Thermal Characterization of Cargt SOMs

Will your product need a heatsink? This page answers that for Cargt SOMs. We ran each SOM under a range of loads, CPU frequency caps and cooling options, in open air and in a sealed plastic enclosure, until its temperature stopped changing. From each test we report the warmest air the setup can run in without the processor reaching its design limit.

SOMs covered

SOM	Processor	Status
00414	i.MX 95	Tested: open air and enclosure, no heatsink and passive heatsink, stress, application and NPU loads
00377	i.MX 8M Plus	Tested: open air and enclosure, no heatsink; open air, passive heatsink; stress, application and NPU loads. Passive heatsink in the enclosure planned
00363-93	i.MX 93	Tested: open air and enclosure, no heatsink, stress and application loads. NPU testing planned
00395	STM32MP257F	Tested: open air, no heatsink, CPU, memory, GPU and video loads. Enclosure and NPU testing planned

Results for each SOM are added here as testing completes. Every test, with its full measurements, is listed in Thermal Test Data.

How to read the results

Each test runs one load until the die temperature is steady, which takes 20 minutes to 4 hours. Two numbers come out of it:

• ΔT is the steady die temperature minus the surrounding air temperature. It depends on the load, the CPU frequency cap and the cooling, but hardly on the room temperature.

• Maximum ambient is the design limit minus ΔT: the warmest surrounding air the setup can run in without the die passing the design limit.

The design limit is 10 °C below the temperature at which the processor starts to throttle (its passive trip point). Below it, the SOM runs at full speed with margin to spare. For the i.MX 95 the passive trip is 105 °C, so the design limit is 95 °C. For the i.MX 8M Plus and i.MX 93 the passive trip depends on the processor’s temperature grade: U-Boot sets it 10 °C below the grade’s maximum junction temperature. The SOMs tested here have 105 °C parts, so the passive trip is 95 °C and the design limit is 85 °C. A part with a lower maximum has a lower trip point and design limit. For the STM32MP257F the passive trip is 110 °C, so the design limit is 100 °C. grep . /sys/class/thermal/thermal_zone*/trip_point_*_t* shows a running board’s trip points and their types, in thousandths of a degree.

Note

The maximum ambient figures come from tests in a 19–24 °C room. Leakage current rises with temperature, so ΔT grows somewhat at high ambient, and figures above about 50 °C are extrapolations. Temperature chamber tests to confirm them are planned.

Summary

For the i.MX 95 (00414):

• With the passive heatsink, the SOM suits indoor products in a sealed plastic enclosure, even at full load. The worst case we tested, CPU, GPU, memory and NPU all fully loaded at 1.8 GHz, still allows 38 °C room air (44 °C with the CPU capped at 1.4 GHz). A touchscreen UI with video playback allows about 64 °C.

• Without a heatsink, the worst case allows only 37 °C air in open air at 1.8 GHz (45 °C at 1.4 GHz). That is enough for light loads but not for sustained full load in an enclosure.

• Capping the CPU frequency is the most effective setting. Each step between 900, 1404 and 1800 MHz changes ΔT by 4–8 °C under load.

• Real application loads run much cooler than stress tests. A touchscreen UI with 1080p30 video runs about 17 °C cooler than CPU, GPU and memory fully loaded.

• The NPU is a significant heat source. At 50–60 % busy it adds about 9–10 °C of ΔT.

• When the SOM does throttle, it degrades gracefully. The kernel holds the die 1–3 °C under the throttle point by lowering the CPU and then the GPU frequency; video playback keeps its frame rate.

For the i.MX 8M Plus (00377):

• In open air, the SOM needs no heatsink. The worst case we tested, CPU, GPU, memory and NPU all fully loaded at 1.6 GHz, allows 45 °C room air; without the NPU, 52 °C. A touchscreen UI with video playback allows about 60 °C.

• A passive heatsink adds 5–12 °C of headroom. It lowers ΔT by about 30 % and raises the open-air worst case from 45 to 57 °C, with no change in performance.

• It runs much cooler than the i.MX 95. Under the same full CPU, GPU and memory load it rises 33 °C above the room, against 58 °C for the i.MX 95 without a heatsink, so it allows 7–15 °C warmer air despite its lower design limit.

