How DMA Firmware Works: Cards and Fusers

You seated a card in a spare PCIe slot and Windows now shows an unknown device with a yellow exclamation mark. Or you are weighing a two computer setup against software that installs on the machine you already own, and trying to work out what the extra hardware actually buys you. Either way, the piece almost nobody explains properly is DMA firmware.

This is the plain-English version: what a DMA setup is, why the second PC exists, what a fuser and a KMBox do, what the 35T, 75T and 100T numbers mean, and what the firmware step is really for.

It is written for someone deciding whether to buy in, and for someone who already owns a card and needs the firmware step to make sense.

What DMA is and why it needs a second PC

DMA stands for direct memory access. PCIe devices are allowed to read and write system memory directly instead of routing every transfer through the CPU, which is a normal capability that network cards and storage controllers use constantly.

A DMA card puts that in your hands. It is an FPGA board seated in a PCIe slot on your gaming PC, reading the game's memory across the bus. It does no analysis of its own; it pushes what it reads down a USB 3 cable to a second computer, which parses the structures, locates players and draws the overlays.

The reason people accept that complexity is what is not on the gaming machine: no process to enumerate, no driver to load, no module injected into the game, no overlay hooked into the renderer. Kernel anti-cheats are very good at inspecting the machine they run on, and far weaker when the software they hunt sits on a computer they have no presence on.

Set against an internal setup, that is the whole trade. Internal software runs on the gaming PC as a process the anti-cheat shares a machine with, which is why it is simple to start and why the pressure on it never lets up. DMA moves the software off that machine entirely, and pays for it with a second computer, a video path, an input path and a rig you have to build before anything works at all. Neither approach touches how a match looks from the server, which is the part covered further down.

That leaves two problems, and every other piece of hardware exists to solve one of them. The second computer has to see your game, and it has to move your mouse. The machine itself does not need to be powerful, because it renders nothing. An old office desktop or a laptop with a USB 3 port and a capture input covers it.

Fusers, KMBox devices and the rest of the chain

The video path, and what a fuser is for

Your second PC has no idea what your game looks like. It has memory addresses, not pixels, so something has to carry the video across.

An HDMI fuser is a splitter built for that job. It takes the gaming PC's output, passes the full signal to your monitor, and sends a copy to the capture input on the second machine.

The obvious question is why a cheap splitter will not do, and the answer is EDID. Every display tells the source which modes it can accept, and a splitter has to answer for everything attached to it at once. Most of them do that by negotiating a single shared mode, so hanging a 1080p60 capture input off a cheap unit drags your high refresh monitor down to match. The ones that instead copy the first port's EDID wholesale fail the other way and hand the capture side a mode it cannot take at all. A proper fuser handles EDID per port, so the gaming PC keeps its native resolution and refresh rate while the second machine gets a copy it can digest.

That is what separates a 1K fuser from a 4K one. At 1080p a standard unit is enough. Above that, 4K capability stops the capture path compromising the picture you play on, and it matters because you spend the whole session looking at the second machine's screen rather than your own.

KMBox class input devices

Here is the fork that decides what your money buys. A card on its own is read-only. It hands you information: overlays, radar, positions, loot. It cannot move your mouse, and no firmware changes that.

A KMBox is a small board on the USB path between your mouse and the gaming PC. Your real mouse plugs into it, and it plugs into the gaming PC. The second computer sends it instructions over its own link and the board emits USB HID reports, so movement arrives as real USB input from a device the gaming PC enumerates like any other peripheral, rather than as something synthesised by local software that anti-cheat can watch for.

Boards in that family differ in how faithfully they pass your own hand through while injecting movement, which is the detail worth asking about before you commit to one. Without an input device, plan for a visuals-only setup and decide up front that you are content with it, because adding aim assistance later means buying more hardware and rewiring the desk.

Cables

The card talks to the second PC over USB 3, so use the cable supplied with the kit. A marginal cable is the most common cause of a board that enumerates fine and then reads slowly or drops mid-session, and it is the last thing anyone suspects.

Card tiers: what 35T, 75T and 100T actually mean

The numbers are not tiers invented by sellers. They name the Xilinx Artix-7 FPGA on the board: XC7A35T, XC7A75T, XC7A100T. The number is roughly how many thousand logic cells the chip carries, so a 100T has substantially more programmable fabric than a 35T.

More logic cells buys firmware room to do more on the card itself, plus buffering headroom that shows up as higher sustained read throughput. Throughput is the part you feel: everything the second PC knows about the world is refreshed by reading memory across the bus, so faster reads mean overlays that track rather than trail. The USB 3 link imposes a ceiling of its own, so a bigger chip is not a free multiplier. It is more room for the firmware to keep that link saturated instead of stalling on it.

