Intel's Kaby Lake CPU: The Good, the Bad, and the Meh | Hacker News Reader

Intel's Kaby Lake CPU: The Good, the Bad, and the Meh

makeuseof.com

61 pointswalterbell76 comments

The author of this article says that Kaby Lake is an 'anomaly' when it comes to the tick-tock strategy that Intel has been following. However, this is not the case. Following the rollout of Intel's 14nm lineup, Intel realized that it was no longer feasible to move to a smaller processing node every 2 years. And so tick-tock died.

It is currently not expected that after Kaby Lake or the 14nm node, that Intel will move back to the tick-tock model. Kaby Lake is no anomaly - it's a normal architecture update in the post 14nm world.

I agree that it's a slightly odd way of describing the situation, but it is true that this is a _different_ kind of architecture update than what Intel has done for the past few years, which is worth pointing out to readers. For people who are not up to date on Intel's roadmap, this update is distinct from what has happened before, and therefore anomalous.

True, but they really should have taken the care to clarify that this is also suppose to be the new normal. Kind of sloppy otherwise.

I am all for such updates, especially when they bring new important features. For some reason people always think it's all about clock speed, performance, etc. It isn't. Kaby Lake new features introduce:

Native USB 3.1 Generation 2 (10 Gbit/s) support (personal favorite)

200 Series chipset (Union Point)

Support for 16 PCI Express 3.0 lanes from CPU, 24 PCI Express 3.0 lanes from PCH (LGA 1151)

Support for Thunderbolt 3

Support for Intel Optane Technology

https://en.wikipedia.org/wiki/Kaby_Lake

You forgot HEVC 10-bit hardware decode, VP9 10-bit hardware decode

Does it mean support for h264 Hi10P profile too?

What will be the simd width? Will it support avx512?

It appears to be the same old 256-bit Skylake core.

Thank you.

Skylake Xeons are supposed to support AVX-512 and Kaby Lake was supposed to be the first desktop platform enabling them, so there is hope ;-)

Hopefully AMD's Zen desktops support AVX512, forcing Intel's hand.

http://www.anandtech.com/show/10578/amd-zen-microarchitectur...

> The FP side of the core will afford two multiply ports and two ADD ports, which should allow for two joined FMAC operations or one 256-bit AVX per cycle.

Still looking forward to Zen

I bought a Haswell Xeon (e5 1630) just this week for a new workstation build, because it was the only Xeon with high ghz that was available at a reasonable price (ie comparable to i7). E.g. the e3 1280 is a skylake xeon but 'only' supports avx2. Maybe the e7 skylake ones have avx512? I didn't look into those as they were out of my budget anyway :)

This thread might help: https://software.intel.com/en-us/forums/intel-isa-extensions...

Be careful about buying into marketing BS. Nothing about the Kaby Lake platform has anything to do with Intel Optane; Optane SSDs are just NVMe SSDs and will work with any PCIe host, and if you think they've added Optane NV-DIMM support to a consumer platform more than a year before such products even exist to be bought by enterprise customers, you're delusional.

Likewise, "Support for Thunderbolt 3" is misleading; Thunderbolt 3 obviously does not require a Kaby Lake processor. Kaby Lake may include a built-in Thunderbolt 3 host controller, but I haven't seen clear confirmation of that. Skylake is also advertised as having Thunderbolt 3 support, but that only means it has PCIe lanes to attach to a Thunderbolt 3 controller.

What does it even mean for a processor to "not support Windows 7"? Does this mean it's not fully x86 backward compatible, or is this just marketing speak for "we have a strategic alliance with Microsoft, and they don't want you to run Windows 7, so we've put hooks into our hardware to prevent it"

Poking around online, it seems kind of like it will run Windows 7, but will never receive updates.

“Windows 7 was designed nearly 10 years ago before any x86/x64 SOCs existed. For Windows 7 to run on any modern silicon, device drivers and firmware need to emulate Windows 7’s expectations for interrupt processing, bus support, and power states, which is challenging for WiFi, graphics, security, and more. As partners make customizations to legacy device drivers, services, and firmware settings, customers are likely to see regressions with Windows 7 ongoing servicing.”

