MacInTouch Amazon link...

benchmarks and benchmarking

Channels
Products


Ric Ford

MacInTouch
I bought a 1TB Crucial MX500 SATA SSD, checked its integrity with SoftRAID Certify, formatted an unencrypted HFS+ partition, and ran some tests.

USB 3.0 (5Gbps) via Sabrent USB 3.0-SATA adapter* (iMac 5K, Anker USB 3 hub):

TestWrite (MB/s)Read (MB/s)
AJA System Test Lite, 4GB file426431
Blackmagic Disk Speed Test, 5GB file418426
* Unfortunately, I've misplaced my newer 10Gbps version, the Sabrent USB 3.1 (Gen 2) SATA adapter, so I just ordered another.
 


Ric Ford

MacInTouch
Here's a 1TB Samsung 970 EVO NVMe SSD in a 10Gbps Sabrent USB-C NVMe enclosure1. 2. 3., connected to the 2017 iMac 5K's USB-C port:

TestWrite (MB/s)Read (MB/s)Rewrite (MB/s)
AJA System Test Lite, 4GB file986982
Blackmagic Disk Speed Test, 5GB file936950
Disk Sensei10011001447

What happens when we move that same storage device to a 2015 MacBook Pro with its 5Gbps USB 3.0 port?

TestWrite (MB/s)Read (MB/s)Rewrite (MB/s)
AJA System Test Lite, 4GB file428427
Blackmagic Disk Speed Test, 5GB file415414
Disk Sensei441440219

Note 1: I can't see SMART data for this enclosure, either with macOS 10.12 Sierra or macOS 10.14 Mojave —I tried DriveDX, Disk Sensei, SoftRAID, Disk Utility and SMART Utility. Trim doesn't seem to be enabled, either.

Note 2: The external temperature of the aluminum enclosure was about 105°F while doing a SoftRAID Certify operation but rose above 120°F (quite hot to the touch) while ambient temperature was about 73°F.

Note 3: Benchmarks were essentially the same when the SSD was moved to a Fledging Shell enclosure (which has an audible, built-in fan).
 


I've experimented some with the same 1TB Samsung 870 EVO NVMe SSD, putting it in a TekQ Rapide Thunderbolt 3 portable external SSD. It runs rather hot to the touch, also, and it can be throttled after a bit of use. This is running on an iMac Pro. (I'm curious whether the 870 Pro version would have behaved more like Apple's built-in SSD in this enclosure, e.g., faster writes, no throttling.)

The 870 EVO does get listed as having Trim support when viewing Apple's System Information for NVMExpress devices, but Apple does not see it as having SMART support. DriveDx however can access its SMART information, as can SMART Reporter.

TestWrite (MB/s)Read (MB/s)
AJA System Test Lite, 4GB file1882
(1282 when throttled
after a few runs)
2454
Blackmagic Disk Speed Test, 5GB file1911
(1562 when throttled
after a few iterations)
2490

Apple's built in 1TB SSD indicates both Trim support and SMART status. DriveDx and SMART Reporter concur.

TestWrite (MB/s)Read (MB/s)
AJA System Test Lite, 4GB file2623 2393
Blackmagic Disk Speed Test, 5GB file27992419
 


Ric Ford

MacInTouch
Geekbench 5 debuted today. Unfortunately (for me), it requires macOS 10.13 High Sierra or later on the Mac.

Currently 50% off, I purchased a "Pro" license but haven't gotten that set up yet.

Results numbers are very different (lower) vs. Geekbench 4 numbers).
 


Geekbench 5 debuted today. Unfortunately (for me), it requires macOS 10.13 High Sierra or later on the Mac. Currently 50% off, I purchased a "Pro" license but haven't gotten that set up yet. Results numbers are very different (lower) vs. Geekbench 4 numbers).
Yes, GeekBench 5 scores annihilate the Classic Mac Pro 5,1 scores for CPU, which I don't understand, because they are 12-Core/24-thread machines, which should still make them fairly competitive. The Metal Score and OpenCL score are actually higher, so thank God for RX580 graphics cards!
 


