Thermal limits - Revision history

ImportBot: Core and HBM temperature thresholds, throttling and drop-off behaviour, what temperatures people run at, how to read the sensors; from the cmpunlocker Discord (2026-10-04 to 10-27), cachenetics/170tune, PixelML/club-170hx, amoghmunikote/170th-Street, matstatman/hbmmon and Ivan's lab notes; gap-fill from #general 2026-10-23 (400 W spike at a 250 W limit), @djpandaman, @xeax

2026-10-30T10:02:38Z

Core and HBM temperature thresholds, throttling and drop-off behaviour, what temperatures people run at, how to read the sensors; from the cmpunlocker Discord (2026-10-04 to 10-27), cachenetics/170tune, PixelML/club-170hx, amoghmunikote/170th-Street, matstatman/hbmmon and Ivan's lab notes; gap-fill from #general 2026-10-23 (400 W spike at a 250 W limit), @djpandaman, @xeax

← Older revision		Revision as of 10:02, 30 October 2026
Line 40:		Line 40:
	== Throttling behaviour ==		== Throttling behaviour ==
	* When the cap or the temperature limit bites, the core clock drops and with it memory bandwidth and tokens per second. On a 10 GB card with all SMs unlocked, the power cap tanked memory bandwidth even with the clock locked. {{Source\|handle=djpandaman\|date=2026-10-12\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1559236914391949484}} See [[Undervolting and power limits]].		* When the cap or the temperature limit bites, the core clock drops and with it memory bandwidth and tokens per second. On a 10 GB card with all SMs unlocked, the power cap tanked memory bandwidth even with the clock locked. {{Source\|handle=djpandaman\|date=2026-10-12\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1559236914391949484}} See [[Undervolting and power limits]].
			* {{Status\|Warning}} '''A 250 W cap did not bound a fully SM-unlocked 10 GB card.''' @djpandaman's card at 122 SMs briefly reported about 400 W of board power during benchmarks with the limit at 250 W (he was unsure whether the reading was measured or calculated), and all of it ends up in the same heatsink and VRM. @xeax worries the VRM MOSFETs get overloaded and doubts he could cool that load. On such a card, watch core and HBM temperatures under real load instead of trusting the cap, and undervolt (see [[Undervolting and power limits#10 GB cards]] and [[Power supplies and cabling#PSU sizing]]). {{Source\|handle=xeax\|date=2026-10-23\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1563222111592521881}}<br />{{Source\|handle=djpandaman\|date=2026-10-23\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1563222404627431497}}
	* Prefill is where cards run hottest: @c133's cards spike over their cap during long prefills {{Source\|handle=c133\|date=2026-10-09\|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1558163141723037798}}; PixelML's worst-airflow card reached 80 to 85 C core on long-context prefill at 180 W, which is why they lowered the cap {{Source\|repo=https://github.com/PixelML/club-170hx}}.		* Prefill is where cards run hottest: @c133's cards spike over their cap during long prefills {{Source\|handle=c133\|date=2026-10-09\|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1558163141723037798}}; PixelML's worst-airflow card reached 80 to 85 C core on long-context prefill at 180 W, which is why they lowered the cap {{Source\|repo=https://github.com/PixelML/club-170hx}}.
	* Check whether the card is in hardware power-brake slowdown: <code><nowiki>nvidia-smi -q \| grep -A1 'HW Power Brake Slowdown'</nowiki></code>. {{Source\|repo=https://github.com/PixelML/club-170hx}}		* Check whether the card is in hardware power-brake slowdown: <code><nowiki>nvidia-smi -q \| grep -A1 'HW Power Brake Slowdown'</nowiki></code>. {{Source\|repo=https://github.com/PixelML/club-170hx}}
Line 85:		Line 86:
	* <code><nowiki>nvidia-smi -q -d TEMPERATURE</nowiki></code> {{Source\|handle=xeax\|date=2026-10-15\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560253301985906692}}, or for logging <code><nowiki>nvidia-smi --query-gpu=index,temperature.gpu,temperature.memory,power.draw --format=csv</nowiki></code> {{Source\|repo=https://github.com/PixelML/club-170hx}}. Watch <code><nowiki>temperature.memory</nowiki></code>: HBM is usually the warmest sensor under load. {{Source\|repo=https://github.com/cachenetics/170tune}}		* <code><nowiki>nvidia-smi -q -d TEMPERATURE</nowiki></code> {{Source\|handle=xeax\|date=2026-10-15\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560253301985906692}}, or for logging <code><nowiki>nvidia-smi --query-gpu=index,temperature.gpu,temperature.memory,power.draw --format=csv</nowiki></code> {{Source\|repo=https://github.com/PixelML/club-170hx}}. Watch <code><nowiki>temperature.memory</nowiki></code>: HBM is usually the warmest sensor under load. {{Source\|repo=https://github.com/cachenetics/170tune}}
	* The card has about 10 temperature sensors, including one per HBM stack. {{Source\|handle=matstatman\|date=2026-10-05\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556770474695069718}} [https://github.com/matstatman/hbmmon matstatman/hbmmon] shows per-stack temperature, but it writes to the HBM debug interface by default and needs a custom driver; use its read-only mode on a stock driver. {{Source\|handle=matstatman\|date=2026-10-05\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288\|repo=https://github.com/matstatman/hbmmon\|note=no license file}}		* The card has about 10 temperature sensors, including one per HBM stack. {{Source\|handle=matstatman\|date=2026-10-05\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556770474695069718}} [https://github.com/matstatman/hbmmon matstatman/hbmmon] shows per-stack temperature, but it writes to the HBM debug interface by default and needs a custom driver; use its read-only mode on a stock driver. {{Source\|handle=matstatman\|date=2026-10-05\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288\|repo=https://github.com/matstatman/hbmmon\|note=no license file}}
	* The 10 GB card's memory runs a few degrees warmer than the 8 GB card's (two more half stacks active): 41 C against 37 C at ~~the same~~ 33 W idle after a repaste; before the repaste the gap was 13 C, which pointed at the thermal interface. {{Source\|handle=ghostinthedatacenter_87281\|date=2026-10-13\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559597067352023152}}<br />{{Source\|handle=colo2442\|date=2026-10-13\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559599442166812784}}<br />{{Source\|handle=djpandaman\|date=2026-10-13\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559649510827757569}}		* The 10 GB card's memory runs a few degrees warmer than the 8 GB card's (two more half stacks active): 41 C against 37 C, both at about 33 W idle, after a repaste; before the repaste the gap was 13 C, which pointed at the thermal interface. The owner used PTM on the 8 GB card and then ran out; what went on the 10 GB card is not stated. {{Source\|handle=ghostinthedatacenter_87281\|date=2026-10-13\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559597067352023152}}<br />{{Source\|handle=colo2442\|date=2026-10-13\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559599442166812784}}<br />{{Source\|handle=djpandaman\|date=2026-10-13\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559649510827757569}}<br />{{Source\|handle=ghostinthedatacenter_87281\|date=2026-10-13\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559651382816936108}}<br />{{Source\|handle=ghostinthedatacenter_87281\|date=2026-10-13\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559653025700909117}}
	* A server BMC cannot read the temperature of a card on a generic passive riser, so it will not ramp the fans for it. {{Source\|handle=fafinet\|date=2026-10-21\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1562300430439157850}}		* A server BMC cannot read the temperature of a card on a generic passive riser, so it will not ramp the fans for it. {{Source\|handle=fafinet\|date=2026-10-21\|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1562300430439157850}}

