In­tel Core i9

£910 inc VAT from

Tech Advisor - - Review -

We’re re­view­ing the first Core i9 chip – In­tel’s 10-core Core i9-7900X – as a ver­i­ta­ble CPU storm looms. Sure, Core i9 blew in as the most pow­er­ful CPU the com­pany has ever sold to con­sumers, and it’s cur­rently the fastest Core-series CPU avail­able. But an ill wind is blow­ing: AMD’s Ryzen 5 and Ryzen 7 chips of­fer stiff com­pe­ti­tion in the low end, while its mas­sive 12- and 16-core Thread­rip­per CPUs loom on the high end. Here, we walk through some of the un­der-the-hood is­sues di­rectly re­lated to per­for­mance, and then we’ll dive into the bench­marks.

Core i9, un­der the hood

Core i9 is the first new ‘Core i’ In­tel has in­tro­duced in 10 years. The com­pany guarded the se­cret so closely that it even in­ten­tion­ally mis­la­belled the chips (in­clud­ing our re­view sam­ple) as ‘Core i7’ to throw off leak­ers. In fact, the re­view sam­ple you see here still iden­ti­fies it­self as Core i7 rather than a Core i9.

Like most ma­jor In­tel launches, the Core i9 fam­ily rep­re­sents a new plat­form, not just a new CPU, which means a new chipset, the X299, and a new socket, the LGA2066, all in­com­pat­i­ble with pre­vi­ous CPUs.

The new plat­form also does some­thing no pre­vi­ous one did by uni­fy­ing two CPU fam­i­lies. Be­fore to­day, if you wanted the com­pany’s lat­est Kaby Lake core, you had to buy a moth­er­board us­ing the LGA1151 socket. And if you wanted to buy, say, a 6-core Sky­lake CPU such as In­tel’s Core i7-6800K, you had to buy an LGA2011 V3-based moth­er­board.

With X299 and LGA2066, you can now pick your poi­son, be­cause the plat­form en­com­passes ev­ery­thing from a 4-core Core i5 Kaby Lake CPU to an 18-core Core i9 Ex­treme Edi­tion, which is a Sky­lake CPU. For clar­ity’s sake, the Kaby Lake CPUs, also called Kaby Lake-X, are the Core i5-7640X and the Core i7-7740X. The rest of the Core i7 and Core i9 CPUs are Sky­lake, col­lec­tively called Sky­lake-X.

This union has been greeted with some con­fu­sion and trep­i­da­tion. It’s likely that X299 mother­boards will be ex­pen­sive. Some are rightly won­der­ing who would buy a £350 moth­er­board to in­stall a £250 CPU.

In­tel’s mo­tives for the Kaby Lake-X may ac­tu­ally be a nod to the over­clock­ing sports. Un­like LGA1151-

sock­eted Kaby Lake chips, Kaby Lake-X chips have no in­te­grated graph­ics ca­pa­bil­ity. In fact, we’ve been told the chips phys­i­cally have no IGP on the die at all. This al­lows the two Kaby Lake-X chips to over­clock po­ten­tially far higher than the LGA1151 ver­sions. At the re­cent Com­pu­tex show in Taipei, in fact, In­tel said a record was set for the high­est over­clock of a Kaby Lake chip, and it was on X299.

In a per­fect world, we’d all have 18-core CPUs, but the truth is there are those who buy cheap CPUs on nice mother­boards. Kaby Lake-X is for you. PCIe ra­tioning Still, hav­ing Kaby Lake-X and Sky­lake-X on the same socket is bound to cre­ate con­fu­sion. Ex­hibit A is the PCIe ra­tioning. With the Core i9-7900X, for ex­am­ple,

you get quad-chan­nel mem­ory sup­port and 44 PCIe Gen 3 lanes di­rectly from the CPU. If you were to drop a Core i7-7740K into the same build, the moth­er­board drops down to dual-chan­nel mem­ory sup­port and, per­haps worse, the PCIe lanes drops down to 16 lanes be­cause the Kaby Lake core doesn’t sup­port more. That means some of the slots on a moth­er­board would fall back in per­for­mance or be com­pletely dis­abled.

While Kaby Lake-X’s 16-lane limit is due to the CPU’s de­sign, In­tel di­als back PCIe lanes on Sky­lake-X in­ten­tion­ally. Rather than the 44 lanes the 10-core ver­sion gets, the 6-core and 8-core ver­sions of Sky­lake-X get just 28 lanes. From what we un­der­stand, there’s no tech­ni­cal rea­son for it, just “mar­ket seg­men­ta­tion,” which is a busi­ness school way of say­ing, “so we can charge you more.”

