Today at Computex 2017, Intel announced… well, I’m not actually sure what I can call it. It’s not a chip family, because there are three different architectures involved. When writing the headline for Intel’s latest product announcements, I picked the “Basin Falls” platform to cover it all in one go, because this is by far the most confusing launch Intel has ever orchestrated. Also announced today, but launching later this year, are the new Intel X299 motherboards that had been rumoured for months, along with the brand new Kaby Lake-X and Skylake-X high-end desktop processors to combat AMD’s Ryzen and Threadripper products. There’s a lot to talk and think about with this launch, and I’ll have it all ready for you after the jump.
Skylake-X CPU comparison
|Core i7-7800X||Core i7-7820X||Core i7-6900K||Core i9-7900X||Core i7-6950X||Core i9-7920X|
|Base clock speed||3.5GHz||3.6GHz||3.2GHz||3.3GHz||3.0GHz||TBD|
|Boost clock speed||4.0GHz||4.3GHz||3.7GHz||4.3GHz||3.5GHz||TBD|
|Turbo 3.0 clock speed||—||4.5GHz||4.0GHz||4.5GHz||4.0GHz||TBD|
|Max DDR4 speed||2400MHz||2666MHz||2400MHz||2666MHz||2400MHz||2666MHz|
|Thermal design point (TDP)||140W||140W||140W||140W||140W||140|
Skylake-X is an interesting chip family because it is increasing the core count and the die size, but it is not built on a new process. The LGA 2066 socket it fits into is only a little larger than LGA 2011-3, with the extra pins serving new functions like powering the extra cores, or allowing the use of new features on the X299 chipset (which I’ll detail in a bit). These new chips are indeed based on the Skylake family, but there’s some special sauce that Intel is not talking about yet. There are supposedly some tweaks that they’ll reveal closer to launch that give them higher performance and throughput compared to the Broadwell-E family. Intel is claiming a 30% raw IPC boost over the previous generation, but there’s both a core count and clock speed increase on the cards here, so again there is likely nothing that has changed architecturally.
Relative to Broadwell-E, there are some major shifts in terms of pricing. The once-mighty Core i7-6950X used to command the highest price for Intel’s consumer parts, but it now is replaced by the Core i9-7920X, which will be detailed closer to launch, but is priced much lower. The Core i9-7900X replaces the Core i7-6900K with an extra two cores and more clock speed to spare, and drops $100 in the process from the tray price. The more interesting chip, the Core i7-7820X, takes the specs of the Core i7-6900K and now retails for $500 less than the outgoing Broadwell-E chip.
I can’t say for sure how Intel is managing this, but I’d guess that part of it is that they’re cutting their own margins with ex-server parts to remain competitive with AMD, as well as saving money from not having so much L3 cache. Intel increased the size of the L2 caches to improve throughput, so we’ll have to see how that turns out when reviews go up. Saving space on that L3 cache is not a trivial thing for a design this large
Something else that got tongues wagging was Intel’s announcement of future Core i9 parts in the family. These are Skylake-X processors with a revamped architecture that Intel isn’t talking about in great detail, which means that its a different die size and a different architecture altogether. Intel is coy about the specs of these products, and they’re only willing to part with the bare minimum of teasers.
The Core i9-7940X will have 14 cores and 28 threads, with a $1399 tray price. The Core i9-7960X has 16 cores and 32 threads, and will have a $1699 tray price. And the behemoth Core i9-7980XE will have 18 cores, 36 threads, 16.5MB of L3 cache, what appears to be two sets of Tensor cores in the die shot pictured above, and a tray price of $1999. The Core i9-7980XE also has AVX-512 support for the first time on a consumer product, which will definitely aid in heavy compute workloads that make use of vector extensions on the CPU. These will all be available to consumers eventually, possibly as a late 2017 launch, or perhaps even as an early 2018 product.
But whichever date Intel decides for these parts, they must surely be thinking of how close AMD might price their Threadripper and EPYC processors. If the 16-core, 32-threads Threadripper flagship costs less than the Core i9-7960X, then I expect that Intel will have a slightly tougher time selling these chips to their customers. For reasons that I’ll discuss later in AMD’s Computex coverage, I think Threadripper has several benefits that make the Core i9 family look like the weaker choice.
