Intel 10 nm Ice Lake is Alive: Server and Desktop Support Added to the Linux Kernel

[AleksandarK]

There were many rumors about Intel's 10 nm CPUs, many of them indicating that Intel will not manufacture 10 nm CPUs for desktop users, due to the 10 nm manufacturing process being in a bad shape. Those rumors were later countered by Intel, claiming that 10 nm is doing very well on improving yields and that we will see desktop CPUs based on the new node very soon.

Thanks to the Linux kernel mailing list (LKML), we now know that support for Ice Lake desktop and server CPUs has been added. A Patch titled "Add more CPU model number for Ice Lake" has many details about variants of Ice Lake with names like Ice Lake X for server Xeon CPU, Ice Lake D for Xeon D CPUs, Ice Lake L for mobile, and regular Ice Lake for desktop series of CPUs. This confirms Intel's claims that Ice Lake is on its way to desktop and server users in the near future. Possible launch date on these CPUs would be sometime in 2020, when Xe graphics cards are launched in July/August, so Intel could bundle both processors on the same 10 nm node.



View at TechPowerUp Main Site

 

[john_]

I guess after 4 years hitting that 10nm wall with their head, they are so dizzy, they think they are finally breaking the wall.

They should have chosen to move faster to 7nm, but nevermind. I like seeing AMD having plenty of time to fully recover. Only then we will have true duopoly in the market (which is not great, but much better than monopoly).

 

[Kaotik]

There's no mention of "-S" models, which is the actual "desktop line", and none of Intel roadmaps includes one for Ice Lake either, so the "desktop" probably means something else than most think
edit:
Intels next desktop chip should be Comet Lake next year followed by Rocket Lake, which supposedly has Xe-based graphics (possibly as separate die)

 

[SL2]

They should have chosen to move faster to 7nm

I'd guess that's easier said than done, and even easier said in hindsight.

I'm not defending Intel, I just think the development of a new process node is hard to speed up. Besides, I don't think Intel wants to take any risks with 7 nm.

There's no mention of "-S" models, which is the actual "desktop line", and none of Intel roadmaps includes one for Ice Lake either, so the "desktop" probably means something else than most think

Mind you, there's no INTEL_FAM6_COMETLAKE_S either...

Edit: I can't find any _S models listed.

Source: https://github.com/torvalds/linux/b...301faeebcd2cb3cd/arch/x86/events/intel/core.c

static const struct x86_cpu_desc isolation_ucodes[] = {
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL,         3, 0x0000001f),
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL_L,         1, 0x0000001e),
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL_G,         1, 0x00000015),
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL_X,         2, 0x00000037),
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL_X,         4, 0x0000000a),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL,         4, 0x00000023),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_G,         1, 0x00000014),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_D,         2, 0x00000010),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_D,         3, 0x07000009),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_D,         4, 0x0f000009),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_D,         5, 0x0e000002),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_X,         2, 0x0b000014),
    INTEL_CPU_DESC(INTEL_FAM6_SKYLAKE_X,         3, 0x00000021),
    INTEL_CPU_DESC(INTEL_FAM6_SKYLAKE_X,         4, 0x00000000),
    INTEL_CPU_DESC(INTEL_FAM6_SKYLAKE_L,         3, 0x0000007c),
    INTEL_CPU_DESC(INTEL_FAM6_SKYLAKE,         3, 0x0000007c),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,         9, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE_L,         9, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE_L,        10, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE_L,        11, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE_L,        12, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,        10, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,        11, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,        12, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,        13, 0x0000004e),
    {}
};

 

[birdie]

This commit means absolutely nothing. Tomorrow they may revert it.

 

[efikkan]

This commit means absolutely nothing. Tomorrow they may revert it.

That's nonsense.
This is pretty much the strongest type of evidence we can have of an unreleased product at this time. Of course even Intel can't know for sure until they get the final stepping back from the factory, but this is evidence that they intend to release the product, and probably have early engineering samples for testing.

Compared to most of the source material for "news" in the tech press, with all kinds of BS pulled right out of thin air by garbage news sites or random YouTubers, this news is actually pretty solid stuff.

 

[birdie]

That's nonsense.
This is pretty much the strongest type of evidence we can have of an unreleased product at this time. Of course even Intel can't know for sure until they get the final stepping back from the factory, but this is evidence that they intend to release the product, and probably have early engineering samples for testing.

Compared to most of the source material for "news" in the tech press, with all kinds of BS pulled right out of thin air by garbage news sites or random YouTubers, these news are actually solid stuff.

Your comment and attitude are nonsense. I've been following Linux development since the mid-'90s when you were probably not yet born. Commits are commits. They are not an indication of any commitment or real product launches. Also, considering the fact that Ice Lake U CPUs are far and between and there are no parts which run faster than 3.9GHz, it takes a lot of wild imagination and belief to think that Intel has suddenly solved their 10nm node and are ready to launch CPUs which are undoubtly faster than Coffee Lake CPUs.

Slow down and calm down, dude, and stop talking from your ar*e. Thanks.