ImportBot: Core and HBM temperature thresholds, throttling and drop-off behaviour, what temperatures people run at, how to read the sensors; from the cmpunlocker Discord (2026-10-04 to 10-27), cachenetics/170tune, PixelML/club-170hx, amoghmunikote/170th-Street, matstatman/hbmmon and Ivan's lab notes; sources: @zorg33, @c133, @markshowa, @matstatman, @xeax, @rh3dai, @analyticeth, @sojun80, @replacedriver47, @spectakkular, @bigptoughneigh, @mountains1605 (+25 more); GitHub PixelML/club-17

2026-10-30T08:51:11Z

Core and HBM temperature thresholds, throttling and drop-off behaviour, what temperatures people run at, how to read the sensors; from the cmpunlocker Discord (2026-10-04 to 10-27), cachenetics/170tune, PixelML/club-170hx, amoghmunikote/170th-Street, matstatman/hbmmon and Ivan's lab notes; sources: @zorg33, @c133, @markshowa, @matstatman, @xeax, @rh3dai, @analyticeth, @sojun80, @replacedriver47, @spectakkular, @bigptoughneigh, @mountains1605 (+25 more); GitHub PixelML/club-17

Show changes

ImportBot: Core and HBM temperature thresholds, throttling and drop-off behaviour, what temperatures people run at, how to read the sensors; from the cmpunlocker Discord (2026-10-04 to 10-27), cachenetics/170tune, PixelML/club-170hx, amoghmunikote/170th-Street, matstatman/hbmmon and Ivan's lab notes; sources: @zorg33, @c133, @markshowa, @rh3dai, @matstatman, @xeax, @analyticeth, @sojun80, @replacedriver47, @spectakkular, @bigptoughneigh, @mountains1605 (+24 more); GitHub PixelML/club-17

