Core-X
We’ve finally got our eager hands on Intel’s new CPU. BENNETT RING crunches the numbers...
When AMD dropped Ryzen on the market, offering the performance of Intel’s highperformance ten-cored CPUs at just half the price, nobody could have expected how quickly Intel would respond. It took them just a few months to release its competitor, the new Core-X series of CPUs, which dominated at this year’s Computex. Along with the new chip came a new chipset, the X299, and it was impossible to visit a motherboard maker’s booth without seeing products based on this. So has Intel been able to bring back its lead when it comes to multi-cored behemoths, or is the Ryzen series the best offer on the market? Also, is it really that important to gamers who aren’t streamers, given that most games only use four cores as it is? Let’s find out.
THE CORE-X FAMILY
Intel has introduced not one, but multiple versions of its new Core-X family, and they’re divided into two main architectures. From four cores up to a massive 18 cores, they’ve got all of their bases covered.
At the head of the lineup is the new i9 series, and these are the chips designed for those who need the most threads possible. The biggest of these comes with 18 cores and 36 threads of performance, and is aimed at content creators and the like. However, it’s not available yet, with the i9-7900X currently the most potent CPU on the block. This ten-cored, 20-threaded behemoth has a base frequency of 3.3GHz, which Boosts up to 4.3GHz.
Like AMD, Intel has introduced a second level of turbo boost if the thermal conditions are right, called Turbo Boost Max 3.0. In the case of the i9-7900X, this adds another 2GHz to the speed, with a maximum speed of 4.5GHz.
All of these cores means a relatively high TDP, at 140W, and Intel does not supply a cooler with this chip. They recommend water cooling if you want to overclock; air cooling if you do not. This is the only chip currently available with the full 44 PCIe lanes from the CPU, well ahead of AMD’s offering. Memory speeds for this CPU are officially rated at DDR4-2666MHz, but we’ve seen several overclockers going well beyond that speed. It also supports quad memory channels for a maximum of 128GB.
This chip is based on the Skylake-X architecture, which is based on a 14nm manufacturing process. Intel has remained mum about the number of transistors within, though it must be a significant amount. Improvements to Skylake-X include an increase of 256kB L2 cache to each core to a huge 1MB, while the bandwidth between the L1 and L2 caches has been increased to 128 bytes per cycle. Intel claims this has quadrupled the associativity of the cache from four ways in Skylake to 16 in Skylake-X. However, there’s a pay-off. The shared-last level cache has dropped from 2.5MB per core in preview architectures down to 1.375MB per core, as Intel felt the extra memory was being wasted by relying on data sharing via L3 caches. As our benchmarks show though, it hasn’t hurt performance.
In the past, Intel used a technology called ring topology to connect the CPU cores, but they’ve ditched that in favour of a new mesh interconnect technology. This apparently has several benefits, firstly being lower latency and higher bandwidth, yet operates at a lower frequency and voltage.
The new CPUs now rely upon a Socket LGA 2066 design, which means you’re going to need to upgrade your motherboard. However, the socket design is the same size as Socket 2011-3, which means your old coolers will still work fine, provided they can handle the 140TDP.
We were supplied with the i97900X, but were also given the ‘entrylevel’ Core-X, the i7-7740X, which is effectively a replacement for the i7-7700K, but gives owners the ability to upgrade to stupidly high-cored CPUs in the future. Think of it as the answer to AMD’s Ryzen 5 series, which is only six-cored, and quite affordable, but opens up an upgrade path to the Ryzen 7 down the track. Unlike the i9, this is based on the Kaby Lake-X architecture. It still uses the Socket LGA 2066 of the i9, and requires the new chipset, but it’s a vastly different offering.
For starters, this is a quad-cored CPU, with Hyper-Thread. It does not support Turbo Boost Max 3.0. It’s also had its PCIe lanes dramatically cut back to just 16 PCIe lanes from the CPU; though the chipset delivers an extra 28, but they share a DMI 3.0 connection with the CPU, which means all 28 won’t be available. It’s also only a dual channel memory board, so if you buy an X299 board with eight memory slots, only half will work. Ditto the PCIe lanes; one or more may not work if you go with a Kaby Lake-X based CPU.
