The card is hardware; the firmware defines the behaviour
A blank FPGA board does nothing. Behaviour comes from firmware that implements a bus master, an address translation scheme and an out-of-band link. Two cards with identical silicon can behave completely differently depending on what firmware is loaded — which is why "which card" is a less useful question than "which firmware, maintained by whom".
01
What DMA actually means at the hardware level
Direct Memory Access is a mechanism that lets a device move data to and from system memory without routing every byte through the CPU. It exists because doing so through the processor would be ruinously slow for storage and networking. The PCIe specification defines how a device becomes a bus master and how it addresses memory.
An IOMMU — Intel VT-d or AMD-Vi — sits between the device and physical memory and enforces which addresses a given device is allowed to touch. When it is enabled and correctly configured, a bus master can only reach memory the operating system has explicitly mapped for it. That constraint is the entire reason an IOMMU matters to this discussion.
02
What is physically on the board
The board carries an FPGA, configuration flash, a PCIe edge connector and a link interface. The FPGA implements a PCIe endpoint: it enumerates with a vendor and device ID, negotiates link width, and exposes BARs. Configuration flash holds the bitstream that defines all of this.
The link interface is what makes the card useful externally: USB 3.x, gigabit Ethernet, or a fibre transceiver. The choice determines bandwidth and latency, which is why fibre-based designs cost more than USB ones and why "80K-class" transfer figures in vendor material refer to a bus-master emulation mode rather than raw link speed.
| Component | Function | Where cost comes from |
|---|---|---|
| FPGA | Implements the PCIe endpoint and transfer logic | Silicon size and speed grade |
| Config flash | Stores the bitstream | Small, but re-flashing risk |
| Link PHY | USB / Ethernet / fibre | Fibre transceivers are the expensive part |
| Firmware | Defines device identity and behaviour | Development and ongoing maintenance |
03
Why firmware is the real product
Firmware determines the vendor and device ID the card reports, the class code it advertises, the BAR layout and how it responds to configuration cycles. It also determines whether the card behaves like a plausible piece of consumer hardware or like an obvious lab instrument.
This is the part that requires continuous work rather than one-time design. Operating systems and anti-cheat vendors update their device catalogues; a firmware that looked generic last year may stand out this year. A card with no firmware maintenance path becomes a paperweight for anything beyond a lab bench, which is why support commitments matter more than the spec sheet.
- Vendor and device ID are firmware-defined, not hardware-defined.
- Class code and BAR layout affect how the OS classifies the device.
- Firmware maintenance is the difference between a tool and a brick.
04
What a DMA setup actually costs
The card is one line item. A working two-machine setup also needs a second computer, a capture or analysis path, storage, and — if the intent is anything beyond a lab bench — a subscription for whatever software layer the vendor provides. Hardware-only quotes routinely understate the total by a wide margin.
For legitimate research use, the honest comparison is against a bench logic analyser or a PCIe exerciser, both of which are single-purpose instruments costing similar amounts without the ongoing maintenance question. If your use case is protocol analysis, buy the instrument. If it is memory forensics, a software-based acquisition path is usually more defensible.
05
Where the legitimate use ends
Reading memory on hardware you own, for driver development, forensic acquisition or anti-cheat research, is legitimate engineering. Reading another application's memory to obtain information that application did not intend to expose, in a context governed by a Terms of Service, is where it becomes a rules question — and every major game operator writes that rule in behavioural terms, not hardware terms.
The practical test is not "is this device detectable" but "does the resulting behaviour violate a rule I agreed to". A device that is never detected still fails that test if the use is prohibited. This is why the "undetected" framing in vendor marketing is a distraction from the actual question.
06
Inspecting what is on your own bus
If your interest is defensive — verifying that a machine you bought does not contain unexpected hardware — the tools are built into Windows and require no downloads.
- Device Manager → View → Devices by connection, to see what physically sits on the PCIe tree.
- msinfo32 → check Kernel DMA Protection and Virtualization-based security status.
- Windows Security → Core isolation for memory integrity state.
- Any endpoint without a recognisable vendor name deserves an explanation.
FAQ
Frequently asked questions
01What does DMA stand for?
Direct Memory Access — a mechanism that lets a device transfer data to and from system memory without the CPU handling every byte. It is used by storage controllers, network cards and GPUs on every modern system.
02Is a DMA card illegal?
The hardware itself is a general-purpose PCIe device and is not illegal. Legality of use depends on jurisdiction and on what you do with it; Terms of Service violations are a separate matter from law.
03Why do DMA cards need firmware updates?
Because the device identity the card presents is defined by firmware. As operating systems and anti-cheat vendors update their device catalogues, firmware that looked generic can start standing out.
04How much does a DMA card cost?
Boards typically run from a few hundred to around a thousand US dollars depending on FPGA size and link type. A complete two-machine setup costs considerably more once the second PC and software are included.
05Can a DMA card read memory on a machine I do not own?
No. The card operates on the host it is physically installed in. Reading memory across a network requires a completely different class of access.
SOURCES
