AMD Ryzen AI Embedded P100 COM Express Type 6 (FP8) - IC Power Dissipation Heatmap

Power density - mW/mm² (log scale)
0.001
0.01
0.1
1
10
100
Range: 15 - 54 (configurable TDP; default TDP 28). Step: 1.
Assembly options fitted:
Estimated dissipation from datasheet data at the selected power limit, on the real placement and stackup; an attention map, not a thermal measurement.

TOP side

W at W CPU limit (incl. LPCAMM2 and SSD) · viewed from above
U1M.2 2280 SSD card zoneJ1 - LPCAMM2 module zoneU81U9U10084U138U134U135U136U137U10078L39U20L38U139U2001L35L36L37L40R9438R9439R9440R9441R9443U10U10085U10080L18R9595U10130

BOTTOM side

W at W CPU limit · X-ray view from above (not mirrored: same orientation as the TOP side)
U2032U2030L46L41Q11U2040U2033U2031U10146R9442U133U10060U2034U10063U22U49U10086U2035U2037U30L45L49L52L42U29U36L51L47U10064U37U10075U10076U10065U10066U10077U3004U3007U10145U10082U10083U3003U10009U10087U10088U3009U2036
Heat spread through the board
Default view excludes the APU, the LPCAMM2 module and the SSD - they have their own heat paths (heatsink or spreader, module, card) and would otherwise hide the board's own hot zones.
0relative attention (1 = maximum of this view)1
o = the largest dissipators in this view   triangle = heat-sensitive part (hover for details)   all views from above
Heat-sensitive parts
RefSidePartWhy it mattersAttention, all sourcesAttention, excl. APU, LPCAMM2, SSDNearest part dissipating >= 0.3 WDistance (mm)
Data table - all modelled parts, with datasheet thermal resistances
RefSideFunctionPartFootprint (mm)W at WmW/mm²θJA (K/W)θJC (K/W)θJB (K/W)Basis and thermal source
Method and limits

What this page is. A placement-times-estimated-dissipation attention map: where the dissipating parts sit, where they cluster, and which heat-sensitive parts sit near them. It is not a thermal simulation: there is no airflow, no ambient, no heatsink or heat-spreader model and no measured current. The spread view is relative (normalised to its own maximum) and must never be read as a temperature.

Placement is the real board: every part's position, rotation, side and body from the layout database, validated against the copper (100 % of pins on their pads, 100 % of pin nets equal to the schematic netlist) and cross-checked against two further placement sources (1044 of 1044 parts agree within 0.01 mm). The SSD card zone is the M.2 2280 outline between the connector and its standoff.

Dissipation model: a power tree, not a list. Each regulator's load is everything its output feeds on this board (parts, modules and the input current of the regulators below it), followed from the schematic netlist; the tree is evaluated from the leaves up to the 12 V input, so the 12 V input power always equals the sum of every part's dissipation (shown above the boards). Inputs: the CPU power limit P, the LPCAMM2 module power, the SSD power, peripheral activity (datasheet maximum or typical) and the fitted assembly options.

  • APU: W = P, the sustained power limit, which covers all APU supply rails. Short excursions above it (package power tracking, seconds) are faster than the board heats up and are not part of this steady-state view.
  • APU rail split (assumption; AMD's actual per-rail currents depend on the workload): CPU core and graphics 70 % (1.0 V, four phases), SoC 20.5 % (1.0 V), PLL and display controller 3.5 % (1.0 V), I/O-memory-misc 6 %, the last group weighted by each rail's AMD current rating. Every rail current stays within AMD's specified limit for that rail at every power setting.
  • Power stages: loss = a + b I + c I2 per stage through two estimated points (vendor loss curves are not public); a = 0.05 W is the driver supply drawn from 3.3 V. Stage inductors: DCR x I2 + 0.10 W core loss (assumed).
  • Buck regulators: total loss read from the vendor's power-loss or efficiency curve at the computed load current; the curve's own test inductor is removed and this board's inductor DCR x I2 is booked on the inductor. For the 5 V-input bucks the curve (1.2 V out) is corrected for the conduction share at the actual output voltage using the published switch resistances.
  • Linear regulators: (Vin - Vout) x I + Vin x Iq. Load switches and the 5 V path MOSFET: R x I2.
  • Peripherals: the datasheet maximum (or typical) supply current x the nominal rail voltage, per supply rail; each part's basis is in its tooltip and in the data table.

