even the base frequencies for a new CPU.

 

[Zubasa]

I am not looking at the boost speed. I am looking at that "Up to" for the base frequencies.
If I am not mistaken, this is something completely new, not being able to warranty even the base frequencies for a new CPU.

The base clock thing has been going on for a while on X299 side.
Intel used to state a seperate AVX clock for their higher core count CPUs.
For the Skylake-X refreshes Intel just dropped that information entirely, because it got to the point that Intel cannot really guarantee a clock speed under AVX work loads.
The whole cluster F about MCE, TDP limits and boost duration etc didn't help either.

 

[Vya Domus]

I am amazed they still don't want to give up the iGPU even in the highest end 10 core models. They are squeezing this node to death but they are still willing to make the yields and binning worse by having it there.

 

[londiste]

I am amazed they still don't want to give up the iGPU even in the highest end 10 core models. They are squeezing this node to death but they are still willing to make the yields and binning worse by having it there.

It is a competitive advantage even if it is a small one. Given the die sizes it does not cost them all that much. Quicksync is a thing for some people, being fine without dGPU is a definite plus in some situations and especially so for OEM machines.

 

[Vya Domus]

Given the die sizes it does not cost them all that much.

The size is considerable, there is enough real state for 4 more cores. To get these to clock upwards of 5.2 Ghz they need some heavy binning, this silicon is simply dead weight. The 10 core parts will be released in dismally low numbers.

Quicksync is a thing for some people, being fine without dGPU is a definite plus in some situations and especially so for OEM machines.

Who are we kidding ? No OEM will use the unlocked parts and interestingly enough the 9400F survived just fine without the iGPU and media components.

 

[londiste]

Who are we kidding ? No OEM will use the unlocked parts and interestingly enough the 9400F survived just fine without the iGPU and media components.

Even Intel does not have that many different dies. 9000-series K CPUs (including 9600K) are all 8-core dies. Other than that Intel does have 4 and 6-core dies but these are more likely thanks to shared mobile use. Separating this further to with and without iGPU would not make any sense. If they want something like 10900 with iGPU then it has to be on die and everything else is cut down from it.

9400F does have iGPU on die, it is just disabled. I have been wondering about which die is inside 9400F but it is a locked mainstream CPU and not many enthusiasts care. The few delid images there are seem to show it is (sometimes a new stepping of) 6-core silicon.

The size is considerable, there is enough real state for 4 more cores. To get these to clock upwards of 5.2 Ghz they need some heavy binning, this silicon is simply dead weight. The 10 core parts will be released in dismally low numbers.

Other than power allocation, iGPU does not factor into binning. It is on a separate voltage and frequency plane. Memory Controller is also separate enough that it does also not care if iGPU is there. The only problem is "dead" die cost, as you said.

 

[Vya Domus]

iGPU does not factor into binning.

You can be convinced that it does. Say all your cores did 5.2 Ghz but the iGPU has a defect, into the recycling bin it goes.

 

[londiste]

You can be convinced that it does. Say all your cores do 5.2 Ghz but the iGPU has a defect, into the recycling bin it goes.

Judging from 9000-series CPUs - into storage it goes and after quarter or two a 10900KF SKU is born

Edit:

The size is considerable, there is enough real state for 4 more cores.

True, but it does not matter. There is no more power budget on 14nm, even 10 cores is stretching it.

 

[pjl321]

Is it definitely the '630' iGPU's and not the 730?