In terms of new chipset features, there isn’t too much that is different, but there are some caveats with the X299 platform. Firstly, the lane allocations are now split up more. The Kaby Lake-X processors allow up to 16 PCIe lanes from the processor, which basically turns off a lot of the motherboard’s PCIe slots when you use these lower-end chips. Two Skylake-X chips only have 28 lanes attached to them, but this is to allow end-users to configure systems with up to two GPUs running off eight lanes each, with a further eight lanes split between two PCIe NVMe SSDs, or one NVMe SSD and one using SATA Express, or two SATA Express ports in lieu of any PCIe storage.
The rest of the Skylake-X family has access to 44 lanes of PCIe 3.0 connectivity, and this will be mostly split up between two full-speed GPUs at 16 lanes each, and up to two PCIe NVMe drives. Intel’s platform still doesn’t allow for bifurcation of the PCIe lanes, so an external chipset like a PLX chip will need to be used to split up the lanes further to allow for up to four GPUs to be added to the system. Motherboards that support this feature will end up costing more.
Kaby Lake-X CPU comparison
|Core i5-7600K||Core i5-7640X||Ryzen 5 1600X||Core i7-7700K||Core i7-7740X||Ryzen 7 1700|
|Chip family||Kaby Lake||Kaby Lake-X||Zen||Kaby Lake||Kaby Lake-X||Zen|
|Base clock speed||3.8GHz||4.0GHz||3.6GHz||4.2GHz||4.3GHz||3.0GHz|
|Boost clock speed||4.2GHz||4.2GHz||4.0GHz||4.4GHz||4.5GHz||3.7GHz|
|Turbo clock speed||—||—||4.1GHz||—||—||3.75GHz|
|Max DDR4 speed||2400MHz||2400MHz||2666MHz||2400MHz||2400MHz||2666MHz|
|Thermal design point (TDP)||91W||112W||95W||91W||112W||65W|
Now we get to the confusing part of the Basin Falls platform. Intel is supplementing their overclockable Core i5 and Core i7 processors from the consumer-bound Kaby Lake/Z270 platform with slightly upgraded versions on the new X299 platform. These are Kaby Lake-X processors, but they’re based off very similar silicon to the chips on the desktop. In fact, Intel admits that it’s rather the other way around – Kaby Lake-X is the fully implemented processor design but with the integrated graphics disabled, and all the extra features that the older products didn’t have now enabled. This means that Kaby Lake always had four memory channels to begin with, it just didn’t use them.
Intel is going to be asking for a lot more on the motherboard pricing side, at least as much as entry-level X99 motherboards (which, at the time of writing, was around R3,700 for MSI’s X99 SLI PLUS). Pricing the chips equally helps a little with adoption, and it’ll be up to the customer to figure out what they want. But is a smidgen more clock speed and a slightly higher TDP going to make up for that increased cost? No, it doesn’t. If you were looking at the Core i7-7740X with mild interest, you’d be better off settling down with a Core i7-7700K and a decent Z270 motherboard for half the price of an entry-level X299. Intel isn’t even using solder on these new processors, just the same thermal grease they’ve had on consumer parts, so heat will similarly be an issue when cooling one of these bad boys.
Kaby Lake-X on the X299 platform, as things stand today, simply doesn’t make much sense. It might overclock higher on more exotic cooling options, but if that’s all Intel has to offer here, that won’t be enough to bring up the adoption rates among consumers.
Of course, there’s always the Ryzen 7 1700 to consider as well. For the same price as a Core i7-7700K, you get more cores, more threads, similar levels of overclocking headroom, more cache, and you pay less for the platform overall because a B350 motherboard still allows for multiplier-based overclocking. And there’s an RGB-lit stock cooler in the box. If your workload is mostly CPU-intensive and likes to load up more cores, it wouldn’t make sense to buy any of these Intel processors in the first place. Hell, even the Ryzen 5 1600X deals quite a blow for the Core i5 family, taking away the value proposition Intel had in the past. I think that Intel’s efforts to bridge the price gap to their high-end desktop processors is admirable, but AMD beat them to the punch with Ryzen and I don’t foresee that changing anytime soon.
In the future, Intel will replace the Skylake-X processors currently in the higher-end Core i7 and Core i9 brackets with Kaby Lake-X processors, but those chips will be fundamentally different to these ones, with a tweaked cache structure and new capabilities. Picking up a Core i7-7740X and a X299 motherboard isn’t a bad decision because you still have an upgrade path all the way to the mental Core i9-7980XE, but these chips will lack any of the optimisations Intel will be applying to those processors, and thus the lower-end Kaby Lake-X processors are outdated even before they launch.