Thermal resistances in the data table are printed only where the vendor's public datasheet publishes them, with the document named. theta-JC is junction to case top unless marked; theta-JB is junction to board; Psi values are characterisation parameters, not resistances, and are given in the notes. An empty cell means the vendor does not publish the value.

Conduction. The in-plane sheet conductance of the real 8-layer stackup, kt = sum over layers of kCu x ti x coveragei + kFR4 x tdielectric, with copper weights from the layout database and the copper coverage of each layer from a 0.05 mm union raster of every copper feature. The source map is spread with the Green's function of a thin plate losing heat from both faces, K0(r / lambda), lambda = sqrt(kt / h); h is a normalisation constant, not a cooling model. Outer layers are credited with 25 um of plating (assumption).

Layerozt (um)CoveragekCu x t x coverage (W/K)
signal 10.33036.50.33970.00483
plane 20.33011.50.91280.00409
signal 30.50017.40.15370.00104
signal 40.50017.40.53640.00364
plane 50.50017.40.83650.00568
signal 60.50017.40.14850.00101
plane 70.33011.50.90620.00406
signal 80.33036.50.41970.00597
FR4, all dielectrics-1742.4-0.00052

kt = 0.03034 W/K (copper) + 0.00052 W/K (FR4: 0.30 W/mK x 1742.4 um) = 0.03086 W/K; h = 20 W/m2K; lambda = 39.3 mm.

Spread view in this page. The spread is linear in the part powers, so each dissipating part's response to 1 W is precomputed on a 0.25 mm grid, averaged to 1 mm cells, and summed with the powers of the selected state. Parts below 5 mW in every state are listed but not spread. Attention = the spread value at a part's centre divided by the maximum of the same view.

Limits.

  • No measured per-device or per-rail current exists for this board; apart from the APU envelope and the peripheral datasheet figures, every number is an estimate at an assumed load.
  • Power-stage, memory-PHY regulator and SATA bridge losses rest on assumptions (no public vendor data); the SATA bridge (1.3 W) is the least certain peripheral figure and should be measured.
  • The DDR PHY core rail is switched by the APU between several voltages; its default (0.78 V) is used.
  • The LPCAMM2 and SSD powers are selector inputs, not module data; their heat is carried by the module or card and drawn over its outline.
  • The APU power is spread uniformly over its package; the die outline is not modelled.
  • Thermal resistance of the APU: AMD specifies the cooling requirement as die-to-ambient thermal resistance per power setting (FP8 platform documentation, under NDA), not as a junction-to-case value.
  • The board is treated as an infinite plate: edges and holes are not boundaries, so parts near an edge (the power-stage rows, the Ethernet and SATA bridges) are under-weighted relative to parts near the centre.
  • The heat-sensitive list is a judgement (crystals, oscillator, polymer bulk capacitors, flash, EEPROM, TPM, temperature sensor).
  • No heatsink, heat spreader, enclosure, airflow or ambient. A heat spreader changes absolute temperatures completely; it changes the ranking mainly near the APU and under the SSD card.
  • Configurable TDP 15-54 W applies to the commercial and industrial P100 parts; the automotive P122a (15-30 W) and P132a (25-45 W) have narrower ranges.
AMD's FP8 IRM doesn't define θJC. It specifies the whole cooling path as die-to-ambient θda: 0.85 °C/W at 54 W and 1.65 °C/W at 28 W, at 25 °C ambient.