In practice the 35T is the entry point and it works, with the least headroom for elaborate firmware. The 75T is the mainstream pick and the usual middle ground. The 100T has the most headroom and the longest useful life, which counts for something because the card outlives everything else in the chain. ZDMA boards are a separate family, so check compatibility for one of those rather than assuming it.

The tier does not change which features you get, only how comfortably the setup keeps up. If you have to economise somewhere, dropping a card tier hurts less than skimping on the fuser, because a compromised video path is visible in every second of play while a slightly slower card is not.

What DMA firmware actually does and why it matters

An FPGA is a blank chip until you load a bitstream into it. That bitstream is the firmware, and it decides what the device looks like to the machine it is plugged into.

Every PCIe device answers a standard set of questions when the system enumerates the bus: vendor ID, device ID, subsystem IDs, class code, capability list, memory window layout. That block is the configuration space, and anything privileged enough to walk the bus can read it, kernel-level anti-cheat included.

Stock, publicly available firmware answers with documented, widely known values, so checking a machine against a published list is not difficult work. Custom firmware changes the answers so the device presents as something else. Emulated firmware goes further: it reproduces the configuration space of a specific real commercial card, field for field, and pairs it with that device's driver.

Field for field is the demanding part, and it is not just the vendor and device IDs. The memory windows have to be the sizes the real card declares, the capability list has to match, and the subsystem IDs have to belong to a board that was genuinely sold. The result also has to be a device that plausibly exists in a machine like yours: enterprise networking hardware sitting in a gaming desktop is an odd answer even when every individual field is correct.

The driver half is what buyers miss, and it is why the yellow exclamation mark matters. A PCIe device with no driver appears in Device Manager as a problem device, a visible anomaly on an otherwise ordinary machine. A driver-backed build loads a driver and shows as normal hardware, which is the difference between a system that reads as clean and one that invites a closer look.

The last piece is the DNA lock. Every Xilinx FPGA carries a unique factory identifier called Device DNA, and a locked build is compiled against your board's so it will not run on anyone else's card. That is why firmware is not an instant download: it is built per order once you supply your DNA ID.

No firmware makes a card invisible, and anyone telling you otherwise is selling a story. What good firmware does is remove the obvious tells, and because detection approaches change, it is something you keep current rather than something you buy once and forget about.

Choosing a firmware build

Firmware builds are separated by which anti-cheats they are made to sit alongside, and that ladder is worth understanding before you spend anything. BattlEye sits at the accessible end. Easy Anti-Cheat and Anti-Cheat Expert are a step up. Vanguard is the strictest of the common ones, because it loads at boot and holds the machine from the moment it starts rather than from the moment the game does. These are coverage grades rather than quality grades, so take the one that covers the games you actually play. Paying up the ladder for an anti-cheat you never meet buys you nothing.

Firmware is bought for a card you already own, not the other way round, and it is compiled against that board's DNA ID rather than pulled off a shelf. Confirm your board is one the build supports before you buy, and expect to open a ticket and hand over the DNA ID afterwards.

The store carries the hardware side and the firmware for it, and it also carries software that runs on the gaming PC without any of this equipment. If you are new to this, that internal route is the sensible place to start, because it tells you whether you like the result before you commit to a second computer and a rebuilt desk. When you do move to hardware, take the shortest commitment available on the software side before signing up for a long one, and check the status page before you buy anything, since a build that is fine today can be pulled without warning. What exists right now, what it covers and what it costs all live on the DMA category page, which is the only place worth trusting for that. A guide cannot keep up with a catalogue and should not pretend to.

Budgeting for a full setup

Firmware is one line on a longer bill, and the usual mistake is pricing the card alone and meeting the rest afterwards. Work through the whole chain before you buy the first piece of it.

The card is the obvious part. The video path needs a fuser rather than a plain splitter, and if you play above 1080p that means a 4K unit rather than a standard one. Aim assistance needs a KMBox class input device, and skipping it leaves you a visuals-only rig, which is a legitimate choice as long as it is a deliberate one. The USB 3 cable is the piece everyone treats as an afterthought and the piece that most often causes trouble. Buying the parts together is usually simpler than sourcing them one at a time, because they arrive matched and you are not left hunting for a fuser that behaves.