2026-10-30T08:05:11Z

Core and HBM temperature thresholds, throttling and drop-off behaviour, what temperatures people run at, how to read the sensors; from the cmpunlocker Discord (2026-10-04 to 10-27), cachenetics/170tune, PixelML/club-170hx, amoghmunikote/170th-Street, matstatman/hbmmon and Ivan's lab notes; sources: @zorg33, @c133, @markshowa, @rh3dai, @matstatman, @xeax, @analyticeth, @sojun80, @replacedriver47, @spectakkular, @bigptoughneigh, @mountains1605 (+24 more); GitHub PixelML/club-17

New page

The CMP 170HX runs hot for its power, and it has two temperatures that matter: the GPU core and the HBM memory. Under
load the HBM is usually the hotter one, and it is the one that limits memory overclocks and corrupts data first.
Keep the core under about 80 C and the HBM under about 80 to 85 C for daily use.

== Thresholds ==
{| class="wikitable"
! Sensor !! Temperature !! What happens !! Source
|-
| Core || about 76 C || throttling starts, according to a rough cooler model; keep core and HBM under 75 C || {{Status|Theory}} {{Source|handle=zorg33|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1562110412395380767}}
|-
| Core || 80 C || PixelML's stop rule for core temperature || {{Source|handle=c133|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1562161732137386127|repo=https://github.com/PixelML/club-170hx}} (Apache-2.0)
|-
| Core || above about 80 C || GA100 leakage rises with temperature and the card runs away thermally || {{Source|handle=markshowa|date=2026-10-18|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561324784598384784|repo=https://github.com/amoghmunikote/170th-Street}} (no license stated)
|-
| Core || 85 C || maximum operating temperature (read from the card) {{Source|handle=rh3dai|date=2026-10-04|link=https://discord.com/channels/1518634379604136178/1518695289278566450/1556287917775655053|repo=https://github.com/cachenetics/170tune}} (MIT); "the cards will throttle at 85" {{Source|handle=matstatman|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558212817121972318}}; "85 degrees is the maximum, so you're most likely bottlenecked by cooling" {{Source|handle=xeax|date=2026-10-15|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560252687839133757}} || Source: [https://github.com/cachenetics/170tune cachenetics/170tune]
|-
| Core || 95 C || hardware slowdown Source: [https://github.com/cachenetics/170tune cachenetics/170tune]; the card falls off the bus and turns off {{Source|handle=analyticeth|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1557491900934791409}} || see cells
|-
| Core || 100 C || one card reached it by accident, cut its power as hard as it could, and still works || {{Source|handle=sojun80|date=2026-10-06|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556855494008643698}}
|-
| HBM || about 75 to 76 C || a memory overclock (NDIV 70 and up) starts to corrupt data; 170tune qualifies NDIV 70 only with HBM at or below about 76 C || Source: [https://github.com/cachenetics/170tune cachenetics/170tune]
|-
| HBM || 80 C || "hbm doesn't have issues until 80c or so" {{Source|handle=replacedriver47|date=2026-10-10|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558399070006157436}}; "anything above 80 is too hot" {{Source|handle=spectakkular|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558215172294840442}} || see cells
|-
| HBM || 85 C || PixelML's stop rule for memory; above 85 C JEDEC halves the refresh interval, so keep stock refresh settings || Source: [https://github.com/PixelML/club-170hx PixelML/club-170hx] {{Source|handle=rh3dai|date=2026-10-04|link=https://discord.com/channels/1518634379604136178/1518695289278566450/1556287917775655053}}, repo [https://github.com/vektorprime/170tune vektorprime/170tune] by vektorprime (fork of cachenetics/170tune) on GitHub (MIT)
|-
| HBM || 95 C || maximum operating temperature Source: [https://github.com/cachenetics/170tune cachenetics/170tune]; "a few minutes" at 95 C corrupts data in the WPR {{Source|handle=bigptoughneigh|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558266699218812929}} || see cells
|}

