The Flagship Mobile Core i9-8950HK

with Intel Thermal Velocity Boost

The peak of Intel’s mobile binning strategy for the Spring 2018 launch is the new Intel Core i9-8950HK. It sits atop the mobile CPU stack with six cores, a high base frequency, a high turbo frequency, an even higher turbo based on thermals, and support for overclocking. Intel calls this its highest-performance mobile 8th Gen Intel Core processor, focused on no-compromise gaming, VR, and content creation. Ultimately the marketing comes across as this is a desktop replacement (DTR) processor for those that need big and beefy mobile systems with one or two high-end graphics cards: the sort of DTR laptops that weigh several pounds.

AnandTech Core i9-8950HK Core i7-7820HK
Generation Coffee Lake-H Kaby Lake-H
Launch April 2018 January 2017
TDP 45 W 45 W
Cores 6C / 12T 4C/8T
Base Frequency 2.9 GHz 2.9 GHz
Turbo Frequency 4.6 GHz 3.9 GHz
Thermal Velocity Boost +200 MHz -
iGPU UHD 630 HD 630
iGPU Base/Turbo 350 / 1200 MHz 350 / 1100 MHz
DRAM Support DDR4-2666 DDR4-2400
Optane Support Yes No
Tray Price $583 $378

The Core i9-8950HK is the only product that will have the Core i9 branding, although unlike the Core i9 products for the high-end desktop, the only things here that separates the Core i9s to the Core i7s are the frequency, the new Thermal Velocity Boost, the overclocking, and the L3 cache per core.

The Core i9-8950HK is a six-core processor with a base frequency of 2.9 GHz, and will have a nominal maximum turbo of 4.6 GHz. The 45W thermal design power (TDP), a staple of a Core-H series processor, will be for that 2.9 GHz figure: any turbo mode is likely to go beyond the TDP for power consumption values, so something like 4.6 GHz on all cores might be pushing nearer 100W. This is important, as Intel will actually list the maximum turbo frequency as 4.8 GHz due to the inclusion of Intel’s Thermal Velocity Boost (TVB) feature.

TVB is essentially another phase on top of Intel Turbo Boost such that if the CPU stays under a certain temperature, in this case 53ºC, the single core turbo frequency can go up to 200 MHz higher than the processor is ‘normally’ rated for (despite the fact that Intel will officially promote the rating at the higher frequency anyway). This is designed to leverage OEM designs that are efficient in their cooling, as well as benefit the lucky individuals that get very nice chips from the silicon lottery. Intel states that if the workload is continuous, or over the age of the device, as temperatures rise then this extra 200 MHz is likely to decrease, and thus promotes the TVB implementation for short and burst-y workloads. A good example of this is something like WebXPRT, which is one of Intel’s favorites, or the ‘Responsive’ workload test in SYSMark 2014. For multi-core workloads, TVB will afford a potential +100 MHz increase if the thermal threshold is not breached. Knowing how most OEMs work, it would be interesting to see if that threshold can be adjusted.

With TVB in mind, and the fact that Intel states +200/+100 MHz based on the core loading, it is worth reiterating that Intel does not state per-core turbo modes any more. As per Intel’s policy from last year that, and I quote,

‘Intel specifies only base and single-core turbo frequencies for its processors and no longer discloses turbo frequencies beyond this level of detail, as it is proprietary to Intel.’

I’m still calling out this atrocious policy, especially when the value can be obtained by just having the chip in front of you. Despite having this policy in place, we were still given the breakdown of per-core turbo frequencies for the Xeon D-2100 family of processors launched earlier this year, which means the policy is not uniformly applied across Intel’s processor deployment. What irks me more however is this new turn of phrase: ‘as it is proprietary to Intel’. It’s a vacuous statement that means nothing. Turbo frequencies can be adjusted by the motherboard manufacturer or the OEM, who do so regularly on mobile platforms, so whatever notion of ‘proprietary’ is meaningless and goes out the window – ultimately we use Intel’s guides to make sure the vendors are playing ball, and not trying to burn laps/desks with abnormal settings.

One of the interesting things here is that Intel states that this chip is using the latest 14nm technology. This basically confirms that this chip is using Intel’s 14++ nm process node, which would also imply a Coffee Lake design, however Intel has not yet formally confirmed this to us.