 

[SL2]

it takes a lot of wild imagination and belief to think that Intel has suddenly solved their 10nm node and are ready to launch CPUs which are undoubtly faster than Coffee Lake CPUs.

Nobody said anything about higher frequency. You did.
The 5775C didn't break any stock clock speed records for pretty much the same reason, but it still exists.
Remember the first 10 nm from last year, Cannon Lake? Max turbo was 3.2 GHZ on a low power 2C and disabled GPU. Ice Lake mobile is a big step forward and severs a purpose among laptops, it might do the same for desktops even if it's not a 9900K successor, Comet Lake will handle that (unfortunately).

when you were probably not yet born.

Slow down and calm down, dude, and stop talking from your ar*e. Thanks.

Easy now. You may be older, but you don't act like it.

 

[efikkan]

Your comment and attitude are nonsense. I've been following Linux development since mid 90s when you were probably not yet born.

Congratulations, you've already managed to embarrass yourself and prove you can't make a well formed argument, and have resorted to personal insults. Well done.
And P.S. I managed to learn coding in the mid 90s, turbo pascal if that rings any bells, so please stay civil and polite or go play somewhere else.

Commits are commits. They are not an indication of any commitment or real product launches.

If you read my post properly, you would have seen the answer to that, instead of rushing to insult.

 

[birdie]

Congratulations, you've already managed to embarrass yourself and prove you can't make a well formed argument, and have resorted to personal insults. Well done. And P.S. I managed to learn coding in the mid 90s, turbo pascal if that rings any bells, so please stay civil and polite or go play somewhere else.

If you read my post properly, you would have seen the answer to that, instead of rushing to insult.

Your comment contains zero facts yet you were talking as if you were damn sure about it. I didn't insult you, I was insulted by the level of confidence you had based on the three lines of code.

Intel promised 10nm based CPUs for 2015 (!!!) and now when year 2019 is almost over, we still don't have a working 10nm node. Base CPU clock rates for Ice Lake CPUs are just laughable: https://en.wikipedia.org/wiki/Ice_Lake_(microprocessor)

I don't know what you choose to believe in but I believe something when I see it. For previous nodes Intel at least used to show yet to be released products in advance - they haven't done that for their 10nm node. Nothing at all. If I'm writing shat, you can refer to it as "shat" pure and simple. And I won't say "someone has insulted me"- instead I will take note and won't try to behave like I'm the smartest soothsayer to ever exist.

If you are willing to speculate about new nodes and Intel uArchs go to Anandtech as I'm done here.

Speaking of embarrassment - I have nothing to be embarrassed about. I was honest and straightforward and I didn't try to make things up, unlike you. And you are still trying to defend your comment without providing a single fact - just speculation on top of speculation.

 

[Kaotik]

I'd guess that's easier said than done, and even easier said in hindsight.

I'm not defending Intel, I just think the development of a new process node is hard to speed up. Besides, I don't think Intel wants to take any risks with 7 nm.



Mind you, there's no INTEL_FAM6_COMETLAKE_S either...

Edit: I can't find any _S models listed.

Source: https://github.com/torvalds/linux/b...301faeebcd2cb3cd/arch/x86/events/intel/core.c

static const struct x86_cpu_desc isolation_ucodes[] = {
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL,         3, 0x0000001f),
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL_L,         1, 0x0000001e),
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL_G,         1, 0x00000015),
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL_X,         2, 0x00000037),
    INTEL_CPU_DESC(INTEL_FAM6_HASWELL_X,         4, 0x0000000a),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL,         4, 0x00000023),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_G,         1, 0x00000014),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_D,         2, 0x00000010),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_D,         3, 0x07000009),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_D,         4, 0x0f000009),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_D,         5, 0x0e000002),
    INTEL_CPU_DESC(INTEL_FAM6_BROADWELL_X,         2, 0x0b000014),
    INTEL_CPU_DESC(INTEL_FAM6_SKYLAKE_X,         3, 0x00000021),
    INTEL_CPU_DESC(INTEL_FAM6_SKYLAKE_X,         4, 0x00000000),
    INTEL_CPU_DESC(INTEL_FAM6_SKYLAKE_L,         3, 0x0000007c),
    INTEL_CPU_DESC(INTEL_FAM6_SKYLAKE,         3, 0x0000007c),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,         9, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE_L,         9, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE_L,        10, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE_L,        11, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE_L,        12, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,        10, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,        11, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,        12, 0x0000004e),
    INTEL_CPU_DESC(INTEL_FAM6_KABYLAKE,        13, 0x0000004e),
    {}
};

Apparently they use different letters there then, compared to the actual chip codenames in roadmaps (S for desktop, X for HEDT, U / Y for low power etc
That still doesn't change the roadmaps not including a desktop Ice Lake, unless they plan to bring it in 2022 or later.

 

[efikkan]

We already know Intel have added driver support for upcoming Tiger Lake mobile and desktop parts, and that intel is intending to launch Ice Lake-X for HEDT. But the production volume and yields on 10nm+ remains to be seen. I would not be surprised if we see a lineup mixed with both 10nm and 14nm on the same platform, but if supplies remains limited, I hope Intel prioritizes 10nm wisely.