=== Where people disagree ===
{{Status|Contested}}
* How hot is too hot: @mountains1605 called @arabus_74064's temperatures on Qwen3.8 Next Flash too hot, while @arabus_74064 thought them normal at full load. {{Source|handle=mountains1605|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558199574974304338}} {{Source|handle=arabus_74064|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558210159942041702}} @replacedriver47: the SMs are not the problem, the HBM is, and it is more prone to heat damage {{Source|handle=replacedriver47|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558217213717254195}}; @rrickk420: "you're frying your mem" {{Source|handle=rrickk420|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558221699038576726}}; @bigptoughneigh: memory is not always hotter than the core under load {{Source|handle=bigptoughneigh|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558266699218812929}}; @dottoreargus barely reaches 65 C on the same model {{Source|handle=dottoreargus|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558215409625333812}}.
* @impactic thought the card shuts itself down above 95 C; @prefix10k thought 85 C is already bad. {{Source|handle=prefix10k|date=2026-10-06|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556853048943452211}} {{Source|handle=impactic|date=2026-10-06|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556861368513331264}}
* The throttle point: 76 C by @zorg33's model {{Source|handle=zorg33|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1562110412395380767}} against 85 C by @matstatman {{Source|handle=matstatman|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528826330454429706/1558212817121972318}} and the card's own maximum operating value Source: [https://github.com/cachenetics/170tune cachenetics/170tune].

== Without airflow the card dies fast ==
* {{Status|Proven}} A card without a fan reaches 95 C quickly even at its 40 W idle, falls off the bus and turns off. {{Source|handle=analyticeth|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1557491900934791409}} {{Source|handle=nikhilr1696|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1557492862344630364}} @programmerboi's passive card stopped showing in <code><nowiki>nvidia-smi</nowiki></code> at 85 C while idle {{Source|handle=programmerboi|date=2026-10-16|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1560492431768813611}}, and a new 10 GB card "clocked out in under 5 mins without cooling. way worse than V100" {{Source|handle=programmerboi|date=2026-10-25|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1563704621388529725}}. The heatsink is designed for high-airflow servers and does not work even at idle without it. {{Source|handle=civiloid|date=2026-10-16|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1560561874188374087}} {{Source|handle=globalapathy|date=2026-10-25|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1563704870073143377}}
* {{Status|Warning}} A failed fan means a dead card or worse; @nikhilr1696 only runs these cards supervised. {{Source|handle=nikhilr1696|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1557491750145232997}} Use a decent fan, not necessarily a blower. {{Source|handle=cibernox|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1557506106509631579}}

== Throttling behaviour ==
* When the cap or the temperature limit bites, the core clock drops and with it memory bandwidth and tokens per second. On a 10 GB card with all SMs unlocked, the power cap tanked memory bandwidth even with the clock locked. {{Source|handle=djpandaman|date=2026-10-12|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1559236914391949484}} See [[Undervolting and power limits]].
* Prefill is where cards run hottest: @c133's cards spike over their cap during long prefills {{Source|handle=c133|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1558163141723037798}}; PixelML's worst-airflow card reached 80 to 85 C core on long-context prefill at 180 W, which is why they lowered the cap Source: [https://github.com/PixelML/club-170hx PixelML/club-170hx].
* Check whether the card is in hardware power-brake slowdown: <code><nowiki>nvidia-smi -q | grep -A1 'HW Power Brake Slowdown'</nowiki></code>. Source: [https://github.com/PixelML/club-170hx PixelML/club-170hx]
* Fixes, in order: improve the air path (see [[Fans and ducts]]), lower the cap ({{Source|handle=xeax|date=2026-10-15|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560253213268250645}}: <code><nowiki>sudo nvidia-smi -i 0 -pl 200</nowiki></code>), then consider water ([[Water cooling]]). These chips run 400 W on SXM4 boards; better cooling is what unlocks more of them. {{Source|handle=xeax|date=2026-10-15|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560253818329899039}}