• In a sealed plastic enclosure, it needs no heatsink unless the NPU runs flat out with everything else. Without the NPU, every load allows at least 42 °C room air; a touchscreen UI with video allows 51 °C. With CPU, GPU, memory and NPU all fully loaded it allows 35 °C at 1.6 GHz and 39 °C at 1.2 GHz.

• The NPU is again the largest single heat source, adding 7–10 °C of ΔT.

For the i.MX 93 (00363):

• It needs no heatsink, in open air or in a sealed plastic enclosure. Under full CPU and memory load it rises only 20 °C above the room in open air and 24 °C in the enclosure, so it allows 64 °C and 61 °C room air.

• It runs much cooler than the i.MX 8M Plus: with all CPU cores loaded it rises 19 °C above the room, against 31 °C for the i.MX 8M Plus at 1.6 GHz.

• The enclosure costs only about 3 °C of ambient. The air inside it ran 2–3 °C above the room.

For the STM32MP257F (00395):

• In open air, the SOM needs no heatsink. The worst case we tested, CPU, GPU and memory all fully loaded at 1.5 GHz, rises 26 °C above the room. With its 100 °C design limit that allows 74 °C room air, although figures that high are extrapolations.

• It runs cooler than the i.MX 8M Plus and i.MX 95. Under full CPU, GPU and memory load it rises 26 °C above the room, against 33 °C for the i.MX 8M Plus and 58 °C for the i.MX 95 without heatsinks.

• The GPU is its largest heat source. Capping the CPU at 1.2 GHz saves only about 2 °C.

i.MX 95 (00414) results

Test setup

Hardware	00414 i.MX 95 OSM-L SOM on the 00365 carrier board, no display connected
Software	Cargt Linux BSP (NXP lf-6.18.20). CPU governor: performance for the full-load tests, ondemand for the others.
CPU frequency caps	900, 1404 and 1800 MHz. The i.MX 95 Cortex-A55 cores have no 1.2 GHz operating point; 1404 MHz is the nearest.
Die temperature	The higher of the i.MX 95 ANA and A55 thermal sensors
Ambient	Dracal USB-PTH450 sensors: one in the room and, for enclosure tests, one in the enclosure air
Passive heatsink	Advanced Thermal Solutions ATS-CPX045045025-162-C2-R0 (45 × 45 × 25 mm) with its pre-applied phase-change interface material, fastened to the 00365 carrier with ATS-HK91-R0 hardware
Enclosure	Polycase WA Series WA-42*16 (12 × 9 × 3 in), lid sealed, no vents

Loads:

• Stress loads: stress-ng on the CPU (all cores, half load, or a single thread) and memory; glmark2 on the GPU; and all of them at once (“CPU + GPU + memory”).

• Application loads:

  • a Chromium kiosk touchscreen UI, with three scrolling charts at the display rate, live value tiles and an event list;

  • 1080p30 H.264 video playback on the VPU.

• NPU loads: TensorFlow Lite inference on the eIQ Neutron NPU:

  • MobileNet v1 image classification;

  • SSD MobileNet object detection.

Open air: no heatsink and passive heatsink

Load	CPU cap (MHz)	ΔT, no heatsink (°C)	ΔT, passive (°C)	Maximum ambient, no heatsink → passive
CPU + GPU + memory	1800	58.4	37.5	37 → 58 °C
CPU + GPU + memory	1404	50.2	32.7	45 → 62 °C
CPU + GPU + memory	900	42.8	28.4	52 → 67 °C
CPU, all cores	1800	46.6	30.9	48 → 64 °C
GPU only	1800	35.9	24.3	59 → 71 °C
CPU, all cores at 50 %	1800	37.1	23.8	58 → 71 °C
Idle	1800	24.1	16.9	71 → 78 °C

• The passive heatsink lowers ΔT by 30–36 %, which allows 7–21 °C warmer air. The gain is largest under the heaviest loads.

• No test throttled. The hottest die reading was 82.6 °C, at full load with no heatsink.

• The CPU governor (ondemand, performance or schedutil) changed temperatures by less than 1 °C at the same cap and load. A governor does not replace a frequency cap.

• Reducing the GPU frequency cap from 1000 to 500 MHz lowers ΔT by only 2–4 °C.

