All Posts by Date or last 15, 30, 90 or 180 days.
also by Lloyd: photography and
Thank you for purchasing through links and ads on this site.
diglloyd Deal Finder...
Buy other stuff at
Capacities up to 56TB and speeds up to 1527MB/s
Handpicked deals...
$30 $15
SAVE $15

$2998 $2498
SAVE $500

$1199 $920
SAVE $279

$1999 $1599
SAVE $400

$2799 $2399
SAVE $400

$89 $59
SAVE $30

$400 $280
SAVE $120

$1798 $1598
SAVE $200

$3297 $2797
SAVE $500

$3397 $2797
SAVE $600

$150 $90
SAVE $60

$1398 $898
SAVE $500

$3698 $2998
SAVE $700

$1799 $1329
SAVE $470

$1999 $1199
SAVE $800

$2299 $1599
SAVE $700

$2399 $2049
SAVE $350

$2799 $1899
SAVE $900

$997 $897
SAVE $100

$2099 $1699
SAVE $400

$1999 $1369
SAVE $630

$1349 $949
SAVE $400

$4499 $3999
SAVE $500

$329 $329

$1499 $1029
SAVE $470

$1499 $1289
SAVE $210

$2199 $1999
SAVE $200

$3399 $2199
SAVE $1200

Meltdown and Spectre CPU Architecture Bugs: Apple Has Partially Addressed with No Performance Impact?

It is my working theory that recent security updates have broken performance on some Macs, both 10.12.6 and 10.13.2. By recent, I mean over the past 8 weeks or so. It is also suspicious that compatibility is broken with some applications, but perhaps that is just the usual Apple Core Rot.

Weird compatibility and performance bugs that I only noticed in the past few weeks to me suggest strongly that the Apple mitigations for Meltdown and Spectre CPU architectural bugs have degraded some software—including my own—more on that below.

Specifically, macOS 10.13.2 and macOS 10.12.6 (16G1114) have compatibility and performance issues that I cannot recall ever observing before. This has been nagging at me for some weeks and why I released performance updates for diglloydTools IntegrityChecker.

As it turns out, I have discovered that the IntegrityChecker performance improvements are now highly machine specific and may be worse on some machines with my updates. Stranger still, the Java version of IntegrityChecker can now (sometimes) outperform the native version by up to 3X at times—bizarre at best—and yet my C++ source code and Java code have not changed. Both are now highly sensitive to thread count and buffer count, and this was never the case before. Buffers for I/O have to be mapped into real memory and that is precisely a situation where kernel vs application address space come into play.

Never before have I been at a total loss as to how to optimize either my C++ or Java code—I’m working on quicksand now. Yet this issue only emerged in recent weeks, right after the updates to 10.13 and 10.12.6. Ten years of solid performance for IntegrityChecker, then a sudden change up to 3X slower? This cannot be random chance.

Meltdown and Spectre

The Meltdown and Spectre CPU bug fixes almost certainly prompted Apple to work overtime over the past few month, and Apple has stated that macOS is already partially secure. As quoted below, it is a misleading statement to claim no performance reductions based on benchmarks, since standardized benchmarks invariably fail to accurately predict real world performance for any particular program or workflow (as my years of testing show without fail).

Benchmarks correlate with real-world performance but only ignorance can excuse anyone claiming they do more than correlate with any particular workflow performance. Since I do not think Apple is ignorant, that necessarily leads me to say that Apple is avoiding discussing the issue and not telling the whole truth.

From About speculative execution vulnerabilities in ARM-based and Intel CPUs, emphasis added:

Meltdown is a name given to an exploitation technique known as CVE-2017-5754 or "rogue data cache load." The Meltdown technique can enable a user process to read kernel memory. Our analysis suggests that it has the most potential to be exploited.

Apple released mitigations for Meltdown in iOS 11.2, macOS 10.13.2, and tvOS 11.2. watchOS did not require mitigation.

Our testing with public benchmarks has shown that the changes in the December 2017 updates resulted in no measurable reduction in the performance of macOS and iOS as measured by the GeekBench 4 benchmark, or in common Web browsing benchmarks such as Speedometer, JetStream, and ARES-6.

Apple necessarily would not have even mentioned changes in the release notes prior to early January 2018, for security reasons. It is after most recent updates when I started seeing odd behaviors across machines.

MPG’s view is that the “average user” doesn’t read this blog, but that professionals do and that high performance computing might well be strongly affected. As per Intel Issues Updates to Protect Systems from Security Exploits, emphasis added:

Intel (INTC) said that "for the average user," the performance impact on products using the processors from the last five years "should not be significant and will be mitigated over time."

More on Spectre and Meltdown

But The Register broke the news in early January, and now we see that Apple has already been hard at work to deal with the issue.

OWC Thunderblade Thunderbolt 3 SSD
Gen 2!

Blazing fast, 1TB, 2TB, 4TB, 8TB.

Lloyd’s all-time favorite SSD!

√ No more slow and noisy hard drives!
OWC Thunderbolt 3 Dock
Ideal for any Mac with Thunderbolt 3

Dual Thunderbolt 3 ports
Gigabit Ethernet
5K and 4K display support plus Mini Display Port
Analog sound in/out and Optical sound out

Works on any Mac with Thunderbolt 3
USB-C Travel Dock

Fast charging with up to 100W!

HDMI, SD card reader,
USB-C port, 2 USB Type-A ports
Built-in cable self-stores neatly.
See also OWC 14-port Thunderbolt 3 Dock"

Deals Updated Daily at B&H Photo
View all handpicked deals...

Pelican 3rd-Generation 1910B LED Flashlight (Black)
$30 $15
SAVE $15 | Terms of Use | PRIVACY POLICY
Contact | About Lloyd Chambers | Consulting | Photo Tours
Mailing Lists | RSS Feeds | Twitter
Copyright © 2019 diglloyd Inc, all rights reserved.
Display info: __RETINA_INFO_STATUS__