What actually goes wrong with cheap DMA hardware
The FPGA itself is rarely the problem. The usual causes are the cable, the bridge chip, the thermal design and the board revision.
By Lethal Hardware Desk
Support tickets in this niche show a pattern that has nothing to do with what people shop for: almost nobody has a broken FPGA. Instead, some have a cable that cannot carry data or a USB bridge of uncertain origin. Others have a board that throttles or drops out once it is warm, or a board that is physically not the revision the firmware was built for. Each of these problems is cheap to diagnose, yet each one gets mistaken for a dead card.
Which failures happen most often?
| Failure mode | What you see | The check that settles it |
|---|---|---|
| Cable cannot sustain USB 3 data | Device enumerates but read rate is a fraction of expected, or the link negotiates as USB 2. | Check the negotiated speed in Device Manager or lsusb -t. Then swap to a known-good short data cable and re-measure. Charge-only USB-C cables are extremely common and look identical. |
| Bad USB path rather than bad cable | Works, then drops out mid-read. Inconsistent benchmark results between runs. | Go direct to a rear motherboard USB 3 port. Remove every hub, dock and extension. Strip other bandwidth-heavy devices off the same controller. |
| Thermal throttle or dropout | Fine for ten minutes, then degrades or disconnects. Recovers after a cool-down. | Touch test after a sustained read, and check whether the failure correlates with time rather than with what you were doing. M.2 cards in a cramped slot with no airflow are the usual suspects. |
| Board revision mismatch | Flash appears to succeed, card never enumerates correctly. Or the flashing tool does not see the device. | Photograph the silkscreen: board model, revision, FPGA marking, bridge chip marking. Compare against what the firmware was built for before you flash again. |
| Wrong socket | The second PC sees nothing at all, LEDs on. | Most boards have two USB-C sockets: one for the flashing bridge, one for the data bridge. Plugging into the flashing port produces exactly the symptom people read as a dead card. |
| Questionable bridge silicon | Driver installs but behaves oddly. Works on one machine and not another. Odd descriptor contents. | Read the descriptors and the chip marking. Compare against the part the board claims to carry. This is the hardest of the six to confirm and the rarest. |
What does the USB bridge chip do?
The USB 3 bridge on most of these boards is an FTDI FT601, a SuperSpeed USB 3.0 to FIFO bridge with a 32-bit parallel interface. It turns the FPGA's data stream into something a host can read over USB. That is why the data port is a USB 3 port, and why cable quality matters as much as it does.
Counterfeiting in the FTDI ecosystem is documented, but forum claims in this area go further than the evidence, so separate what is documented from what is only claimed. The well-established case is the FT232 serial bridge. In October 2014, an FTDI driver shipped through Windows Update that detected non-genuine chips and reprogrammed their USB product ID to zero. That made the chips unrecognisable to Windows, Linux and macOS alike. The incident is real and was widely reported at the time, and it is why the phrase counterfeit FTDI carries weight in hardware circles.
Reports of relabelled, remarked or otherwise questionable FT601 parts on budget boards circulate widely in this niche. We have seen boards whose markings did not match their behaviour, but we have found no authoritative published analysis of FT601 counterfeiting to match the FT232 case. So we do not present it as established: treat it as community inference, not documented fact. The practical advice is the same either way. Buy from a seller who will tell you what is on the board, and photograph the markings when it arrives. Be suspicious of any board priced well below every other board that claims the same bridge.
Does one product name mean one board?
The most expensive mistake in this category is assuming that a product name identifies a board, because it often does not. Two cards listed as the same class with the same FPGA can still differ in PCB revision, USB bridge, flash chip, LED wiring and connector placement. Firmware is built against specific hardware, so a revision mismatch is not a minor incompatibility that mostly works. It is a card that does not start at all.
- 1Before you order firmware, photograph the silkscreen, including the board name, the revision number and the markings on the FPGA and the bridge chip.
- 2Ask the firmware seller to confirm compatibility against that photograph, not against the product name you bought the board under.
- 3Keep the photograph, because when something breaks in six months you will not remember which revision you have.
- 4If a seller cannot tell you which revision they are shipping, that tells you something important about the seller.
A short, known-good, data-rated USB 3 cable costs almost nothing, and it fixes more reported faults than any component swap. Buy two and keep one sealed as a reference. Then, when something goes wrong, you have a cable you know is good rather than four cables you suspect. Testing with a known-good reference part is the difference between fixing a problem in ten minutes and replacing a board that was working.
What do you give up with a cheap board?
Mostly, you do not give up reliability of the silicon, because the Artix-7 on a budget board is the same commodity part as on an expensive one. What you lose is everything around it: a cable that works, accurate notes on which revision you received, and a seller who can identify their own board. You also lose a thermal design with some margin, and someone to talk to when the card does not start. Those are the items that go missing first, and they are the ones that decide whether your build works.
Read next
Questions about this? Ask on Discord
All posts