Sealed enclosure, passive heatsink

Load	CPU cap (MHz)	ΔT, open air → enclosure (°C)	Enclosure air above room (°C)	Maximum ambient, open air → enclosure
CPU + GPU + memory	1800	37.5 → 48.3	5.2	58 → 47 °C
CPU + GPU + memory	1404	32.7 → 42.7	4.7	62 → 52 °C
CPU + GPU + memory	900	28.4 → 36.2	3.7	67 → 59 °C
CPU, all cores	1404	26.5 → 33.5	3.4	69 → 62 °C
GPU only	1800	24.3 → 32.8	3.7	71 → 62 °C
Idle	1800	16.9 → 23.4	3.2	78 → 72 °C

• The enclosure costs 6.5–11 °C of maximum ambient. Only 3–5 °C of that is the enclosure air warming up. The rest is the heatsink working less well in still air.

• An enclosure takes several hours to settle. The 4-hour tests ran about 1.3 °C warmer than 40-minute tests at the same settings. The CPU + GPU + memory rows above are from 4-hour tests.

• No test throttled. The hottest die reading was 73.2 °C.

Application and NPU loads in the enclosure

All with the passive heatsink, CPU capped at 1800 MHz unless stated.

Load	ΔT (°C)	Maximum ambient	Performance
Idle	23.4	72 °C	–
1080p30 video playback	24.5	71 °C	30 fps, no dropped frames
Touchscreen UI	30.6	64 °C	49 fps
Touchscreen UI + 1080p30 video	31.2	64 °C	UI 50 fps, video 30 fps
Touchscreen UI + video, CPU at 1404 MHz	30.0	65 °C	UI 49 fps, video 30 fps
Touchscreen UI + video, CPU at 900 MHz	28.4	67 °C	UI 39 fps, video 30 fps
Object detection on the NPU	30.2	65 °C	194 inferences/s
Image classification on the NPU	33.2	62 °C	821 inferences/s (1.2 ms each)
Touchscreen UI + video + object detection	40.2	55 °C	UI 48 fps, video 30 fps, 216 inferences/s
CPU + GPU + memory + NPU (4 hours)	56.7	38 °C	all at full load
CPU + GPU + memory + NPU, CPU at 1404 MHz (4 hours)	50.6	44 °C	CPU −16 %, memory −24 %, NPU −4 %, GPU unchanged

• Video playback costs almost nothing: the VPU adds about 1 °C to ΔT.

• For a UI, 1404 MHz costs nothing: the UI keeps its frame rate and the chip runs 1.2 °C cooler than at 1800 MHz. At 900 MHz the UI frame rate drops 22 % for only 1.6 °C more.

• The NPU adds 9–10 °C of ΔT at 50–60 % busy, whatever else is running. MobileNet runs in 1.2 ms per inference on the NPU, against 13 ms on the CPU.

• CPU + GPU + memory + NPU is the worst case tested. It allows 38 °C room air in the enclosure, against about 47 °C for the same load without the NPU. Over 4 hours it settled under 1 °C hotter than in a 1-hour run.

• A 1404 MHz CPU cap recovers 6 °C of that worst case, to 44 °C. It costs 16 % of CPU and 24 % of memory throughput; the NPU loses only 4 % and the GPU nothing.

Throttling behavior

No real test throttled, so we lowered the throttle point (passive trip) to make the SOM throttle at room temperature and see how it behaves. All runs were in the enclosure with the passive heatsink, capped at 1800 MHz.

Load	Throttle point	CPU MHz while throttled	Performance while throttled
CPU + GPU + memory	65 °C	1169	CPU −18 %, memory −27 %, GPU unchanged
CPU + GPU + memory	60 °C	980	CPU −35 %, memory −41 %, GPU unchanged
CPU + GPU + memory	55 °C	707	CPU −57 %, memory −61 %, GPU −35 %
Touchscreen UI + video	50 °C	631	UI 50 → 25 fps; video still 30 fps

• Throttling is orderly. The kernel holds the die 1–3 °C under the throttle point by stepping the CPU down first and the GPU only when that is not enough.

• The SOM ran steadily for over 40 minutes while throttled, with no oscillation, errors or crashes.

• The cost grows with how far the load overshoots: 5 °C over costs 20–35 % of CPU performance, 15 °C or more costs over half.

Long-term stability

• 36 hours at full load: CPU + GPU + memory at 1800 MHz with the passive heatsink in open air, using stress-ng --verify. There were no errors and no throttling. Throughput stayed constant: CPU 59.4–60.0 operations/s, GPU 79 fps.

