Re­views & Rat­ings

I just patched the Sur­face Book with the Melt­down and Spec­tre patches. Here’s how much it hurt.

PCWorld (USA) - - Contents - BY GOR­DON MAH UNG

As a per­for­mance junkie, I’m less con­cerned about the se­cu­rity risks of the Spec­tre and Melt­down vul­ner­a­bil­i­ties (see page 115)—after all, there are no known ex­ploits in use to­day—than I am about a per­for­mance hit from the fixes.

And from what I’m see­ing, my con­cerns are war­ranted.

My sole ex­pe­ri­ence with a fully up­dated

plat­form so far is with Mi­crosoft’s orig­i­nal Sur­face Book. It’s based on an In­tel Sky­lake Core i7-6700u and has 16GB of LPDDR3 and a 512GB Sam­sung 950 Pro NVME drive. The Sur­face Book is run­ning the 64-bit Win­dows 10 Pro Fall Cre­ator’s Up­date.

I ba­si­cally drove the ma­chine for most of a week at CES, and on that Fri­day morn­ing when I fired it up at home, I found that Mi­crosoft had pushed out two pairs of firmware up­dates that ad­dress the Spec­tre and Melt­down vul­ner­a­bil­i­ties.

I’ll ad­mit, I’m still try­ing to play catch-up on just what the hell is go­ing on with Spec­tre and Melt­down, but this was a great op­por­tu­nity to run be­fore and after bench­marks on a pro­duc­tion ma­chine.

I know from read­ing Steve Wal­ton’s write-up at Techspot ( go.pc­ that the per­for­mance of games and most Cpu-in­ten­sive apps doesn’t change. But Wal­ton found stor­age read/write per­for­mance to be an is­sue, so this was my first point of in­ves­ti­ga­tion on the Sur­face Book. I started my tests with a not-yet­patched ma­chine.


I started by run­ning the de­fault test for Crys­tald­iskmark 5.5.0. It’s the slightly older ver­sion, but the re­sults are still valid for two ar­eas: small 4K reads and writes as well as 4K reads and writes us­ing a queue depth of 32. Above is the re­sult be­fore Sur­face Firmware 91.1926.768.0 and 90.1837.256.0 were in­stalled.

I made three re­peated runs with two to five min­utes of rest time in be­tween to let the SSD re­turn to nor­mal tem­per­a­ture. SSDS, as you may know, can slow when heated. This is one par­tic­u­lar re­sult, but rep­re­sen­ta­tive of the re­sults I saw.


Once I com­pleted my tests, I was able to re­boot the Sur­face Book and let the firmware patches in­stall.

As you can see in the next chart, the se­quen­tial read and write per­for­mance doesn’t change much (in fact, Crys­tald­iskmark no longer uses a low-queue-depth se­quen­tial test).

But just as Techspot found, 4K per­for­mance ain’t pretty. While 4K read per­for­mance was sim­i­lar, the write per­for­mance dropped by 26 per­cent.

Far worse, though, 4K read and write with high queue depth take a per­for­mance hit of 42 per­cent and 39 per­cent, re­spec­tively. Ouch.

As with the pre-patch state, I ran the test three times with sim­i­lar re­sults.


Stor­age per­for­mance hits in syn­thetic read/write tests, of course, can be dif­fi­cult to re­late to. So, to see how the patches man­i­fested in per­for­mance that’s easy to re­late to, I also ran Prin­ci­pled Tech­nolo­gies WEBXPRT 2015 on the Sur­face Book, us­ing the lat­est ver­sion of Mi­crosoft Edge. WEBXPRT 2015 is a browser-based bench­mark that mea­sures per­for­mance in var­i­ous sce­nar­ios in HTML5 and Javascript.

Un­patched, I saw an over­all score of 450. Once patched, the score dropped, though not sub­stan­tially, to 433. That’s about 4 per­cent slower. In­tel’s own find­ings with that same test show about a 10 per­cent re­duc­tion.

So is the sit­u­a­tion not as dire as the syn­thetic stor­age bench­marks make it out to be? Yes and no. We’re still very early in test­ing the patches, but it’s safe to as­sume that per­for­mance drops will be de­pen­dent on what you ac­tu­ally do with your ma­chine.

As Techspot found, most clas­sic tests (such as pure 3D ren­der­ing) and most games won’t see a change. But the greater-than-20per­cent stor­age penalty that both Techspot and I ob­served will rear its head oc­ca­sion­ally.

In fact, one test that In­tel presents is BAPCO’S SYSMARK 2014 SE. One of the most ad­vanced bench­marks around, SYSMARK uses

real-world ap­pli­ca­tions such as Word and Pho­to­shop, and then runs them through tasks that mir­ror real-world ac­tiv­ity.

Even bet­ter, BAPCO has de­vel­oped a method for test­ing where only the re­sponse time of an ac­tion is mea­sured. In Word, for ex­am­ple, old-time bench­marks would type and per­form ac­tions in Word well be­yond what even the fastest typ­ist could ever hit.

But SYSMARK 2014 SE mea­sures the things that can truly an­noy you, like how long it takes to start the ap­pli­ca­tion, or have it per­form a search and re­place, or im­port pho­tos.

As you can imag­ine, SYSMARK will lean more heav­ily on stor­age re­spon­sive­ness.

In­tel’s tests on SYSMARK 2014 SE mostly in­di­cate about an 8 per­cent over­all hit, but in the de­tails for the

Sys­tem Re­spon­sive­ness test (where you’d ex­pect the stor­age per­for­mance to mat­ter more), In­tel says its see­ing a 21 per­cent drop. Ouch.

We’ll try to in­de­pen­dently ver­ify In­tel’s re­sults on our own builds, but ev­ery­thing I’m see­ing so far says the per­for­mance penalty will prob­a­bly run the gamut from “no big deal” to “this is re­ally test­ing my pa­tience.” Again: It’ll come down to what you’re do­ing, and how you’re do­ing it.

If we’re talk­ing an ex­tra 500ms to launch an ap­pli­ca­tion that takes 1,500ms to launch, no big deal. But if we’re talk­ing 34 sec­onds to im­port or copy pho­tos in­stead of 27 sec­onds, it’s go­ing to get an­noy­ing re­ally fast and that’s what scares me.

Crys­tald­iskmark 5.5.0 per­for­mance re­sults on a Mi­crosoft Sur­face Book be­fore be­ing patched for Melt­down and Spec­tre.

Crys­tald­iskmark 5.5.0 re­sults on the Mi­crosoft Sur­face Book once it had been up­dated for the Spec­tre and Melt­down ex­ploits.

Be­fore Melt­down and Spec­tre patch in WEBXPRT 2015.

Two sets of patches for the Sur­face Book ad­dress the Spec­tre and Melt­down ex­ploits.

After Melt­down and Spec­tre patch in WEBXPRT 2015.

Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.