Testing DMA read speed and checking the USB path
How to benchmark a DMA card's read rate, what a healthy result looks like, and the four things that cost you most of your throughput.
Short answer
Run your firmware's read-speed tool from the second PC, with the card in its data port and nothing else on that USB controller. A healthy result is tens to low hundreds of MB/s, depending on card class and firmware. Anything far lower almost always means a USB path fault: charge-only cable, front-panel header, hub or USB 2 link.
On this page6
Why measure read speed before changing anything?
Almost every "my ESP is laggy" report has one of two causes. Either the read rate is far below what the hardware can do, or a missing memory map makes the reader work far harder than necessary. Both can be measured in minutes, while guessing at either is how people spend money on the wrong layer. Buying a bigger card to fix either problem is the expensive version of that mistake.
How to run a clean read-speed test
The test must be clean, because a benchmark taken while a webcam and external drive share the controller measures your webcam, not your card.
- Boot the game PC first and let it reach the desktop
- Confirm the cable runs from the card's data port to a rear USB 3 port on the second PC
- Unplug every USB device on the second PC that the test does not need
- Close the game and any capture or streaming software on the game PC
- From the second PC, run the read-speed tool supplied with your firmware
- Write down the figure, then cold-boot the game PC and run the test once more for a second data point
What does your result mean?
Absolute numbers vary by card class, firmware and tool, so treat these as bands rather than targets: what matters is which band you land in.
| Band | What it usually means | What to do |
|---|---|---|
| Far below expectation, single-digit MB/s | The link negotiated USB 2, or the cable is charge-only | Swap the cable, change port, verify the link shows as USB 3 in Device Manager |
| Low but not absurd, unstable between runs | Shared controller contention, a hub, a dock, or a marginal cable | Go direct to a rear port, strip other devices off that controller |
| Reasonable but ESP still lags | Read rate is fine; the reader is probing badly | Generate a memory map — this is the classic MMAP symptom |
| In band and stable | The USB and card layers are healthy | Stop tuning hardware; anything left is firmware or software configuration |
If two runs disagree wildly, the instability is a result in itself, and it points to the cable or controller rather than the card.
What costs you the most throughput?
They are listed in the order they show up in our support tickets.
- A charge-only or thin USB-C cable that cannot sustain USB 3 data rates
- A front-panel USB header instead of a rear motherboard port
- A hub, dock or extension cable anywhere in the path
- Other bandwidth-hungry devices sharing the same USB controller
Tip
When does a bigger card actually help?
A bigger card rarely helps, and even then only after the USB path is proven clean and a memory map is in place. If a stable, direct USB 3 link with a good cable and a valid MMAP still falls short, FPGA headroom is your limit. At that point a higher class is a real upgrade for the features you want. Until then, a 100T on a bad cable is slower than a 35T on a good one.
Common questions about read speed
Does read speed affect detection risk?
Not directly, although very aggressive read patterns are a behavioural signal in principle. In practice, the relationship people care about, the idea that faster reads are safer, does not exist. Read rate affects how smooth your overlay is.
Should I overclock anything?
No, because there is nothing to gain and a marginal USB link is already the most common failure. In this setup, stability matters more than throughput every time.
Why did my read speed drop after a Windows update?
The usual cause is a USB controller driver rollback on the second PC. Re-check the FTDI driver and the port's negotiated speed before you assume the card has degraded.
Which tool should I benchmark with?
Use the read-speed tool supplied with your firmware order and run it from the second PC. Do not compare your result with a figure someone posted from another tool, because tools measure different things and their absolute numbers are not interchangeable.
Is a lower read rate ever fine?
Yes. A radar-only build asks far less of the link than a full ESP overlay, so a figure that limits the overlay can still suit radar. Judge the rate against the features you actually run, not against the highest number you have seen quoted.
Sources
The bands are our own measurements, taken from cards on our bench and from support tickets rather than from a published spec.
Read next
Still stuck? Open a ticket on Discord
All guides