• Consistent results: 4-hour tests matched the 20-minute tests within 0.2 °C in open air. Over the 36 hours the room varied by 3 °C while ΔT stayed within 1 °C.

i.MX 8M Plus (00377) results

Test setup

Hardware	00377 i.MX 8M Plus OSM-L SOM on the 00365 carrier board, in open air on a desk and in a sealed Polycase WA-42*16 plastic enclosure. For the UI and video tests the HDMI output was enabled with no monitor connected.
Cooling	No heatsink, or the passive heatsink used on the i.MX 95 (ATS-CPX045045025-162-C2-R0)
Software	Cargt Linux BSP (NXP lf-6.18.20). CPU governor: performance for the full-load tests, ondemand for the others.
CPU frequency caps	1200 and 1600 MHz, the two Cortex-A53 operating points of this part. The GC7000UL GPU has no frequency levels to cap.
Die temperature	The higher of the i.MX 8M Plus CPU and SoC thermal sensors, which report whole degrees
Ambient	A Dracal USB-PTH450 sensor in the room, 22–24 °C. In the enclosure, a second sensor measured the air inside it.

The loads are those used for the i.MX 95, with the NPU loads running on the VIP8000 NPU through the TensorFlow Lite VX delegate. Every test ran to a steady die temperature and none throttled.

Open air, no heatsink

Load	CPU cap MHz	ΔT °C	Max ambient	Performance
CPU + GPU + memory + NPU	1600	40.4	45 °C	209 inferences/s, CPU 23 ops/s, GPU 16 fps
CPU + GPU + memory	1600	33.4	52 °C	CPU 25 ops/s, GPU 17 fps
CPU + GPU + memory	1200	29.6	55 °C	CPU −19 %, memory −20 %, GPU unchanged
CPU, all cores	1600	30.6	54 °C	52 ops/s
CPU, all cores	1200	25.8	59 °C	40 ops/s (−22 %)
Image classification on the NPU	1600	29.0	56 °C	289 inferences/s
CPU at 50 %	1600	25.2	60 °C	–
GPU only	1600	25.1	60 °C	19 fps
Touchscreen UI + 1080p30 video	1600	25.3	60 °C	UI 33 fps, video 30 fps
Touchscreen UI + 1080p30 video	1200	23.6	61 °C	UI 30 fps, video 30 fps
1080p video decode, maximum rate	1600	20.3	65 °C	249 fps
1080p30 video playback	1600	18.7	66 °C	30 fps, no dropped frames
Idle	1600	18–19	66 °C	–

• No heatsink is needed in open air, even at full load: the hottest die reading in any test was 64 °C.

• The NPU adds 7–10 °C of ΔT, about as on the i.MX 95. With everything running, the NPU rate drops 28 % against the NPU alone.

• Capping the CPU at 1200 MHz saves 4–5 °C under CPU load and costs about 20 % of CPU and memory throughput. For the UI it costs 8 % of the frame rate for under 2 °C.

• Video costs almost nothing: the VPU adds about 1 °C at 30 fps.

• The UI is limited by the GPU, not temperature: the touchscreen UI runs at 30–33 fps on the GC7000UL, against 49 fps on the i.MX 95.

• The governor makes no difference: ondemand and performance give the same temperature within 0.3 °C at the same cap and load.

Open air, passive heatsink

Load	CPU cap MHz	ΔT °C (no heatsink → passive)	Max ambient, no heatsink → passive
CPU + GPU + memory + NPU	1600	40.4 → 28.4	45 → 57 °C
CPU + GPU + memory	1600	33.4 → 23.9	52 → 61 °C
CPU + GPU + memory	1200	29.6 → 21.5	55 → 63 °C
CPU, all cores	1600	30.6 → 20.7	54 → 64 °C
CPU, all cores	1200	25.8 → 18.1	59 → 67 °C
Image classification on the NPU	1600	29.0 → 19.5	56 → 65 °C
CPU at 50 %	1600	25.2 → 17.4	60 → 68 °C
GPU only	1600	25.1 → 18.6	60 → 66 °C
Touchscreen UI + 1080p30 video	1600	25.3 → 18.6	60 → 66 °C
Touchscreen UI + 1080p30 video	1200	23.6 → 17.4	61 → 68 °C
1080p30 video playback	1600	18.7 → 13.8	66 → 71 °C
Idle	1600	18–19 → 12.9	66 → 72 °C

• The heatsink lowers ΔT by 26–33 %, which allows 5–12 °C warmer air. As on the i.MX 95, the gain is largest under the heaviest loads.