Ric Ford

MacInTouch
Yes, GeekBench 5 scores annihilate the Classic Mac Pro 5,1 scores for CPU, which I don't understand, because they are 12-Core/24-thread machines, which should still make them fairly competitive. The Metal Score and OpenCL score are actually higher, so thank God for RX580 graphics cards!
I'm seeing bizarrely bad GeekBench 5 Compute scores for a 2017 MacBook Air, below the scores for an iPhone SE, for example. (Unfortunately, there's something wrong with Geekbench's results storage/display, and I can't see my Geekbench 4 results now. )-:
 


Ric Ford

MacInTouch
USB 3.0 (5Gbps) via Sabrent USB 3.0-SATA adapter* (iMac 5K, Anker USB 3 hub):

TestWrite (MB/s)Read (MB/s)
AJA System Test Lite, 4GB file426431
Blackmagic Disk Speed Test, 5GB file418426
* Unfortunately, I've misplaced my newer 10Gbps version, the Sabrent USB 3.1 (Gen 2) SATA adapter, so I just ordered another.
I just received the Sabrent USB 3.1 (Gen 2) SATA adapter, and it improves performance significantly for the Crucial MX500, despite SATA's limitations:

TestWrite (MB/s)Read (MB/s)
AJA System Test Lite, 4GB file469523
Blackmagic Disk Speed Test, 5GB file457513
DriveDx shows a maximum MX500 temperature so far of 154°F. It's currently at 113°F while doing a 700GB Carbon Copy Cloner backup to a clean FileVault2 volume, writing > 4 GB/min.

Updates:
  1. The MX500 temperature is now up to 136°F after writing 480 GB.
  2. Later, mostly reading hundreds of gigabytes (writing only a few gigabytes), DriveDx shows an internal temp of 122°F, while the hottest part of the case is 99°F (says my Raytek Mini Temp).
  3. Carbon Copy Cloner can do a Backup Health Check of 700 GB in about an hour on the iMac 5K (with SSD, not fusion drive) to the MX500 on the 10Gbps USB-SATA adapter. (This is actually a test of the adapter quality/integrity.)
 


Ric Ford

MacInTouch
I'm seeing bizarrely bad GeekBench 5 Compute scores for a 2017 MacBook Air, below the scores for an iPhone SE, for example. (Unfortunately, there's something wrong with Geekbench's results storage/display, and I can't see my Geekbench 4 results now. )-:
I guess the Geekbench folks fixed something — I'm getting my Geekbench 4 results again now, and they aren't as weird as Geekbench 5 with the MacBook Air; Geekbench 4 shows the 2017 MacBook Air faster than the iPhones in Compute results (while being slower than the 2018 Mac Mini and 2015/2018 MacBook Pros).
 


My GeekBench 5 scores for my 2012 Mac Pro 5.1, Dual 3.46Ghz CPU are:

Single-Core667
Multi-Core6616
OpenCL41317
Metal39336

My GeekBench 4.2.2 scores are:

Single-Core3000
Multi-Core27112
OpenCL132160
Metal134260
 


Ric Ford

MacInTouch
Geekbench "calibration" seems unnecessarily confusing:
Primate Labs said:
Interpreting Geekbench 5 Scores
Geekbench 5 CPU scores are calibrated using an Intel Core i3-8100 processor as a baseline [at 1000 points?!?].

Interpreting Geekbench 4 Scores
Geekbench 4 CPU and Compute scores are calibrated using a Microsoft Surface Book with an Intel Core i7-6600U processor as a baseline with a score of 4,000 points.

Interpreting Geekbench 3 Scores
Geekbench 3 scores are calibrated using a Mac mini (Mid 2011) with an Intel Core i5-2520M @ 2.50 GHz processor as a baseline with a score of 2,500 points
I wonder if Poole is purposely making the numbers very different from version to version to prevent people from inadvertently comparing results from one version against another when the benchmark programs are so different. But I haven't see any clear explanation of any of this, so far.
 