 

[Patriot]

We already know Intel have added driver support for upcoming Tiger Lake mobile and desktop parts, and that intel is intending to launch Ice Lake-X for HEDT. But the production volume and yields on 10nm+ remains to be seen. I would not be surprised if we see a lineup mixed with both 10nm and 14nm on the same platform, but if supplies remains limited, I hope Intel prioritizes 10nm wisely.

Sure that would make limited sense, low end 10nm+ chips to combat amd APUs, 14nm++ superclocked to stave off AMD 8c Zen for gaming.
Would match the laptop clusterfuck. So far the 10nn+ sunnycove architecture is "faster" better IPC but lacks clocks.

Just a reminder, linux patch notes have served as a notice of incoming products but not who they would be targeting or when they would be released.
Intel has slid 10nm 5 years and changed roadmaps every 3mo for the past 2yrs... Not enough salt in the ocean for something like this

 

[claes]

Your comment contains zero facts yet you were talking as if you were damn sure about it. I didn't insult you, I was insulted by the level of confidence you had based on the three lines of code.

Intel promised 10nm based CPUs for 2015 (!!!) and now when year 2019 is almost over, we still don't have a working 10nm node. Base CPU clock rates for Ice Lake CPUs are just laughable: https://en.wikipedia.org/wiki/Ice_Lake_(microprocessor)

I don't know what you choose to believe in but I believe something when I see it. For previous nodes Intel at least used to show yet to be released products in advance - they haven't done that for their 10nm node. Nothing at all. If I'm writing shat, you can refer to it as "shat" pure and simple. And I won't say "someone has insulted me"- instead I will take note and won't try to behave like I'm the smartest soothsayer to ever exist.

If you are willing to speculate about new nodes and Intel uArchs go to Anandtech as I'm done here.

Speaking of embarrassment - I have nothing to be embarrassed about. I was honest and straightforward and I didn't try to make things up, unlike you. And you are still trying to defend your comment without providing a single fact - just speculation on top of speculation.

Can you point to an example of a Linux kernel commit by Intel indicating new processors that Intel later pulled and never released?

Because that’s your argument, and I don’t see any evidence here.

I have no horse in this race towards speculation, but I do think it’s important to play nice

Have fun out there!

 

[TheGuruStud]

There's no mention of "-S" models, which is the actual "desktop line", and none of Intel roadmaps includes one for Ice Lake either, so the "desktop" probably means something else than most think
edit:
Intels next desktop chip should be Comet Lake next year followed by Rocket Lake, which supposedly has Xe-based graphics (possibly as separate die)

"Desktop" all in ones and quad core xeons...WHAT YEAR IS IT?

Maybe some glued together, terrible TDP, xeons, too.

 

[efikkan]

Sure that would make limited sense, low end 10nm+ chips to combat amd APUs, 14nm++ superclocked to stave off AMD 8c Zen for gaming.
Would match the laptop clusterfuck. So far the 10nn+ sunnycove architecture is "faster" better IPC but lacks clocks.

Intel haven't released any products on 10nm+ yet. Ice Lake-U/-Y is still on the first generation 10nm, the same as the nearly forgotten Cannon Lake.
Except for low-power laptops, it would make more sense to save the 10nm+ capacity for server and higher desktop models, as the 35-45W TDP latop CPUs does just fine on 14nm++ for now. It's easy to forget that Intel's 14nm++ have excellent yields and is quite a bit better than TSMC's and GloFo's 12/14/16nm nodes, and it's when you push clocks higher and get a lot of heat that 14nm++ starts to fall behind TSMC 7nm.

There are many who claim that Intel should give up 10nm and wait for 7nm, but they are just clueless. 7nm will start limited production in 2021, and will only be able to reach a majority of volume by late 2022 or 2023, at the earliest. Intel will launch their 10nm+ in 2020 and 10nm++ in 2021, and they have no option but to fix it. And there is technically no reason why they shouldn't be able to get similar results as TSMC's 7nm eventually. And if Sunny Cove provides a similar ~18% IPC gain for desktop and HEDT, but can only boost to ~4.5-4.6 GHz, it would still outperform a 5 GHz Coffee Lake.

As to your claim of "14nm++ superclocked to stave off AMD 8c Zen for gaming", that really doesn't make much sense;
Firstly, in games CPU performance only matters to the point where the CPU is no longer bottlenecking the GPU, and for current games and the i9-9900K, it's already faster than it needs to for most games, so pushing beyond 5 GHz isn't going to give a significant advantage over AMD.
Secondly, each hundred MHz beyond 5 GHz is going to cause significant increases in power requirements, I'm not even sure 14nm++ would be able to do that. Intel is much better off getting more of their cores to scale better on clock speed rather than pushing extreme clocks. Even "backporting" Sunny Cove to 14nm would make much more sense and perform much better than pushing Skylake beyond 5 GHz.