• Performance is unchanged: no test throttled with or without the heatsink. The hottest die reading with the heatsink was 52 °C.

• Tests of the heatsink in the sealed enclosure are planned.

Enclosure, no heatsink

The same loads inside the sealed enclosure. ΔT is still measured against the room air outside the enclosure. Each test ran about 40 minutes to settle.

Load	CPU cap MHz	ΔT °C (open air → enclosure)	Max ambient	Performance
CPU + GPU + memory + NPU	1600	40.4 → 50.3	35 °C	211 inferences/s, CPU 22 ops/s, GPU 15 fps
CPU + GPU + memory + NPU	1200	45.5	39 °C	207 inferences/s, CPU −20 %, GPU 14 fps
CPU + GPU + memory	1600	33.4 → 42.7	42 °C	CPU 25 ops/s, GPU 17 fps
CPU + GPU + memory	1200	29.6 → 38.4	47 °C	CPU −20 %, GPU unchanged
CPU, all cores	1600	30.6 → 37.1	48 °C	51 ops/s
CPU, all cores	1200	25.8 → 31.4	54 °C	40 ops/s (−22 %)
Image classification on the NPU	1600	29.0 → 37.1	48 °C	288 inferences/s
Touchscreen UI + 1080p30 video	1600	25.3 → 33.7	51 °C	UI 32 fps, video 30 fps
GPU only	1600	25.1 → 33.2	52 °C	19 fps
CPU at 50 %	1600	25.2 → 30.6	54 °C	–
1080p30 video playback	1600	18.7 → 25.4	60 °C	30 fps
Idle	1600	18–19 → 24.0	61 °C	–

• No test throttled. The hottest die reading was 76 °C, with everything loaded at 1600 MHz: 9 °C under the design limit.

• The enclosure costs 5–10 °C of ambient: about 5 °C at idle and under light loads, 8–10 °C under heavy ones. The air inside the enclosure ran 3–7 °C above the room.

• Only the NPU worst case may need a heatsink. CPU, GPU, memory and NPU all fully loaded allow 35 °C room air at 1600 MHz. Capping the CPU at 1200 MHz raises that to 39 °C, at a cost of 20 % of CPU throughput; the NPU rate is unchanged. For warmer rooms this case needs a heatsink.

i.MX 93 (00363) results

Test setup

Hardware	00363 i.MX 93 OSM-L SOM on the 00365 carrier board, no heatsink, in open air on a desk and in the same sealed Polycase WA-42*16 enclosure, mounted as the i.MX 8M Plus was. The display output was enabled with no panel connected.
Software	Cargt Linux BSP (NXP lf-6.18.20)
CPU frequency	The two Cortex-A55 cores run at a fixed 1.7 GHz; the i.MX 93 has no CPU frequency levels to cap or governor to choose. It has no 3D GPU or video decoder, so the touchscreen UI is drawn by the CPU and there are no GPU or video tests.
Die temperature	The i.MX 93 CPU thermal sensor
Ambient	A Dracal USB-PTH450 sensor in the room, 22–24 °C. In the enclosure, a second sensor measured the air inside it.

The loads are those used for the other SOMs, without the GPU. Every test ran to a steady die temperature and none throttled. The open-air tests of CPU and memory, all CPU cores and idle were repeated two or three times; repeats agreed within 0.5 °C, and the tables give their mean. NPU tests on the Ethos-U65 are planned.

Open air and enclosure, no heatsink

Load	ΔT °C (open air → enclosure)	Max ambient, open air	Max ambient, enclosure	Performance (open air → enclosure)
CPU + memory	20.5 → 24.0	64 °C	61 °C	CPU 14 → 13 ops/s, memory 18,700 → 18,400 ops/s
CPU, all cores	19.4 → 22.5	66 °C	63 °C	29 → 28 ops/s
Touchscreen UI (drawn by the CPU)	20.0 → 23.2	65 °C	62 °C	13 → 12 fps
CPU at 50 %	16.5 → 19.7	69 °C	65 °C	–
Idle	12.8 → 16.1	72 °C	69 °C	–

• No heatsink is needed in open air or in the enclosure: the hottest die reading in any test was 48 °C, 37 °C under the design limit.

