CORE i9-7900X
Intel’s first Skylake-X CPU reviewed
SOMETHING WEIRD is happening at Intel. Recent years have seen small, incremental upgrades to its top-drawer enthusiast chips. A speed bump here, a couple extra cores there. And that was if you were lucky. Out of nowhere, however, Intel announced a whole box full of new CPUs, a new LGA2066 socket to stick ’em in, and a new X299 chipset to underpin the whole shebang.
The new chips aren’t the usual minor massaging of specs, either. In one fell swoop, Intel has gone from a maximum of 10 cores for desktop PCs to 18 cores, with additional ’ tweener options filling the newly opened chasm between the old and new realities of Intel’s high end.
That’s all remarkable enough. Where the real weirdness comes in is that Intel has announced all those new CPUs, but hasn’t been able to provide full specifications. For the 14- to 18-core models, we know the model names and core counts, but that’s pretty much it. We don’t know clock speeds or TDPs. The reason, almost certainly, is that those chips are a hasty reaction to the announcement from AMD, Intel’s only rival in the performance PC processor market, involving a shocking new 16-core CPU known as Threadripper. From nowhere, Intel is faced with some real competition. Its reaction feels like a loss of composure.
Intel also announced a pair of very odd quad-core models, which also slot into this new platform, but are actually based on the existing Kaby Lake mainstream quadcore CPU die, complete with integrated graphics. Except the integrated graphics will be disabled in this application. Like we said, weirdness abounds.
But what of this Core i9-7900X CPU? It’s a 10-core chip based on the Skylake-X microarchitecture—likely the CPU Intel had intended to roll out as its top offering for the new X299 platform. That’s why it’s ready for launch now, and everything above arrives later this year, at the earliest.
Across all 10 cores, the base clock and Turbo speeds are 3.3GHz and 4.3GHz respectively. However, the revised favored core mode, also known as TurboMax, can crank up the clocks on the two highest performing cores to 4.5GHz. The idea is to ensure that these multicore models deliver the best of both worlds—lots of cores and threads for multithreaded software, and high clocks when performance depends on one or two really intensive threads. The equivalent clock speeds for the old 10-core Core i7-6950X are 3GHz, 3.5GHz, and 4GHz.
In terms of features, one headlinegrabber is the addition of AVX-512 support. That was previously the preserve of Intel’s big-ticket Xeon processors, to which all these new Core chips are, of course, closely related. AVX-512 is part of a long line of floating-point extensions to the x86 instruction set, which date right back to the likes of MMX and SSE. In reality, few, if any, desktop applications currently support AVX-512. But the feature is now available, and that will change over time.
More important, however, are revisions to the cache memory hierarchy. The obvious change amounts to more L2 cache and less L3 cache per core. However, the details involve a shift from inclusive to non-inclusive cache, which in turns means a shake-up in terms of policy, latency, and efficiency. In short, it will take time for the full implications of the changes to the cache to shake out, but very likely some apps will benefit significantly more than others. It’s a fairly radical change.
Unpicking all that will take time. What we can initially say is that the big jump in the all-cores Turbo clock speed certainly contributes to a healthy uptick in performance of around 20 percent in multithreaded apps, such as Cinebench. Factor in the $999 price tag, where the old 6950X was a $1,700 chip, and the new 7900X looks very compelling. It is the fastest PC processor yet, and by some margin. But that’s only for now. When AMD’s Threadripper arrives, that $999 figure won’t look nearly so clever.