There might also be hot new important and/or whizz-bang features on new silicon that Windows 7 does not support and Microsoft doesn't want to support on 7.

Ok so the real explanation is that Windows 7 will not support the new Intel chip. Not the other way around.

It's going to be very difficult to get Windows 7 working if Intel won't release things like USB or video drivers for it.

There are already problems with Skylake where you'll do a fresh install of Windows 7 and none of your USB ports or network devices will work because they require newer drivers. Microsoft is doing their best to drive everyone to Windows 10 by not doing another service pack or releasing updated ISOs.

Intel has no reason to not release those drivers for it. Windows might not ship the drivers themselves, or make it "just work." But that's not in Intel's interest, they want to sell chips. They don't care what you run on it, and will support whatever has a large enough install base.

Microsoft is the one not doing the supporting. Which means that if something doesn't work, and you create an issue with them (think enterprise customers) their official response is going to be that you need to upgrade. This is perfectly reasonable and Apple does this all the time. There is no reason why older iDevices can't install newer IOS, besides pushing people to upgrade.

I don't see Windows 7 "not working" with Kaby Lake. Maybe things that require deeper OS support, like Optane as DDR chips, or booting off Optane. Maybe Thunderbolt 3 won't work right. But for the most part, most things will work just fine.

Think of it this way, if OSes had to actively support new chipsets, you would need to buy/get new install disks with every release. That just doesn't happen.

Linux definitely does have to get updates for new chipsets, mostly around power management features. iGPUs also tend to get updated, which needs new drivers. Old kernels will still run, but they will use more power than necessary and not have access to many new features.

Indeed. In the past when I've needed to add support for something during installation, I've never had a problem slipstreaming it in.

It's more than that. Windows 7 doesn't natively support XHCI (USB3) and Intel dropped EHCI support in Skylake. EHCI was expected to be dropped for a long time from hardware support since modern Operating Systems have supported XHCI for quite sometime.

Here's AMD chance to say that it will support Windows 7 on its Zen and Zen+ processors.

Windows comes with patches for the CPUs like https://support.microsoft.com/en-gb/kb/3064209

Probably means there won't be any for < Win10

That's not a "patch for the CPU", that's a microcode update. Both the CPU, and Windows work just fine without it.

It means that it's not fully compatible with the existing PC architecture in terms of aspects like power management, peripherals, etc. Skylake was the same to some extent and it apparently caused really obnoxious, hard-to-fix compatibility issues with Linux too. I'm not sure those have been fully dealt with even now.

Intel removed EHCI support for USB in it's chips with Skylake in favor of xHCI. This has been a long time coming and not a malicious or insidious act. Windows 7 doesn't natively have XHCI support and Microsoft isn't adding features to Windows 7. So, on Skylake, you can't install Windows 7 via USB and you can't use a USB keyboard or mouse during installation because the USB drivers are incompatible.

I suspect there are many other incompatibilities that exist and Microsoft has just drawn a line in the sand.

Hardware and software don't always evolve in lockstep. Sometimes new hardware arrives that breaks something or it doesn't keep up with the evolution of software. Microsoft has taken somewhat bizarre actions in the past to correct this situation when it happens.

In the 90s just a year after Windows 95 was launched you saw USB, AGP, UDMA, PCI and other hardware advances come to market. Windows 95 supported none of it. They released Windows 95 OSR2 to OEMs to add support for all the new hardware. Windows 98 was a long way out and they needed to take action to support new hardware or face being displaced in the market.

With all of the hardware advances of the late 90s and the rapid decline of the 16bit CPU, ISA, and a lot of other legacy systems, Microsoft wanted to make a clean break in the 00s. Windows XP wasn't going to support any of the old cruft of the 80s and early 90s. The problem was that in the years leading up to XP, OEMs refused to start shipping PCs that met the minimum requirements. They were still shipping with legacy hardware that wasn't compatible. So Microsoft stopped shipping Windows 98 and instead replaced it with the crap fest known as Windows ME. WinME was a rebranded Win98 with one key difference, it had all of the "legacy" drivers removed giving it the same minimum specs as WinXP. Unfortunately some of that legacy was so ingrained that WinME was incredibly unstable.