The Core i9-8950HK is likely to appear in the refreshes of all the high-end DTR systems, such as MSI’s Titan line, some massive Acer Predator thing, a big ASUS ROG system, and perhaps someone might try and slip it into something slimmer with an NVIDIA Max-Q.  We know of at least one set of machines that will have the Core i9 as a processor option, however those devices are still under embargo as this announcement is made. A few hours after this announcement you can follow this link to see examples of the i9-8950HK in devices:

https://www.anandtech.com/show/12605/

(We’ll update this page in a bit)

It is also worth noting that the Xeon E-2186M, the first member of the so-far-unannounced Xeon-E family that is being launched today, mirrors the Core i9-8950HK in almost every aspect, but swaps the overclocking functionality for the ability to use ECC memory and be vPro enabled.

Coffee Lake-H: Intel’s New 45W Processor Range

The rest of Intel’s 45W processor line mimics previous generations, with the noticeable bump up in core count, memory support, and Optane support. This is best demonstrated in a table:

AnandTech Cores Base
Freq
Turbo
Freq
L3 vPro DRAM
DDR4
iGPU iGPU
Turbo
Core i7-8850H $395 6 / 12 2.6 4.3 9 MB Yes 2666 24 EUs 1150
Core i7-8750H $395 6 / 12 2.2 4.2 9 MB No 2666 24 EUs 1100
Core-i5-8400H $250 4 / 8 2.5 4.2 8 MB Yes 2666 24 EUs 1100
Core i5-8300H $250 4 / 8 2.3 4.0 8 MB No 2666 23 EUs 1000

All the CPUs are rated at 45W for their base frequency, all support Optane as a drive accelerator, and all support DDR4-2666 in dual channel operation. Moving down from Core i9 to i7 and then to i5 gives a gradual decrease in the L3 cache per core (Intel still uses an inclusive cache in this market), as well as a steady drop in frequencies. The Core i5 parts are quad-core processors, but still retain DDR4-2666 compatibility.

We are told that all of these CPUs are Coffee Lake based, marking the arrival of the Coffee Lake-H family. That being said, there is a disparity between Coffee Lake desktop-based Core i5, which run at six cores and six threads, compared to the Coffee Lake mobile-based Core i5 parts which are quad cores with eight threads. This would allow for the Core i5 parts to be Kaby Lake-Refresh processors, however Intel has confirmed this is not the case, making these the first quad-core with hyperthreading-based Coffee Lake parts.

One of the more anomalous processors in this list is the Core i7-8850H, which Intel has listed as being ‘partially’ unlocked, compared to the Core i9-8950HK which is fully unlocked. Intel has clarified that this means the Core i7-8850H has the ability to overclock the DRAM and up to 400 MHz on the CPU beyond the standard specifications. From the devices we have seen so far, almost all the laptops that use the Core i9 also have a Core i7-8850H option.

Intel is also enabling vPro on a couple of SKUs: the Core i7-8850H and the Core i5-8400H. The only thing these have in common is that they’re the highest-rated SKUs for each segment, but beyond that there is nothing in the naming to suggest there is a pattern.

The Belated Birth of Xeon E

Alongside the consumer processors, Intel is also pushing out two Xeon mobile CPUs with the rest of the 45W stack. This is a little confusing, given that Intel hasn’t even officially announced the Xeon E processor line, and that we have known about the name change from Xeon E3 to Xeon E for a good few months now. Nonetheless, the two Xeon E mobile processors are both set for the high end, emulating the Core i9 in performance and support.

AnandTech Cores Base
Freq
Turbo
Freq
L3 vPro DRAM
DDR4
iGPU
 
iGPU
Turbo
Xeon E-2186M $623 6 / 12 2.9 4.6 + 0.2 12 MB Yes 2666 ECC 24 EUs 1200
Xeon E-2176M $450 6 / 12 2.7 4.4 12 MB Yes 2666 ECC 24 EUs 1200

The two big advantages for the Xeon E processors are vPro support and ECC memory, however they lose overclocking capabilities. Pricing is, as yet, unknown, however devices with these components will be announced this week.

* Base IGP frequency for all 45W processors is 350 MHz.

