How to diagnose USB dropouts and read-rate collapse
The card works then stops, drops at random, or its read rate falls mid-session. How to tell selective suspend, a shared controller, a bad cable and a hot card apart, with evidence a supplier can use.
Short answer
Intermittent DMA failures are almost always the USB path, not the card, and have four causes: selective suspend, shared controllers, marginal cables and heat. Change one variable at a time and record when each drop happens, because the timing pattern tells them apart.
On this page11
- Are intermittent dropouts really random?
- Which cause matches your symptom?
- What is USB selective suspend?
- Do other USB devices share the card's controller?
- Could the USB cable be the problem?
- Can heat make the card drop out?
- How to capture evidence of a dropout
- How to keep a dropout log a supplier can use
- What to rule out, in order
- Common questions about USB dropouts
- Sources
Are intermittent dropouts really random?
A fault that comes and goes feels random, but it usually is not. Four causes account for nearly all such faults, and each one has a timing signature. Before you change anything, spend one session just watching. Note the clock time of each drop, what you were doing and how long the machine had been running. That one note will save you three evenings of swapping parts.
Which cause matches your symptom?
Match the shape of your failure rather than its wording, then use the right-hand column, which gives the cheapest test that tells that cause apart.
| Shape of the failure | Likely cause | Test |
|---|---|---|
| Works, then dies after a few minutes of idle, comes back after a replug | USB selective suspend or link power management putting the bridge to sleep | Turn off the power-saving checkbox for that device and the USB selective suspend setting in the active power plan, then repeat the idle |
| Drops when something else on the machine gets busy — a webcam, a capture card, an external drive | Shared USB controller: bandwidth or interrupt contention | Move the card to a port on a different controller, or unplug the other device and repeat |
| Random, unrelated to load or idle time, worse if the cable is moved | Cable or connector: charge-only, thin, too long, or a tired socket | Swap to a short known-good data cable and do not touch it for a session |
| Fails after a consistent warm-up period, every session, roughly the same number of minutes | Thermal: the card, or the bridge chip, in a case with no airflow | Run with the side panel off and air moving over the board. If the time to failure changes, it is heat |
| Read rate is low from the first second and never recovers | The link negotiated USB 2, or a hub or front-panel header is in the path | Check the negotiated speed and go direct to a rear USB 3 port |
| Everything drops the moment the game changes scene or loads | Reads into unmapped physical memory killing the path, not USB at all | Use a memory map and retest. This one is misdiagnosed as a cable constantly |
Two causes can stack, and a marginal cable on a shared controller fails in a way that looks like neither cause on its own.
What is USB selective suspend?
Windows is allowed to power down a USB device that looks idle, and USB 3 links have their own low-power states on top of that. For a mouse this makes no visible difference, but a bridge that is about to be asked for a sustained stream wakes up slowly. It wakes slowly enough that the software sees a failed transfer and gives up. Two switches need clearing: the checkbox on the device's Power Management tab in Device Manager, and USB selective suspend in the active power plan. Clear both on the second PC and retest before you suspect anything physical.
Could the USB cable be the problem?
USB-C cables are not interchangeable, because many are wired for charging only and have no SuperSpeed pairs at all. Plenty of cheap data cables also cannot hold a USB 3 link over a long run. A socket that has been pulled sideways for six months will drop out under vibration. Keep one short cable that you know is good as your reference, and test with it whenever a fault appears. If switching to it changes the behaviour, stop investigating everything else.
Can heat make the card drop out?
An FPGA and a USB 3 bridge both dissipate real power. A low-profile board pressed against a hot GPU, in a case with no side intake, has nowhere to put that heat. The sign is timing: the failure arrives after roughly the same interval every session and clears once the machine has been off for a while. We do not quote a temperature figure, because it varies by board and none of the numbers people post were measured on yours. Test the theory instead by opening the case, aiming a fan at the board and checking whether the interval changes.
How to capture evidence of a dropout
These read-only commands run on the SECOND PC. Run them before a session and again straight after a drop, and keep both outputs, because the difference between them is the evidence.
REM Which USB controllers exist, so you can see what shares what
powershell -c "Get-CimInstance Win32_USBController | Select-Object Name, DeviceID"
REM Every USB device and its current status. Look for Error or Unknown
powershell -c "Get-PnpDevice -Class USB | Select-Object Status, Class, FriendlyName, InstanceId"
REM Is the FTDI bridge still present right now?
powershell -c "Get-PnpDevice | Where-Object FriendlyName -like '*FT601*' | Format-List"
REM Recent device-installation and PnP configuration events
powershell -c "Get-WinEvent -LogName Microsoft-Windows-Kernel-PnP/Configuration -MaxEvents 40 | Format-Table TimeCreated, Id, Message -Auto"
REM Surprise-removal and driver-load failures in the System log
powershell -c "Get-WinEvent -FilterHashtable @{LogName='System'; StartTime=(Get-Date).AddHours(-6)} -MaxEvents 60 | Where-Object ProviderName -match 'USB|PnP|Disk' | Format-Table TimeCreated, Id, ProviderName"
REM Which sleep states this machine supports, which tells you whether suspend can be in play
powercfg /aHow to keep a dropout log a supplier can use
Three sessions of this log are worth more than three pages of description, because the log turns a ticket from a conversation into a fix.
- Write down the start time of the session and the machine's uptime
- Note the read-speed figure at the start of the session
- When it drops, note the clock time and what was happening: idle, loading, a scene change, or another device in use
- Copy the exact error string the software printed, as text
- Run the status and event commands and save their output next to the timestamp
- Change exactly one variable, such as the cable, the port, the power setting or the airflow, and repeat the session
- Send the log with that one change named, so whoever reads it can see cause and effect rather than a list of everything you own
What to rule out, in order
The list starts with the cheapest and most common causes, and most dropout tickets end at item three.
- USB selective suspend is off, in both the per-device checkbox and the power plan setting
- The card is plugged directly into a rear USB 3 port, with every hub, dock and extension removed
- One short known-good data cable is the reference for every later test
- The card is alone on its USB controller, with busy devices moved elsewhere
- A memory map is in use, so the read path is never walked into unmapped ranges
- The card has airflow over it, tested with the side panel off before anything is bought
- Both machines are cold-booted properly between tests, not restarted
Common questions about USB dropouts
Could a dropout mean I have been detected?
No. A USB device disappearing from the second PC is a local hardware-path event, of the same kind as an external drive dropping off. It also happens on machines with no game installed.
Would a powered hub help if I am short of ports?
A hub is still an extra link in the chain, and it is the first thing anyone will ask you to remove. Use the rear ports for the card and put everything else on the hub, not the other way around.
Does a longer cable really matter at these speeds?
Yes. USB 3 signalling has a modest length budget, and cheap cables use most of it before you add a metre. Keep the card's cable short and treat length as a variable worth testing.
My read rate halves mid-session but nothing disconnects. Is it the same problem?
It belongs to the same family: a collapse without a disconnect usually means contention on the controller or thermal throttling somewhere in the path. Use the timing log to tell them apart.
Should I update the FTDI driver to the newest version?
Use the version shipped with your order. A newer driver running against an older software build is a known source of odd intermittent behaviour, which is exactly what you are trying to eliminate.
Sources
The four causes and their timing patterns are bench work, from real drops the team matched to suspend, controller sharing, cables and heat.
- 01Windows debugging tools (Microsoft)
where the free USB tree viewer the page mentions comes from
learn.microsoft.com
Read next
Still stuck? Open a ticket on Discord
All guides