== What temperatures people run at ==
{| class="wikitable sortable"
! Cooling !! Cap !! Hottest reading !! Source
|-
| Water block, full load || 250 W || 50 C || {{Source|handle=leenodie|date=2026-10-26|link=https://discord.com/channels/1518634379604136178/1545982693903306813/1564340979065163910}}
|-
| Water block, 360 mm radiator, fans at minimum || 180 W || 45 C || Source: [https://github.com/amoghmunikote/170th-Street amoghmunikote/170th-Street]
|-
| Water, general || not given || 50 to 60 C || {{Source|handle=endsjeremy|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1561901936494649345}}
|-
| Air, unlocked 64 GB card || 175 W || 60 C core, 65 C HBM || {{Source|handle=rrickk420|date=2026-10-08|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1557800123457736795}}
|-
| One 80 mm fan per pair, no duct || 185 W || 63 C || {{Source|handle=c133|date=2026-10-09|link=https://discord.com/channels/1518634379604136178/1528513735575081070/1558160764370690100}}
|-
| S12038-4K push only, four cards, cold garage || 200 W || 62 C || {{Source|handle=alanoo|date=2026-10-26|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1564379568801058877}}
|-
| Blower at 50 to 60 % || 200 W || 60 C || {{Source|handle=68x0|date=2026-10-15|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560301937675599944}}
|-
| P12 Pro on a shroud || 175 W || 65 C || {{Source|handle=rrickk420|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1518691940646781070/1557456081607860236}}
|-
| Sealed funnel || 150 W || 70 C || {{Source|handle=globalapathy|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1518691940646781070/1557451493379154061}}
|-
| S12038-4K push-pull, 24 C room || 300 W || HBM 72 C || {{Source|handle=wakeupmbrc1|date=2026-10-16|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1560655488331546715}}
|-
| P12 Pro push and pull on 3-card plenums, 10 minutes full power || full || 80 C || {{Source|handle=reala7096|date=2026-10-08|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1557593651037737051}}
|-
| One card, 6-way inference soak for 10 minutes || 200 W || HBM 84 to 85 C, GPU 78 C || Source: [https://github.com/cachenetics/170tune cachenetics/170tune]
|-
| Four cards, GLM-5.3-Flash TP4 || 165 W || HBM 82 to 83 C || Source: [https://github.com/PixelML/club-170hx PixelML/club-170hx]
|-
| Eight cards in a server with a fan wall, same cap, by slot || 180 W || 58 to 68 C on two slots per CPU, 81 to 84 C on the other two || {{Source|handle=darkmaniac7|date=2026-10-28|note=lab notes}}
|-
| Four idle cards, one 80 mm blower on a printed duct || idle || core 37 to 38 C, memory 41 to 51 C || Source: [https://github.com/PixelML/club-170hx PixelML/club-170hx]
|-
| Sandwiched cards with shared intake || under load || 34 to 42 C || {{Source|handle=moliminous|date=2026-10-11|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1558686919033094145}}
|}

== Reading the sensors ==
* <code><nowiki>nvidia-smi -q -d TEMPERATURE</nowiki></code> {{Source|handle=xeax|date=2026-10-15|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560253301985906692}}, or for logging <code><nowiki>nvidia-smi --query-gpu=index,temperature.gpu,temperature.memory,power.draw --format=csv</nowiki></code> Source: [https://github.com/PixelML/club-170hx PixelML/club-170hx]. Watch <code><nowiki>temperature.memory</nowiki></code>: HBM is usually the warmest sensor under load. Source: [https://github.com/cachenetics/170tune cachenetics/170tune]
* The card has about 10 temperature sensors, including one per HBM stack. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556770474695069718}} [https://github.com/matstatman/hbmmon matstatman/hbmmon] shows per-stack temperature, but it writes to the HBM debug interface by default and needs a custom driver; use its read-only mode on a stock driver. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288|repo=https://github.com/matstatman/hbmmon}} (no license file)
* The 10 GB card's memory runs a few degrees warmer than the 8 GB card's (two more half stacks active): 41 C against 37 C at the same 33 W idle after a repaste; before the repaste the gap was 13 C, which pointed at the thermal interface. {{Source|handle=ghostinthedatacenter_87281|date=2026-10-13|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559597067352023152}} {{Source|handle=colo2442|date=2026-10-13|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559599442166812784}} {{Source|handle=djpandaman|date=2026-10-13|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1559649510827757569}}
* A server BMC cannot read the temperature of a card on a generic passive riser, so it will not ramp the fans for it. {{Source|handle=fafinet|date=2026-10-21|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1562300430439157850}}

== Room and inlet temperature ==
* Your intake air sets the floor. Ivan's eight-card server raises its power caps only while the inlet is below 24 C and lowers them above 27 C; at about 2 kW into a room, the room became the limit before the fans did. See [[Chassis builds]]. {{Source|handle=darkmaniac7|date=2026-10-27|note=lab notes}}
* Twelve cards on dedicated circuits with an air conditioner pointed at them is still the warmest and loudest room in @gpmickey's house. {{Source|handle=gpmickey|date=2026-10-19|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1561664734179364944}}
* {{Status|Warning}} Do not feed cold winter air or sub-dew-point coolant to the cards: condensation. {{Source|handle=alanoo|date=2026-10-20|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1562111068204044420}} {{Source|handle=gnasherrr|date=2026-10-26|link=https://discord.com/channels/1518634379604136178/1534370835220004885/1564185826538160159}}
* Paste and pads matter for the last few degrees; see [[Water cooling]] for opening the card.

== See also ==
[[Cooling and power]] · [[Fans and ducts]] · [[Water cooling]] · [[Undervolting and power limits]] ·
[[P-states and idle power]] · [[Troubleshooting]] · [[Tools]]

{{ImportFooter}}
[[Category:Cooling and power]]

[[Category:8 GB]]
[[Category:10 GB]]