Stock vs Medius
Stock MAKCU firmware V4.122 and Medius 3.5.0, tested the same way on two MAKCU boxes. The main results from each area are below, and each area's page has the rest.
DevicesAll results
| Test | Stock | Medius | Result |
|---|---|---|---|
| Native data reaching the PCKeyboard with five HID interfaces, an IN on each | An interface sends nothing | Not cloned | MediusStock defect |
| Interval the PC polls atFour mice with a 10 ms intervalIncluding a two-mouse device and two mice with two configurations | 1 ms | Native interval | MediusStock defect |
| HID driver on the PC768-2,047 bytes | Doesn't load | Loads | MediusStock defect |
| Requests given the native replySTM32's USB library | 924/948 | 948/948 | MediusStock defect |
PassthroughAll results
| Test | Stock | Medius | Result |
|---|---|---|---|
| Mouse reports reaching the PC1 kHz mouse | All, once, in order | All, once, in order | Equal |
| Report latency, p501 kHz mouse | 1.32-1.40 ms | 1.03-1.07 ms | Medius |
| Report latency, p50125 Hz mouseStock's clone declares 1 ms, the mouse 8 ms (see Poll interval) | 1.34-1.41 ms | 4.45-4.49 ms | StockStock defect |
| Reports read from a 1 kHz mouse in ~15.5 sInjecting 1,000 moves/s | 4,706-4,719 | 15,553-16,051 | MediusStock defect |
| PC traffic reaching the real deviceSwitch to configuration 2, then back | Clone reconnects instead | Passed on | MediusStock defect |
| Counter values skippedVendor-page HID interrupt IN | 80-140 | 0 | MediusStock defect |
| Tries with a request left unansweredSET_PROTOCOL, SET_IDLE and GET_PROTOCOL to 2 interfaces | 6/300 | 0/300 | MediusStock defect |
Injected motion All results
| Test | Stock | Medius | Result |
|---|---|---|---|
| Default move latency, p50Resting mouse, 1 count | 64.07-64.17 ms | 2.41-2.43 ms | Medius |
| Immediate move latency, p50Moving 125 Hz mouse, 1 count | 1.220-1.328 ms | 4.373-4.866 ms | StockStock defect |
| Default moves with their first count within 0.5 sMoving 125 Hz mouse, 1 count | 600/600 | 600/600 | Equal |
| Reports carrying only injected motion from 15,000 default movesMoving 1 kHz mouse | 4,619-4,622 | 0 | Medius |
| Empty reports in 2 s of one-count moves, mouse still30/s | 1,927-1,928 | 0 | Medius |
Control link All results
| Test | Stock | Medius | Result |
|---|---|---|---|
| Query round trip, p50Versionkm.version against QUERY(VERSION), whose reply also carries the box's MAC and name | 172-173 µs | 205-206 µs | Stock |
| Query round trip, p50Statuskm.moving against QUERY(HEALTH) | 172-173 µs | 144-145 µs | Medius |
| Valid move after the malformed inputHalf a move, the valid move 100 ms later | Lost | Arrives | Medius |
| Counts arriving from the malformed input5-count move with one bit flipped | 4 | 0 | Medius |
Buttons, keys and locks All results
| Test | Stock | Medius | Result |
|---|---|---|---|
| Native press after the injectionNative left held, box injects a left release | Still held | Still held | Equal |
| Native input reaching the PCX locked, box injecting Y at 1 kHz | None | None | Equal |
RecoveryAll results
| Test | Stock | Medius | Result |
|---|---|---|---|
| Left button 3 s after the program exitsPressed, then the port closed | Still held | Released | MediusStock defect |
| Replug times, p50Unplug to clone gone | 0.16-0.19 s | 0.08-0.15 s | Equal |
| Immediate move latency once the clone is back, p50Mouse reporting every 2 s | 3.17 ms | 2.06-2.37 ms | Medius |
| Held input 3 s after the program stops, 3 tries a runKey held, program silent | Still in effect | Released | MediusStock defect |
| Real device while the PC sleepsMouse | Suspended | Suspended | Equal |
FeaturesAll results
| Test | Stock | Medius | Result |
|---|---|---|---|
| Absolute move on a virtual screenMoves | km.moveto | No | Stock |
| Drive the pad state the PC getsPad commands | km.controller_state | No | Stock |
| Scale motion, 0-255%Motion scaling | No | LOCK | Medius |
| Control transfer to the real deviceDevice commands | No | TRANSFER | Medius |
SoftwareAll results
| Test | Stock | Medius | Result |
|---|---|---|---|
| KMBox Net programsProgram supportMove, wheel, three buttons and keys only | Toolkit Forwarding mode | No | Stock |