No way to add/override driver on pre-install stage ? (memories of 2K era setups)

Yes, you can still add drivers, however, someone (Intel probably) would have to release those drivers for Windows 7.

If they are built for Windows 10 and use Windows 10 kernel-mode APIs, they won't work on Windows 7.

However, I don't think Intel won't build drivers for Windows 7. It's against their own interest. Intel just wants to sell hardware.

I was thinking of a non corporate effort, maybe some people from the linux world. I don't think Intel cares because they surely have ties with MS ensure Win10 on Kaby Lake boxes.

In today's mobile world I'd guess most Kaby Lake boxes are sold to enterprise customers, who care a lot about Windows 7. It's just a guess though.

Porting/using as reference Linux drivers (a lot of which are written by Intel people anyway) to Windows just doesn't seem to happen very often. Technical difficulty aside, which is significant (I've written both Windows and Unix (although not Linux) kernel mode drivers for years), nobody seems to care about such things, so I wouldn't get my hopes up.

One option could be running Linux as the host operating system on Kaby Lake hardware, then running Windows 7 in a VM where drivers would mainly need to support virtual hardware. CPU/microcode compatibility may still be an issue.

> then running Windows 7 in a VM

With the added benefit of being able to run on immutable disk images or to roll them back to safe states in case some malware decides to take residence.

My mom works for a health insurance company and all of their machines boot off immutable network images. The only downside seems to be longer boot times and immutable desktop icons.

Lots of enterprises end up in this situation. Tax office had Windows machine only to run an AS400 terminal emulator.

xHCI is a non-issue, Intel even provides an official tool to patch Win 7 images to enable USB3 support during install [1]

The main issue is processor errata and architectural oddities that need to be fixed in the kernel itself, such as problematic (buggy) instructions or unexpected behavior in the scheduler (e.g. moving execution from core to hyperthreaded core), deep sleep states. If those bugs aren't fixed, users might experience random freezes/reboots or even being unable to boot.

[1] https://downloadcenter.intel.com/download/25476/Windows-7-US...

This article is written by someone who clearly doesn't understand the industry well. And 10 nm for Intel is not directly comparable with 10 nm for TMC. Intel still maintains their fab lead by pretty much a whole process generation.

Can you explain the differences in a muggle-friendly way?

"TSMC’s 10nm shrink would retain a 20nm minimum feature size, while its 7nm would deliver a 14nm minimum feature size (10/20 and 7/14, respectively)."[1]

Intel is the only company with a "true" 14nm and 10nm process.

[1] http://www.extremetech.com/computing/228806-arm-announces-ne...

Microsoft could improve sales of Windows 10 by leaps and bounds by coming out with a No Analytics edition. They may think its a techy problem, but I keep having normal people tell me they won't upgrade because "it spies on you and snoops your credit cards". This is getting silly and stupid.

Those same people use Google and Facebook without any qualms. Microsoft should hire their marketing psych people.

I don't disagree, but it feels a bit different when its something you supposedly own. Frankly, some of the Facebook posts are a bit on the fringe side, but we're talking about PCs here so they're instruments of the Twilight Zone to begin with.

I keep thinking some Linux / BSD group would get it together and go for this still big market, but they are worse at marketing than Microsoft. I had such high hopes for ARM-based PCs.

You mean Google Chromebook (Linux) or Apple Mac (BSD)?

Well, I did have some hope of Chromebook's but they are phone-home only machines (We had a discussion in another thread on them and my trying to buy them institutionally and all the fail). Is there even an ARM Chromebook still sold?

Apple doesn't seem too interested in taking over the PC market or even releasing new PCs.