In­tel VROC

Even more con­tro­ver­sial than PCIe ra­tioning is In­tel’s VROC, or Vir­tual RAID on CPU. It’s a nifty fea­ture on Sky­lake-X that al­lows a user to con­fig­ure up to 20 NVMe PCIe drives in RAID into a sin­gle bootable par­ti­tion. The prob­lem? It seems that In­tel in­tends to charge more money for the fea­ture. De­tails haven’t been re­leased, but ven­dors have told us they be­lieve RAID 0 would be free, RAID 1 to cost around £99, while RAID 5 and RAID 10 could cost around £299. You’ll get a hard­ware don­gle that un­locks the fea­ture.

It gets worse: VROC will work only with In­tel SSDs and pricier Sky­lake-X parts. If you buy Kaby Lake-X, you’re shut out. VROC also ap­plies only to PCIe RAID that runs di­rectly through the CPU’s PCIe lanes.

X299 still sup­ports var­i­ous RAID 0,1, 5, 10 through the chipset, but the chipset RAID won’t touch the per­for­mance you get from VROC.

How Core i9 changes Sky­lake

Once you’ve got be­yond the plat­form con­fu­sion and con­tro­versy, there’s a re­ward. The Sky­lake-X chip it­self is in­deed some­thing to ad­mire, be­cause it’s built un­like any pre­vi­ous high-end In­tel con­sumer chip.

Pre­vi­ous ‘en­thu­si­ast’ or ‘ex­treme’ CPUs have mostly been the same. That is, a 4-core Haswell Core i7-4770K wasn’t all that dif­fer­ent from an 8-core Haswell-E Core i7-5960X ex­cept for the sup­port of quad-chan­nel RAM.

With Sky­lake-X, In­tel breaks from tra­di­tion, with some ma­jor tin­ker­ing un­der the hood. The most

no­tice­able is an in­crease in Mid-Level Cache (MLC), or L2 cache: In­tel has quadru­pled it to 1MB per core, up from 256MB in last year’s Broad­well-E and the ma­jor­ity of In­tel’s CPUs. The Last-Level Cache (L3) ac­tu­ally gets smaller, with 1.375MB per core ver­sus the 2.5MB of the pre­vi­ous Broad­well-E chip, but In­tel com­pen­sates with the larger MLC and also the use of a non-in­clu­sive cache de­sign. Com­pared to the in­clu­sive de­sign in Broad­well-E, which might keep data that’s not needed, non-in­clu­sive cache at­tempts to track what should be in the cache so it can more ef­fi­ciently use the avail­able space.

In­tel also ditches the ring bus ar­chi­tec­ture it has used for sev­eral years (in­clud­ing Kaby Lake and Sky­lake) for a new mesh ar­chi­tec­ture. If you think of a quad­core CPU as four homes con­nected by a bus line that

makes stops at each home, it’s per­fectly fine un­til you add, say, 12 or 18 homes to the com­mu­nity. You could con­nect two bus lines to­gether, but that still isn’t as fast as sim­ply driv­ing from one home to the next, which is what the new mesh ar­chi­tec­ture does.

In­tel’s use of a mesh de­sign clearly puts it on a bet­ter foot­ing to com­pete with Thread­rip­per, as more and more cores are added to CPUs. AMD’s Ryzen series uses some­thing it calls an In­fin­ity Fab­ric, which is es­sen­tially a su­per-high-speed mesh net­work.

The last fea­ture worth not­ing is the im­proved Turbo Boost Max 3.0. This is the fea­ture wherein In­tel iden­ti­fies the ‘best’ CPU core at the fac­tory and gives it a lit­tle more boost speed. With Broad­well-E CPUs, only one core was cho­sen. With Sky­lake-X, two cores are iden­ti­fied as the ‘best’ and al­lowed to run a cou­ple of hun­dred mega­hertz faster.


In­tel sent us the Core i9-7900X in an Asus Prime X299-Deluxe moth­er­board. We ran the test bed with the An­niver­sary Up­date build of Win­dows 10. Yes, we know, the world has moved on to the Cre­ators Up­date, but in or­der to com­pare it with past CPUs we stuck with this ear­lier build.