• The enclosure costs about 3 °C of ambient. The air inside it ran 1.6 °C above the room at idle and 2.6 °C at full load.

• The touchscreen UI loads the CPU almost fully, because the i.MX 93 has no 3D GPU: it runs about as warm as the CPU stress test, at 12–13 fps.

• Throttling works as on the other SOMs. With the throttle point lowered to test it, the kernel held the die at the throttle point by idling the CPU, at a cost of 18–30 % of CPU throughput.

STM32MP257F (00395) results

Test setup

Hardware	00395 STM32MP257F OSM-L SOM on the 00365 carrier board, no heatsink, in open air on a desk, no display or cameras connected. For the GPU and video tests the LVDS display output was enabled with no panel connected.
Software	Cargt OpenSTLinux BSP (scarthgap, kernel 6.6.78). CPU governor: performance for the full-load tests, ondemand for the others.
CPU frequency caps	1200 and 1500 MHz, the two Cortex-A35 operating points. The GC8000UL GPU has no frequency levels to cap.
Die temperature	The higher of the STM32MP25 CPU and SoC thermal sensors
Ambient	A Dracal USB-PTH450 sensor in the room, 22–24 °C

The loads are those used for the other SOMs: stress-ng on the CPU and memory, glmark2 on the GPU and 1080p30 H.264 playback on the hardware video decoder. The touchscreen UI and NPU loads were not run. Every test ran to a steady die temperature and none throttled. Where a test was repeated, the table gives the range.

Open air, no heatsink

Load	CPU cap MHz	ΔT °C	Max ambient	Performance
CPU + GPU + memory	1500	25.6–25.8	74 °C	CPU 8.5 ops/s, memory 11,200–11,500 ops/s, GPU 5 fps
CPU + GPU + memory	1200	24.0	76 °C	CPU 7.0 ops/s, memory 9,400 ops/s (both −18 %), GPU 5 fps
GPU only	1500	21.5	79 °C	6 fps
CPU + memory	1500	19.8	80 °C	CPU 5.7–5.9 ops/s, memory 14,800–15,000 ops/s
CPU + memory	1200	17.7	82 °C	CPU 4.8 ops/s (−18 %), memory 12,500 ops/s (−16 %)
CPU, all cores	1500	18.8	81 °C	18.8 ops/s
CPU, all cores	1200	17.1	83 °C	15.0 ops/s (−20 %)
1080p30 video playback	1500	17.4	83 °C	30 fps
CPU at 50 %	1500	16.5–16.9	83 °C	–
CPU at 50 %	1200	15.5–15.6	84 °C	–
Idle, display output enabled	1500	15.2	85 °C	–
Idle, no display output	1500	14.0–14.3	86 °C	–

• No heatsink is needed in open air: the hottest die reading in any test was 49 °C, 51 °C under the design limit. The maximum ambient figures are far above the 22–24 °C room the tests ran in, so treat them as extrapolations (see the note under How to read the results).

• The GPU is the largest heat source. The GPU alone raises ΔT 6 °C above idle, more than all CPU cores do. With CPU, GPU and memory together the GPU runs glmark2 at only 5–6 fps.

• Capping the CPU at 1200 MHz saves only about 2 °C and costs 16–20 % of CPU and memory throughput. The Cortex-A35 cores make little heat.

• Video costs about 2 °C over idle with the display output enabled.

• These tests are listed in full here; they are not in Thermal Test Data. Enclosure and NPU tests are planned.

Choosing cooling for your product

To estimate the cooling a product needs:

1. Find the row closest to your product’s load and CPU frequency cap.

2. Take the maximum ambient for your environment and cooling.

3. Compare it with the warmest air your product will see. Allow for the inside of your enclosure being warmer than the room: 2–7 °C in the sealed enclosure tested here.

For example, an indoor i.MX 95 product in a sealed plastic enclosure, with the CPU capped at 1404 MHz and the passive heatsink, allows 52 °C room air at the worst-case stress load. That is a margin of about 20 °C over a 30 °C room.

These results come from one SOM of each type in one orientation on the 00365 carrier board. A different carrier, enclosure, mounting orientation or airflow changes them. Treat them as a starting point and verify your final design. For help choosing cooling for your product, contact Cargt.

Test data

Thermal Test Data lists every test behind this page, with its temperatures, clock frequencies, loads and performance, and has them as a CSV file to download.