Two costs sit outside the hardware entirely. The second PC is yours to supply, no extra outlay if you have an old machine in a cupboard and a real expense if not. It renders nothing, so it does not need to be fast, but it does need USB 3 and a capture input, and a laptop lacking either is not the saving it looks like. The other is the software licence for whichever game you are playing, which is a separate purchase from the hardware and set by the listing you choose rather than by the card. That is the line people forget when they cost a rig: the hardware is bought once and then keeps working across every game you point it at, while the software side is a decision you make again each time you change what you play.

Getting set up

Order matters here more than on anything else the store sells, because firmware is built for a card you have to own first.

1. Get the hardware working: card seated in the gaming PC, second machine on, video path built, input device wired between your mouse and the gaming PC.
2. Read your card's Device DNA ID. The setup pages at guides.camunlocks.com cover how.
3. Choose the build matching the anti-cheats you play against, and confirm you own a card it supports before buying.
4. Open a ticket in the Discord with your DNA ID. Firmware is compiled per order rather than pulled from a shelf, so it is not an instant download.
5. Flash the build, then confirm in Device Manager that the device shows as normal hardware rather than a problem device.
6. Buy your game software last, once the rig is proven.

Support runs through the same Discord, and setup questions are the most common thing it handles.

Staying safe with a DMA setup

A DMA rig changes what runs on your gaming machine. It does not change how you behave in a match, and account risk is real regardless of how clean the hardware looks.

Server-side analysis has no interest in where your software runs. It looks at hit rates, reaction timings, movement and engagement patterns, and those numbers come from you rather than from your card. A rig that is invisible to the machine it sits in is still perfectly visible in a report describing a crosshair that snapped through a wall. Player reports and admin spectating are equally indifferent to your hardware.

What reduces risk is unglamorous. Keep firmware current rather than running whatever you flashed when the card first arrived. Check the public status page before every session, because something fine on Tuesday can be pulled on Thursday. Keep recordings and streams off, including your friends' clips, since a clip you never posted yourself can still end up in front of the wrong person. Stay inside what a strong player could plausibly do, and use an account you would not mind losing, because the card survives a ban and the account does not.

Frequently asked questions

What is DMA firmware and why do I need it?

A DMA card is an FPGA, a blank programmable chip until you load a bitstream onto it. That bitstream is the firmware, and it determines what the device reports about itself to the machine it is plugged into. Stock firmware answers with publicly documented values that are trivial to look up. Custom and emulated builds change those answers so the card presents as a different device, field for field, and a driver-backed build also loads a driver so the card shows in Device Manager as ordinary hardware rather than as an unknown device carrying a yellow exclamation mark.

Do I need a KMBox to use a DMA card?

Only if you want aim assistance. A card by itself is read-only, so it gives you overlays, radar and positional information but cannot move your mouse. A KMBox class board sits between your mouse and the gaming PC and delivers movement as real USB input. Without one you have a visuals-only setup.

What is the difference between a 35T, 75T and 100T card?

Those names come from the Xilinx Artix-7 FPGA on the board, and the number is roughly its logic cell count in thousands. More fabric means more firmware headroom and higher sustained read throughput, which shows up as overlays that track closely rather than trailing. The tier does not add features.

Why is DMA firmware not delivered instantly?

Because it is built for your specific board. Every Xilinx FPGA carries a unique factory identifier called Device DNA, and a locked build is compiled against yours so it will not run on any other card. You supply the DNA ID by opening a Discord ticket, and the build is compiled after that, so it is never an instant download.

What does a complete DMA setup involve?

It depends on how much of the chain you already have. A working rig needs the card, a second computer to run the software, an HDMI fuser for the video path, a KMBox class input device if you want aim assistance, and the USB 3 cable between the card and the second PC. The second PC is yours to supply, and the software licence for each game you play is a separate purchase from the hardware.

Can I use a cheap HDMI splitter instead of a fuser?

You can try, and most people regret it. A basic splitter negotiates one shared display mode across every attached output, so a 1080p60 capture input drags your high refresh monitor down to match. A fuser manages that per port, letting the gaming PC keep its native resolution and refresh.

Will a DMA setup stop me getting banned?

No, and nobody can honestly tell you otherwise. DMA removes software from the gaming machine, which addresses one category of detection. It does nothing about server-side behavioural analysis, player reports or admin spectating, all of which respond to how you play. Keep firmware current and check the status page.

If you already own a card, the DMA category is where the current builds and their anti-cheat coverage are listed. If you are still deciding, cost the whole chain before you buy the card, because the card is the part people reach for first and the rest of the chain is what catches them out. The Discord is open for a second opinion either way.