All of the sys­tems (ex­cept where noted) used a GeForce GTX 1080 Founders Edi­tion, 32GB of DDR4/2133 RAM, and HyperX 240GB Sav­age SATA SSDs. For our Adobe Pre­miere CC 2017 test, the source project and the tar­get drive used a Plex­tor M8pe PCIe SSD in all but the Core i5 and the Ryzen 5 CPUs. This ex­cep­tion is due to a prob­lem with the Ryzen 5’s

moth­er­board, which failed to rec­og­nize the Plex­tor drive. A Sam­sung 960 Pro NVMe SSD was swapped in.

Where we sourced from our pre­vi­ous tests, those tests used the same Nvidia driv­ers, the same OS, and the same hard­ware that we used for this Core i9-7900X re­view. We did, how­ever, de­cide to up­date the test bed for the orig­i­nal 10-core Broad­well-E Core i7-6950X. That test was orig­i­nally con­ducted on a very early As­rock X99 moth­er­board that didn’t fully sup­port In­tel’s new Turbo Boost Max 3.0 tech­nol­ogy. This time around, we used the same Asus X99-Deluxe II that we used for test­ing the two Broad­well-E chips.

Cinebench R15

Our first test is Maxon’s Cinebench R15. It’s a free bench­mark based on the same ren­der­ing en­gine used in Maxon’s Cine­ma4D prod­uct. It scales well with core count and fre­quency and is pretty much a pure CPU test. The re­sults speak well for the 10-core CPUs when com­pared to the 8-core parts. Even though we’re in­creas­ing thread count by only 22 per­cent, we’re see­ing al­most a 30-per­cent in­crease in per­for­mance.

The dif­fer­ence be­tween the 10-core Broad­well-E Core i7-6850X and the 10-core Sky­lake-X Core i9-7900X is less than ex­pected. Ac­cord­ing to In­tel, you might see up to a 10 per­cent dif­fer­ence in mul­ti­threaded tasks and up to 15 per­cent in sin­gle-threaded tasks when com­pared to the Broad­well-E 10-core. In Cinebench, we’re see­ing just about 3.5 per­cent.

What changed? The moth­er­board. What we’re likely see­ing is a re­sult of more than a year of tun­ing

by Asus of its X99 plat­form. It just pushes the CPU far harder and far faster than the first moth­er­board. Our ini­tial re­view of the CPU in this test gave it a score of 1,792, which is quite a bit off from the 2,107 we’re see­ing from it now. Other ini­tial re­views put the chip in the low-1,800s. If that score re­mained true, Sky­lake-X would be al­most 20 per­cent faster than Broad­well-E.

We also ran Cinebench R15 limited to just a sin­gle thread. Be­cause the vast ma­jor­ity of ap­pli­ca­tions and games still rely on a sin­gle thread, the per­for­mance here is just as im­por­tant as it is on mul­ti­threaded tests. The 10-core Broad­well-E now drops back a few spots, as its clock speeds can’t keep up with the 10-core Sky­lake-X chip’s. You can also see that the CPUs with

the higher clock speeds move ahead of the 6- and 8-core chips. All ex­cept for the 10-core Core i9-7900X.

Large 8- and 10-core chips have had trou­ble keep­ing up with the spry quad-cores in high clock speeds. In­tel started fix­ing that in Broad­well-E, but if this Cinebench re­sult holds true, Sky­lake-X has the po­ten­tial to hang with Kaby Lake just fine. WinRAR Mov­ing on to com­pres­sion tests, we used WinRAR’s built-in bench­mark to mea­sure the com­pres­sion per­for­mance of the var­i­ous chips. We no longer break out the per­for­mance of the Ryzen 5 1600X and the Core i5-7600K CPUs. That’s be­cause both of those were tested with the RAM set at DDR4/2933. Mem­ory band­width doesn’t mat­ter that much in 3D ren­der­ing tests, but it def­i­nitely can tilt the scales in com­pres­sion tests. Rather than cloud the re­sults, we’re drop­ping

them. One thing you’ll no­tice is that the 10-core Core i9 sud­denly takes sec­ond place to the 10-core Core i7 pro­ces­sor.

