Three independent observation layers
Device layer: what the PCIe endpoint reports. Behaviour layer: what the account does. Statistical layer: how the account compares to its bracket. A device can hide from the first and cannot hide from the other two.
01
What is genuinely observable at the device layer
A kernel-mode anti-cheat can walk the PCI bus and read configuration space. That gives vendor ID, device ID, class code, revision and BAR information for every endpoint present. It can also enumerate loaded drivers and their signatures.
What it generally cannot do from inside the protected OS is inspect the runtime behaviour of a device that presents valid metadata and loads no driver. That gap is real and is the technical premise of external hardware.
| Signal | Observable | Notes |
|---|---|---|
| PCI vendor / device ID | Yes | Straight from config space |
| Class code | Yes | Groups the device |
| Loaded driver + signature | Yes | Absence is itself notable |
| Firmware version string | Often | Depends on implementation |
| Runtime memory transactions | Rarely | Needs IOMMU-level telemetry |
| Behaviour in game | Always | Independent of hardware |
02
Why behaviour detection makes device invisibility irrelevant
If a player consistently acts on information the client did not display — pre-aiming at occluded targets, pathing around unseen threats, reacting inside human limits — the resulting pattern is measurable regardless of how the information arrived.
Modern anti-cheat systems increasingly weight this layer, because it is expensive to fake and cheap to compute. Server-side telemetry already exists for matchmaking and anti-griefing; reusing it for integrity analysis is a small marginal cost.
- Behavioural signals survive any hardware approach.
- Server telemetry already exists and is cheap to analyse.
- This is why enforcement waves include accounts the client driver never flagged.
03
Why ban rate percentages are meaningless
A ban rate is a ratio: banned accounts over total accounts using a given method. Nobody outside the operator has the denominator. Vendors have the numerator only for their own customers, and only for customers who report back — which is a self-selected sample.
Any percentage quoted is therefore either manufacturer marketing or forum anecdote. The honest statement is that risk exists, is non-zero, and cannot be quantified from outside.
04
How hardware enforcement changes the calculus
An account ban costs an account. A hardware ban costs the machine's identity, which is much harder to replace than a login. When operators apply hardware enforcement for severe cases, the expected cost of an enforcement event rises sharply even if the probability is unchanged.
This is the part vendors systematically understate: the question is not only "will I be caught" but "what does it cost if I am".
05
Inspecting your own machine for unexpected hardware
If your interest is defensive — verifying that a machine you bought does not contain hardware you did not install — the built-in tools are sufficient and require no downloads.
- Device Manager → View → Devices by connection: shows the physical PCIe tree.
- msinfo32: Kernel DMA Protection and Virtualization-based security status.
- Windows Security → Core isolation: memory integrity state.
- An endpoint with an unfamiliar vendor name or no driver deserves an explanation.
06
Why enforcement arrives in waves, not per detection
Operators batch enforcement deliberately. Banning on first detection teaches the other side which signal fired; delaying and banning in volume removes that feedback loop and makes attribution harder. It also produces a period where a technique appears to work, followed by a large ban wave — which is exactly the pattern forum reports describe.
For anyone reading vendor claims, this is the practical consequence: a technique that has been clean for three months is not evidence of safety. It may be evidence that the next wave is being assembled.
07
The false-positive constraint
Every enforcement system trades detection against false positives, and hardware-level signals are particularly prone to them. A shared machine, a refurbished board, a cloned OEM image — all can present identifiers that were previously flagged for reasons unrelated to the current user.
That constraint is why operators weight behavioural evidence heavily and apply hardware bans mainly for severe or repeat cases. Getting it wrong locks out a paying customer, and the support cost of that mistake is immediate.
- Hardware signals produce false positives through shared and second-hand hardware.
- Behavioural evidence is more specific and carries lower false-positive risk.
- This is why hardware bans concentrate on severe and repeat cases.
FAQ
Frequently asked questions
01Can anti-cheat detect a DMA card?
It can read PCIe configuration space, device IDs, class codes and loaded drivers. Runtime behaviour of a device that presents valid metadata and loads no driver is much harder to observe from inside the protected OS.
02What is the ban rate for DMA hardware?
There is no verifiable figure. Nobody outside the game operator has the denominator, and any percentage quoted is marketing or anecdote rather than measurement.
03Does a hardware ban cost more than an account ban?
Yes. An account ban costs one account; a hardware ban associates enforcement with machine identifiers, so a replacement account on the same machine is also affected.
04Is behavioural detection independent of hardware?
Yes. It analyses match outcomes and timing, which are produced by the account regardless of what hardware generated the input.
05Why do bans arrive in waves?
Batch enforcement removes the feedback loop that would tell the other side which detection signal fired, and it makes attribution harder. It also creates the illusion that a technique works until the wave lands.
SOURCES
