The company says they will offer a 40 percent speed boost over the previous generation
Really??? Apparently Intel's marketing department has trouble distinguishing between the meaning of the terms "speed" and "performance".
...with performance increases coming via the two new cores and improvements to their design and manufacturing (higher clock speeds)
OK, at least they're changing their claim from "+40% speed" to "+40% performance", but even there I think their math is way off:
-- Manufacturing somehow providing at least a +5% improvement, if not +10%? Sorry, gotta call them on this from the getgo. First off, most performance "improvements" that are technically due to the manufacturing process show up in other areas (# of cores you can fit in, HyperThreading support, # of transistors, base/Turbo clock speeds, etc.). Second, the majority of those improvements, as well as those that don't readily show up (I.e. shortening the distance between cores & cache) primarily happen when you have a die shrink. I suppose they could have left that in from their Kaby Lake figures, since that was
technically a 5% improvement in clock speeds (since they didn't change the # of cores)...but since we're seeing changes in the # of cores & a drop in clock speed to boot, I don't know if I can believe even a 5% claim.
-- Design improvements? Where? They're calling these 8th-generation CPUs, but they're not actually Coffee Lake CPUs, they're a Kaby Lake "refresh". Normally, that should imply the same features sets (I.e. AVX 2.0, VT-x, etc.). However, it turns out that these chips are missing 2 of the features in their 2C/4T forebearers (http://ark.intel.com/products/124968,
http://ark.intel.com/products/97466): Intel's vPro technology, & Intel's Stable Image Platform Program (SIPP). Now, maybe only enterprise customers & support staff are going to care about those features, or maybe it's just that old of a protocol (since they've had it in their CPUs since the Core 2 days)...but it almost makes it look like they had to remove some features in order to fit everything else in. Not to mention that enterprise purchases are still more of a source of purchasing than consumers will be, at least in terms of volume. In the end, though, I think they may have mislabeled this section, because architecture generally refers to the internal structure of how each core is designed (I.e. "Sandy Bridge", "Ivy Bridge", "Haswell", etc.), where "design" is how much you can put on the CPU (I.e. a desktop Core i3 is "designed" to have 2 cores & use HyperThreading, a desktop Core i5 is "designed" to have 4 cores but not use HyperThreading, a desktop Core i7 is "designed" to have 4 cores & use Hyperthreading, etc.).
-- Design providing a +25% improvement? First off, that's going to vary quite a bit from application to application. And right off the bat, we run right back into the "cores vs. frequency" debate. On single- or lightly-threaded applications, these CPUs are going to
at best see maybe a +5% improvement because a) they'll only be using 1 or 2 cores, & b) the listed Turbo speeds (which, remember, only apply when a
single core is in use) are only 5% faster than the older CPUs they listed. On multi-threaded applications that only need maybe 4 threads (I.e. most gaming applications), especially if you run into thermal throttling, you're going to see these CPUs
lose performance, because their stock clocks run about 25%
slower (& no, we don't have listed what the max Turbo is for any of these CPUs with all 4 cores in use...however, given that the 2C/4T versions have a +1500MHz step, & a similar Haswell version, i7-4550U, had a +1200MHz step with both cores in play, I'm guessing that the i7-7660U can probably reach at least 3.5GHz, if not 3.7GHz, with both cores; the i7-8650U would have to manage a +1600 to +1800MHz step with 2 or more cores to match that performance). The only situations where these 8th-generation CPUs might be able to outperform their faster-with-fewer-core brethren is in applications that can readily take advantage of 5-8 threads. But again, I figure the lower clock speeds are going to have an affect there, especially since once you hit 5 threads you're into HyperThreading usage (which is why you usually see only a +25-30% improvement when going from a 4C/4T CPU to a 4C/8T CPU
of roughly the same clock speeds).
The only impressive thing about these CPUs is that they managed to squeeze them down to a 15W TDP. But even that's not as impressive as you might think. They got a 4C/8T CPU down to 25W previously (i7-6822EQ, 2GHz base/2.8GHz Turbo with 1 core), so it's not quite as much of an improvement as with their normal low-power 47W CPUs. And even then, I'm betting the only way they can get such a huge Turbo gap is because that's with only 1 core working; I'm betting the 3- and 4-core max Turbo speeds are much, much less.