Intel Expands 8th Gen Core: Chipsets High-Performance Mobile: Coffee Lake with Iris Plus at 28W
Comments Locked

123 Comments

View All Comments

  • BillBear - Tuesday, April 3, 2018 - link

    Actually, we've already seen ARM chips surpass Intel in the server space where Intel is strongest.

    https://blog.cloudflare.com/arm-takes-wing/
  • FunBunny2 - Tuesday, April 3, 2018 - link

    "2020 from Intel x86 CPUs to some sort of higher performance ARM processor "

    well, they could stamp 12 or 24 or 48 ARM cores on a chip, and call it a CPU. but that would mean they've abandoned single thread performance. it only makes sense if they've some secret multi-threaded sauce, built into macOS (or whatever they end up calling it), that runtime converts from single to multi. that's some Catch 22.
  • PeachNCream - Tuesday, April 3, 2018 - link

    It sounds like the current intent is to consolidate under iOS across all of their computing devices. I do wonder how you'd reach a x86-comparable performance level with ARM cores. You're right they'd almost have to go crazy into multi-threaded stuff, but there are still some workloads that just don't benefit much. Like I said above though, I'm not really praising the move or particularly excited about it. It's just something that'd be interesting to watch happen.
  • HStewart - Tuesday, April 3, 2018 - link

    A non Intel based Mac will never replace high end mac - that may try it say MacBook Air line - but it likely be very unsuccessful.
  • SaturnusDK - Tuesday, April 3, 2018 - link

    You underestimate the baa! of the iSheep. They'll buy whatever crap Apple launches.
  • serendip - Tuesday, April 3, 2018 - link

    I was a Macbook and Powerbook user around the time of the PowerPC/x86 transition. Apple handled it well by having the Rosetta translation library; the Intel Core Duo chips back then had enough grunt to handle translated code at a decent speed too. It took a few years before big programs like Photoshop had native Mac Intel ports.

    An ARM Macbook with 20 hour battery life and Intel-equivalent performance would be a big seller provided code translation works well and popular apps have native ARM versions. Most Macbooks are used for Web and app development so it won't be hard to recompile code for ARM. I'd say Apple could do this better than the half-baked efforts by Windows OEMs to make Qualcomm PCs.
  • BurntMyBacon - Wednesday, April 4, 2018 - link

    Are we talking low end Apollo Lake or high end Coffee Lake equivalent performance. The former isn't going to have compelling performance in anything beyond (perhaps) Web Apps. The later is a pipe dream. I don't much care for Intel as a company, but do you seriously believe they are so incompetent at designing processors that a company with only 8 year experience designing processors (Apple) is going to double battery life (Macbook) with the same performance while using translation? Catching up to Intel, while not easy, is doable. After all, many of the things Intel did to get to their current performance are now known quantities. Very suddenly passing Intel up doesn't seem plausible as that would require Apple to have design expertise Intel doesn't have. Very suddenly passing them up by such a wide margin ... well I hope it happens as it would spur on competition. However, hope is not in fact a strategy, so I'm not counting on it. More realistically, they could take the AMD approach of offering more cores and trading off single threaded performance for mult-ithreaded performance. That may make it compelling for some use cases.
  • serendip - Wednesday, April 4, 2018 - link

    20 hours based on the Qualcomm PCs which a few vendors have demoed. An Apple ARM chip could achieve similar battery life with similar x86 translation performance. If Qualcomm can do it, I would bet that Apple can too, especially as they control both hardware and software on their devices.
  • fteoath64 - Thursday, April 5, 2018 - link

    Not only this. Apple can add special "sauce" in TensorCore chips into the mix and moe specialised co-processors within the Arm CCN that Intel cannot really match. There is much to innovate in that space where Intel just went into "brute force" speed and power-savings just to market their chips over the last few generations. Apple's GPU expertise is also coming to speed so there is much hope there....
  • HStewart - Saturday, April 7, 2018 - link

    Using AMD for CPU is non-sense - it obvious that Intel has lead on AMD especially in the Mobile CPU and Apple has pretty much abandon the desktop area. Only with iMac Pro with monitor based iMac's are basically now mobile cpu because of size and power.

Log in

Don't have an account? Sign up now