When looking at the specifications of the Kaby Lake X, it strikes us how similar it is to the former Kaby Lake architecture. The main differences are that it must have the LGA 2066 form factor, even though around 1000 of the pins on the Kaby Lake X chip aren’t used. Secondly, they’re slight faster when it comes to base frequencies. However, the i7-7740X has the exact same Boost frequency of 4.5GHz as the i7-7700K. Finally, the integrated GPU in the i7-7740X has been disabled. In all other regards they seem to be identical, with none of the new mesh interconnect technology or changes to caching. In effect, you’re buying a CPU that performs, and is basically, identical to the i7-7700K, just in a new form factor. Oh, you’ll also have to pay $30 extra for the privilege.
PRICING
The big surprise about the Ryzen 7 series was its huge price advantage over Intel at the time. Intel’s ten-cored i7-6950X was selling for – and is still selling for – around $2600. For AMD to come in with a ten-cored chip that could keep up with it in most tasks for just $680 was an absolute revelation. Throw in cheaper motherboards for the Ryzen, and Intel took a beating when it came to price and performance. So how do its new chips compare?
The new ten-cored i9-7900X is currently retailing in Australia for $1,480, just over half the price of its predecessor, but still twice the price of the Ryzen 7 1800X. As for the i77740X, this is now retailing for $519, which puts it on par with the i7-7700K
remember that the X299 is for Intel, and the X399 is for AMD. We think AMD’s naming is a little bit dodgy
it’s replacing. So at the ten-core level, AMD still has a huge advantage over Intel. Its Ryzen 5 chips are also considerably cheaper, but their gaming performance hasn’t been exactly overwhelming, which is why we still recommend going with an i7-7700K if you’re serious about gaming.
There’s one other cost to factor in though – the new motherboards required for the Core-X series, all based on the X299 chipset.
THE X299
Please, please, please remember that the X299 is for Intel, and the X399 is for AMD. We think AMD’s naming practices here are a little bit dodgy, and are sure there will be many customers who end up buying the wrong chipset. Enough about that though – what does the new X299 deliver?
Formally codenamed as Basin Falls, the X299 chipset first and foremost has the LGA 2066 socket required for every single new Core-X processor. Despite some of these new Core-X chips seeming to be identical to their predecessor (cough i7-7740X), Intel has ensured that the Core-X chips will only fit within this socket.
In the past, Intel’s Z-series of chipsets have used a High Speed IO (HSIO) design that makes the chipset act like one large PCIe switch. The DMI 3.0 link from the CPU, which is basically a PCIe 3.0 x 4 link, the chipset could support up to 24 lanes of PCIe 3.0 from various configurations. This meant motherboard makers could go straight to the chipset instead of the CPU for its PCIe lanes, used for USB, M.2, Ethernet and the like.
X299 now mirrors this concept, bringing HSIO to the new chipset. In the past, motherboard makers had to buy PEX PLX switches to make the most of this scenario. However, X299 moves the DMI 2.0 speed in the X99 platform up to the DMI 3.0 speeds of the Z-series, giving them much more bandwidth to play with. This is why the i9-7900X now has 44 lanes of PCIe 3.0 available to it, on top what the chipset offers, whereas the X99 only had eight PCIe 2.0 lanes, separate to the 40 lanes direct to the graphics lanes. However, the number of lanes supported is reliant upon the CPU you buy, with either 16, 28 or 44 PCIe lanes depending on the Core-X CPU installed. Once again, this will end up with consumers having to figure out just how many devices they can install on their system before buying it, looking at confusing tables in motherboard manuals to see which features are disabled when others are in use.
Obviously X299 also supports Optane, but if you’ve read previous issues of PCPP you know how we feel about this caching technology; it holds promise, but currently is too immature for the mainstream market. X299 natively supports eight SATA 3.0 ports and up to 10 USB 3.0 ports.
X299 supports quadmemory lanes, with up to eight DIMMS for a maximum of 128GB. The speed of these will depend on the processor, between DDR4-2400 and DDR4-2666. This new chipset also supports Intel’s VROC (Virtual Raid on CPU), which supports up to 20 devices in a bootable raid partition. However, to enable this feature, you’ll need to buy an extra key that plugs into the board, which looks to us just like a header. To be frank, we feel this technology should be offered free of charge, as it’s built into every X299 board and Core-X CPU.
TWEAK IT HARD
Overclockers will appreciate the fact that the X299 brings both multiplier and base clock frequency adjustments to the table. From what we can tell, all Core-X CPUs are also