Ric Ford

MacInTouch
Here's some backup performance data for what it's worth, an example of backing up a full production system that also includes a lot of archival files to an empty external SSD:

System:​
2015 MacBook Pro 15-inch 1TB, macOS 10.12.6, Carbon Copy Cloner 5.1.14​
Target:​
2TB Samsung T5​
Size:​
1.9 million files (all sizes)​
774 GB
Time:​
72 min.​

For comparison, here's an incremental backup of the same system to a roughly equivalent backup SSD:

Target:​
2TB Samsung 850 EVO, USB 3 enclosure​
Size:​
2331 files​
5.6 GB
Time:​
6.8 min.​

#backup #benchmarks
 


Data point, for what it's worth: My main drive is a 1TB SSD with 457GB of space used. (I don't see a file count under Mojave.) Carbon Copy Cloning to an SSD takes 9-11 minutes. To a spinner, it would usually take 30-40 minutes.

This is an daily incremental backup of a source drive that is 1TB nVME with 457 GB of files*; destination is EVO 860 located in "toaster" dock connected via USB 3.

*4.6 million of them; about 45,000 files copied; about 8 GB.

#backup
 


Ric Ford

MacInTouch
GeekBench 5.1 updates the popular benchmarking utility (with its online database) but produces somewhat different results vs. the previous release.
Primate Labs said:
For the Compute Benchmark, Geekbench 5.1 fixes several issues with the Metal and Vulkan implementations of the Compute workloads. These fixes improve the performance (and increase the scores) of the Compute Benchmarks on Metal- and Vulkan-powered GPUs.

The changes to the CPU and Compute Benchmarks can change their scores, so we recommend exercising care when comparing Geekbench 5.0 scores with Geekbench 5.1 scores.
Even more confusingly, GeekBench 5's results/scaling are radically different from GeekBench 4 results and cannot be compared directly, which is problematic when trying to compare a computer that only supports the newer version (which requires macOS 10.13 or later) vs. a computer that doesn't.
 


Ric Ford

MacInTouch
Here's a 1TB Samsung 970 EVO NVMe SSD in a 10Gbps Sabrent USB-C NVMe enclosure1. 2. 3., connected to the 2017 iMac 5K's USB-C port...
... What happens when we move that same storage device to a 2015 MacBook Pro with its 5Gbps USB 3.0 port?
Samsung Portable SSD T7 Touch, with the default (unencrypted) volume and macOS 10.12 Sierra – Blackmagic Disk Speed Test, 5GB file

Write (MB/s)Read (MB/s)host connection
4034105Gbps USB 3.0 (2015 MacBook Pro 15")
83490310Gbps USB-C (2017 iMac 5K)

The Samsung T7 Touch doesn't appear to be any faster than a Samsung T5 for a computer limited to 5Gbps USB 3.0 (a.k.a. "USB 3 Gen 1"), though it's obviously much faster when you have a 10Gbps USB 3 Gen 2 host, as you can see in the numbers here and above.
 


Ric Ford

MacInTouch
Here's an interesting technical article about benchmarking storage drives using the open-source FIO (Flexible I/O tester) command-line program:
Ars Technica said:
How fast are your disks? Find out the open source way, with fio
Storage benchmarking—much like Wi-Fi benchmarking—is a widely misunderstood black art. Admins and enthusiasts have for decades been tempted to just "get the big number" by reading or writing a large amount of data to a disk, getting a figure in MB/sec, and calling it a day. Unfortunately, the actual workload of a typical disk doesn't look like that—and that "simple speed test" doesn't reproduce a lot of the bottlenecks that slow down disk access in real-world systems.

The most realistic way to test and benchmark disks is, of course, to just use them and see what happens. Unfortunately, that's neither very repeatable, nor is it simple to analyze. So we do want an artificial benchmarking tool—but we want one that we can use intelligently to test storage systems across realistic scenarios that model our day-to-day usage well. Fortunately, we don't have to invent such a tool—there's already a free and open source software tool called fio, and it's even cross-platform!

We're going to walk you through some simple but effective uses of fio on Windows, Mac, and Linux computers—but before we do that, let's talk a little bit about storage systems from a more basic perspective.
 


Amazon disclaimer:
As an Amazon Associate I earn from qualifying purchases.

Latest posts