| Package downloadsPyPI, last 30 daysmakcu, pymakcu and makxd against medius | 3,854 | 691 | Stock |
| MembersDiscord | 28,612 | 767 | Stock |
| Previous build kept as a fallbackUpdates | No | Bootloader rollback | Medius |
Defects
| Box | Defect | Section |
|---|---|---|
| Stock | Report descriptors >512 bytes aren't served, so the PC's HID driver doesn't load for that interface. | Large report descriptors |
| Stock | While injecting, stock reads a 1 kHz mouse whose speed varies only ~300-550 times/s. At 10 moves/s it reads 2,962-3,035 of 5,500 reports in 5.5 s. | Reads while injecting |
| Stock | If a program closes the port or stops sending while a button or key is held or an input locked, it stays in effect until another program releases it or the box restarts (27/27 tries). | Program exit |
| Stock | When the PC switches the clone's configuration, stock reconnects the clone instead of relaying the switch to the real device (3/3 tries). The real device gets 2 bus resets switching to configuration 2 and back. | PC to real device |
| Stock | Moves arriving faster than stock takes them (~30 or more binary moves in one write, ≥10,000 binary moves/s, 8,000 text moves/s, 32 per write every 16 ms, or back to back) are lost. Stock refuses most of them at ≥10,000 binary moves/s, and none in the other cases. Its parser then stays one command behind until a power cycle (in 2/3 tries at 8,000 text moves/s). Typing 30 characters as key presses sent back to back loses characters. | Clumped and back-to-back writes |
| Stock | Neither an injected move nor a native remote wakeup wakes a sleeping PC. | PC sleep |
| Stock | Mouse clones declare a 1 ms poll interval even when the mouse declares 2-10 ms (a 125 Hz mouse's clone is polled at 1 kHz). Stock's lower latency below 1 kHz comes from this. | Poll interval |
| Stock | Requests to the clone sometimes go unanswered: SET_PROTOCOL in 6/300 tries. While the PC sets the clone up, report descriptor reads sometimes go unanswered too. The PC's HID driver waits 5 s before asking again, so the clone comes up ~5 s late. | Boot protocol requests |
| Stock | With the mouse moving, stock refuses moves past ~1,325/s, replying to each one it refuses (1,325/5,000 arrived at 5,000/s). | Paced moves, mouse moving |
| Stock | In a stream of RAW_MOVEs whose receipts go unread, stock refuses all but the first few (16/200 counts arrive), then every later RAW_MOVE until a power cycle. Ordinary moves are unaffected. | Unsmoothed moves |
| Stock | When the PC pauses reading an interrupt endpoint, stock drops reports: 80-140 counter values skipped in a run's 8,000 reports on each endpoint. Natively, a report the PC hasn't taken stays in the real device. | Reads after a pause |
| Stock | Where the real device replies past wLength (LUFA, STM32's USB library, V-USB), stock STALLs the request. | Standard requests across device stacks |
| Stock | Clones have no control endpoint above 0, so every request to one fails. On endpoint 0, stock STALLs reads whose reply is ≥1,024 bytes, a 1,024-byte write, and requests that natively end in an overflow or protocol error. If the real device STALLs a read partway, the PC gets the STALL without the data before it. | Relayed requests |
| Stock | While the PC reads nothing from an interrupt IN, stock still polls it at its declared rate. Natively, an endpoint the PC doesn't read isn't polled. | Polls the real device gets |
| Stock | Stock leaves interfaces it can't relay out of the clone's configuration descriptor (the vendor interface holding 5 of a mouse's 6 interrupt INs, an interface with a second control endpoint). On a keyboard with five HID interfaces, the fifth is declared but emits nothing. | Medius with imperfect clones on |
| Stock | Descriptors are rewritten to fit the box: IN endpoint 0x87 becomes 0x82, and 65- or 1,024-byte packets are declared as 64 bytes. | Medius with imperfect clones on |
| Stock | Back-to-back bulk is relayed at 28-39% of the native rate. | Medius with imperfect clones on |
| Medius | When the real device STALLs a write partway, the PC reports 0 bytes written (once 192), against 64 natively. The clone ACKs each OUT packet before the real device sees it. | Relayed requests |