The game was lost when UEFI came out with certs required. Or maybe earlier, when Windows began phoning home.

> I had such high hopes for ARM-based PCs.

If you haven't given up entirely on ARM Linux-based desktops and laptops, you might want to check out this project:

https://www.crowdsupply.com/eoma68/micro-desktop

You own the physical media. You are free to try to install it on your computer (and, if you have a PS/2 meyboard/mouse, that can even work)

> some Linux / BSD group would get it together and go for this still big market

I don't think they'd be interested in writing Windows drivers and the NDAs they'd need to sign would probably end their careers in open-source. What you can do is to install a thin supported OS that can host VMs and install Windows 7 on one. Oracle's VirtualBox even has some support for GPUs.

Uhm... I must of really missed on that - I don't want them writing Windows drivers - I wanted someone to come up with an amazing alternative to Windows boxes for end users.

The ReactOS guys are trying.

And if Windows stayed free, we would have our apples to apples comparison.

you can limit your usage of google/fb but you can't escape windows (without switching the os).

This is something of a strawman tbh. The computing equivalent of implying anyone who has ever taken their cloths off and had sex must be addicted to pornography.

Just because people use some software or websites that publicly display themselves, does not mean they would ever be willing to accept everything they do on their computer being visible to random strangers.

Kids on the other hand... Stripping away privacy has them so unihibited the average not-even-teenagers-yet cyberkids are sexting friends and total strangers alike instead of drinking an old fashioned beer.

And I used to worry they would never learn to write by hand :(

Well, there's that, but I had in mind the fact that Google and FB and Apple and others gather as much data or more on people than MS. Granted, it's not necessarily all the stuff MS might be getting through a compromised OS, but it is the kind of stuff the parent mentioned.

But it's different data. People use google and facebook with zero expectation of "privacy" - in fact, not being "private" is the actual reason they use them.

Its a very big step to be in a situation where [i]everything[/i] you do on a computer comes with zero expectation of privacy, despite the corporate attempts to machinate the two because cloud.

I'm seeing a greater and greater move to "airgapped" (network but not internet connected) machines in the home for exactly this reason. (even if it is still very small scale, that's where these trends start).

I'm currently looking for a solution that blocks all connections by default for the entire network, with a simple gui to allow failed hostnames/ip addresses and with the option to demilitarise one or two server style machines for torrents and cloud services that then serve only that content to the rest of the internal network. (like hostfile blocking only at the IP level and defaulting to whitelist). Something like a blend of IPTABLES and wireshark with a simple ui.

I know they exist (I've seen something similar in large corporate offices), but not seen anything for SHO.

May I recommend OpenBSD and pf to you? No gui but still relatively straightforward to set up what you're describing.

There's a difference between your OS spying on you vs. a website that you can choose not to use (and can easily limit the spying capabilities of) spying on you.

Most people can't easily switch OSes. You can easily install uBlock or some other ad and tracking blocking addon to prevent Google and Facebook from spying on you.

...and they would call it "Windows 10 Professional"

Nope, you have to have Enterprise to actually turn a lot of the stuff off.

> Microsoft could improve sales of Windows 10 by leaps and bounds by coming out with a No Analytics edition.

It exists. It's called Windows 10 Enterprise Long Term Servicing Branch (LTSB).

Even better, it doesn't push bullshit new features down your throat. It only comes with security updates. And even those, you can chose not to install on a case-by-case basis.

The best part about it is that it doesn't come with Metro apps either. It's basically like an improved Windows 7 (but you can install the (Metro) Windows Store if you make a custom image, and then you can have Metro apps).

However, you can't get it. No sir, you need a special contract with Microsoft to get it. It's ridiculous. No, Microsoft hasn't changed a bit. I cringe when I see all this hype about "the new Microsoft".

Yeah, we ended up starting to image Windows 10 Enterprise and I am taking my time setting up the first box to base it all on.

Doesn't help much for my home box.

What's all this talk of "the CPU won't run Windows 7"? Is this just a confused blogspam site?

