Accurately Determining Storage Speeds Seems Impossible

I just spent the last two weeks measuring the speed of a Thunderblade SSD RAID from OWC. After running more than 200 speed tests, what I learned is that accurately determining the speed of any storage device seems impossible.

NOTE: Here’s the product review that triggered this research.

Why? Because no two testing software presents the same results. Let me explain.

SETTING THE STAGE

Thunderbolt 3 or 4 has a maximum data transfer speed of 2850 MB/second. Think of that as the size of the “pipe.” The speed of your storage determines how full that pipe will be. Hardware is measured by how fast it writes (records) and reads (plays back). By way of comparison:

• A single new hard drive transfers data around 175 MB/second
• A single PCIe SSD transfers data around 400 MB/second
• A single NVMe SSD transfers data around 2800 MB/second. However, not all NVMe SSDs run at full speed.

These speeds can vary depending upon how the storage is designed and manufactured.

CHALLENGE 1: MEASUREMENT TOOLS DON’T AGREE

Here are four examples of different software measuring the speed of the same SSD hardware, formatted and connected the same way, on the same day.

Which one is correct? No one knows.

• Measuring writes varies by up to 1367 MB/second (roughly 47%)
• Measuring reads varies by up to 1271 MB/second (roughly 44%)

Whichever software you choose to use, there’s a real possibility that it’s wrong.

CHALLENGE 2: TRANSFERRING MULTIPLE FILES AFFECTS SPEED

Transferring multiple smaller files is far slower than transferring one large file. Why? Because each time a file is created or moved from one spot to another, the operating system needs to write both the data and update the disk directory. Directories are stored separately from the data. Transferring a single large file only updates the directory once. Transferring multiple files updates the directory each time a new file transfer starts.

When I was testing server speeds earlier this month, I could transfer a large 100 GB file at 150 MB/second to the server. Transferring 50 20 MB files to the server slowed data rates to about 15 MB/second.

While not as big a deal for SSDs, this constant shuttling between updating the directory and writing the data on an HDD seriously slows things down as the drive heads need to constantly shift between directories and data.

CHALLENGE 3: HOW A RAID IS FORMATTED AFFECTS SPEED

RAID 0 provides the fastest transfer speeds, but without any protection in the event a drive dies.

RAID 4, 5, and 1+0 provide security in the event of a drive failure, but there’s a trade-off in terms of performance of up to 40% and available storage capacity.

CHALLENGE 4: OTHER FACTORS THAT AFFECT DATA TRANSFER SPEED

The speed of transferring files also depends upon a number of additional factors:

• Codec. Some media codecs transfer faster. I used ProRes for my tests, but ProRes is not the fastest codec for transfers. However it is often the best codec for editing.
• Computer system speed. If you use a software-based RAID, the speed of your storage will increase as computer speed increases.
• Background activity. Speeds slow down as background activity increases. They can also slow down if the application doing the transferring switches to running in the background.
• Thunderbolt cable quality. Many inexpensive cables are unable to carry the full bandwidth of Thunderbolt signals.
• Operating System. It seems that Monterey is about 20% faster than Ventura when reading or writing files to SSDs. (Link) Yet Ventura is the newer software.

As media creators, we want the best performance from our system that we can get. The problem is that performance varies widely based on conditions over which we have very limited control.

It is “relatively” easy to determine a “maximum data transfer speed” under ideal conditions.. The problem is that conditions are rarely ideal for very long; especially during video editing. Which means the actual transfer speed will vary — a lot.

SUMMARY

As storage gets faster, focusing solely on performance metrics will lead you astray; not least because the software we use to measure data transfer speed is highly inaccurate.

After running hundreds of speed tests, here are my thoughts.

1. The most important takeaway is: Don’t get hung up solely on performance metrics. I’m not saying ignore them, but take them with a large grain of salt. There’s no guarantee the software you are using to measure the speed of your storage is accurate.
2. Measuring speed without considering the work you want to do is not meaningful. Data transfer speeds in excess of about 400 MB/second are more than fast enough to edit any video format up to 12K.
3. No NLE is able to import, edit, render, export or compress at the full speed your storage supports. Video editing requires a lot of calculations – each of which takes time. You can edit faster with faster storage. But don’t expect to max it out.
4. How you format a RAID makes a difference in speed and capacity.
5. SSDs are always faster than spinning hard disks, but they don’t hold as much and they cost more.
6. SSDs survive extreme abuse more easily than spinning hard disks. Expect to get 4-6 years of life from an SSD.
7. If you are doing multicam editing, SSDs are the storage of choice because they have no drive heads to move, no latency, no seek times. SSDs make multicam editing fly!
8. Be very cautious when a manufacturer says: “This device supports speeds up to 2800 MB/second!”  That phrase covers all speeds from 1 MB/s to 2800 MB/s; which is to say it is essentially meaningless.

While I don’t expect marketers to stop using the phrase “Up to 2800 MB/second,” there is something else you can rely on: Don’t make your decision solely on benchmark speeds. Make your decision based on the editing you need to do.

• Editing HD media needs storage capacity more than speed. A single hard drive, or PCIe SSD, is sufficient in most cases.
• 4K media needs more speed and much more capacity. An HDD RAID is a good option providing massive storage and sufficient speed. SSDs are fine, but often don’t provide enough capacity.
• Larger frame sizes, greater bit depth require larger storage and faster RAIDs. The Thunderblade is very useful here.
• Multicam editing benefits enormously from the speed of SSDs and SSD RAIDs. The Thunderblade, or other SSD RAIDs, is tailor-made for multicam editing.

As this chart from Blackmagic Disk Speed Test makes clear, any storage running faster than 1,200 MB/second is enough to handle all video formats up to 12K. If you’re editing uses smaller file sizes, you can use slower storage. We don’t need to fill a Thunderbolt pipe to edit successfully.

As you navigate storage websites, keep your specific editing in mind. It will help to simplify your choices. And remember that whatever storage you buy today will need to be replaced in 4-5 years. Tech keeps getting bigger and faster.

---

Bookmark the permalink.