We also tried this test with the lat­est beta ver­sion of WinRAR and saw no change. We sur­mised this might be the cache de­sign of the new chip, but after talk­ing with In­tel, the com­pany sug­gested it could be the new mesh de­sign. Hand­brake Our en­cod­ing test uses the free Hand­brake to con­vert a 30GB 1080p MKV file us­ing Hand­brake 0.9.9 An­droid Tablet pre­set. The test is mul­ti­threaded and scales well with clock speed. The win­ner is the Core i9-7900X, which comes in about 10 per­cent faster than the 10-

core Core i7-6950X. We’re also see­ing nice scal­ing: the 10-core is about 30 per­cent faster than the 8-core Core i7-6900K and 60 per­cent faster than the 6-core Core i7-6800K. 3DMark Fire Strike For gam­ing per­for­mance, we first run Fu­ture­mark’s 3DMark Fire Strike. We’re re­port­ing only the physics por­tion of the test, as that’s the only one that mat­ters for the CPU. The test uses a real-world physics en­gine that scales well with core count. Oddly, the Core i7-6950X nudges the Core i9-7900X out of the way, per­haps be­cause of the cache dif­fer­ence be­tween the chips or the mesh ar­chi­tec­ture. Note, though, that this is a the­o­ret­i­cal test of what a game could do if it stressed all those cores. In re­al­ity, games don’t de­vote this much to game physics.

Tomb Raider Mov­ing on to a real game, we use Ubisoft’s older Tomb Raider to mea­sure CPU per­for­mance by run­ning the game at 1920x1080 res­o­lu­tion and the nor­mal pre­set. At this low game set­ting and rel­a­tively low res­o­lu­tion for a GeForce GTX 1080, the only dif­fer­ence we’re likely see­ing is clock speed. Each CPU’s cache can oc­ca­sion­ally move the nee­dle, too.

The Ryzen 7 1800X chip per­for­mance is off, likely due to code that isn’t op­ti­mized for its mi­cro ar­chi­tec­ture. Case in point, Rise of the Tomb Raider re­cently re­ceived an up­date that greatly helped Ryzen out. And for the most part, it’s not an is­sue at higher game set­tings where the GPU is the bot­tle­neck on per­for­mance.

You can also see from our re­sults why Ryzen’s per­for­mance was so con­fus­ing: CPUs don’t mat­ter

in con­ven­tional gam­ing as much as peo­ple wish they would.


If you point your eyes at that last chart, which in­cludes just about ev­ery In­tel de­sign rep­re­sented since Sandy Bridge, you can only con­clude that the new Core i9-7900X is the fastest con­sumer CPU ever pro­duced by In­tel. There’s just no ar­gu­ment. The fact that it’s be­ing of­fered at £910 com­pared to the £1,600 tag on the pre­vi­ous 10-core part is an­other rea­son for the PC com­mu­nity to cheer.

The prob­lem is, it’s a dif­fer­ent world now. At £910 for a 10-core chip, you’re pay­ing about 100 per­cent over an 8-core Ryzen 7 chip for about 30 per­cent more per­for­mance. Even worse, we still don’t know what price AMD set for its 12-core and 16-core Thread­rip­per chips. If AMD in­tro­duces a 12-core CPU at £850, as some pre­dict, a 10-core Core i9 for £910 loses its lus­tre. For now, the Core i9-7900X reigns as the fastest con­sumer CPU on the planet. But it should be look­ing over its shoul­der, as will we, for Thread­rip­per. Gor­don Mah Ung


• 10 cores • 20 threads • 3.3GHz Pro­ces­sor Base Fre­quency • 4.3GHz Max Turbo Fre­quency • 13.75MB L3 cache • 8GT/s DMI3 bus speed • 4.5GHz In­tel Turbo Boost Max Tech­nol­ogy 3.0

• 140W TDP • 128GB max mem­ory size • DDR4-2666 mem­ory type • 4 mem­ory chan­nels • In­tel Op­tane Mem­ory Sup­ported • In­tel Turbo Boost Max Tech­nol­ogy 3.0 • In­telTurbo Boost Tech­nol­ogy 2.0 • In­tel Hy­per-Thread­ing Tech­nol­ogy • In­tel Vir­tu­al­iza­tion Tech­nol­ogy • In­tel Vir­tu­al­iza­tion Tech­nol­ogy for Di­rected I/O • In­tel 64 • 64-bit in­struc­tion set • SSE4.1/4.2, AVX 2.0, AVX-512 • En­hanced In­tel SpeedStep Tech­nol­ogy

Sky­lake is very dif­fer­ent from Sky­lake-X, and much of that has to do with the cache, AVX512, and a new mesh in­ter­face

AVX 512 in the Sky­lake-X prom­ises far more per­for­mance, but only if the code sup­ports it

The Core X series is made up of CPUs con­structed with Sky­lake-X cores and Kaby Lake-X cores. The mon­ster 18-core part is due in Oc­to­ber

Newspapers in English

Newspapers from UK

© PressReader. All rights reserved.