I believe it comes from here: https://blogs.windows.com/windowsexperience/2016/01/15/windo...

"Going forward, as new silicon generations are introduced, they will require the latest Windows platform at that time for support....For example, Windows 10 will be the only supported Windows platform on Intel’s upcoming “Kaby Lake” silicon, Qualcomm’s upcoming “8996” silicon, and AMD’s upcoming “Bristol Ridge” silicon."

This isn't saying anything either.

That clearly says that if you have a Kaby Lake CPU, you can only run Windows 10. Not 7. Not 8 or 8.1. Only 10.

The difficulty people have with the statement is that the relationship is inverted: it's Windows 7 which does not support the new processor. The processor does not have some explicit blacklist that prevents Windows 7 from running. Rather, the processor does things that Windows 7 was never designed to deal with, as Windows 7 is a decade old.

But can you run MS-DOS on it? Lilo? Grub? Does it no longer boot in 8086 mode?

If windows 10 is the only supported OS, why not get rid of 16bit real mode and 32bit protected mode, leaving only 64bit long mode as the only execution environment?

While Microsoft has been trying to push for being able to run 64-bit Windows without the 32-bit parts installed (starting with 2008 R2, the WoW64 bits are optional for Server Core), I'm pretty certain that you can't (yet) get a complete interactive environment using only 64-bit mode.

Just looking at the list of Microsoft applications currently running on my 64-bit Win10 desktop, there's still at least a half-dozen different Windows desktop services that are running in 32-bit mode - including the "Microsoft Bing Service".

I'm sure they'll eventually want to reach a point where they can cut that off, presuming that a hybrid approach like x32 doesn't turn out to have phenomenal benefits, but it's not yet here.

When I said removing 32bit protected mode, I meant the original 32bit protected mode which is now called "legacy mode". I believe you cannot exit long mode after enabling it, but there's a sub-mode in 64bit long mode that allows running in 32 bits - which is distinct from legacy 32bit protected mode.

I guess I should have said, "leaving only long mode" (which can be 64bit or 32bit)

Ah, my apologies.

I wonder how much they'd actually save, in die area, if they removed 32bit mode while leaving in the ability to run 32bit code in 64bit mode. If it's just decoding to similar microops with slightly different parameters depending on the mode they're returning to, probably not that much, except in the decoder itself...

Nobody's been clear on how this will be accomplished, but I can really only see three options.

1) Actual blacklisting of the CPU in a Win7 update.

This would be foolish at best, since people would just decline the patch or use modified install media.

2) Lack of support through lack of drivers for new hardware

Maybe the most likely, but still not guaranteed to prevent people from running Windows 7 and either working around missing support or disabling the signed driver requirement and hacking up newer drivers until they do what's desired on Win7.

3) Requiring Win7 backward-compat hooks in BIOS/UEFI not be present on Kaby Lake+ motherboards.

This would probably be the most effective, though it would require more coordination and signoff than I'd necessarily guess they'll muster.

Probably means it won't have microcode updates to patch the CPU https://support.microsoft.com/en-gb/kb/3064209

Both Kaby Lake and Bristol Ridge are APUs. Probably it's about graphics part. There is no breaking changes in these CPUS.

It sounds like there are a bunch of drivers/controllers on the software side that don't support Win7. I seriously doubt that you won't be able to install/boot Win7 on this (I mean modern x86 is still backward-compatible with 16-bit mode!) but will probably end up with sub-optimal power performance and graphics capabilities, and any other niceties that come from platform drivers.

Disclaimer: I work at MSFT, views are my own, etc.

I believe this is pretty much it. It is my understanding that there are a ton of features on a processor today that are not simply the x86/amd64 instruction set. These things require the Hardware Abstraction Layer (HAL) to be updated in order to support them. This is all code written in x86/amd64 ASM in order to interface with the processor and MSFT simply will no longer backport this code to its older OSes.

The linked website makeusof.com goes on the "never visit" list right next to Forbes.com due to a buzzing full screen with no close options ad.

Reply on news.ycombinator.com