ImportBot: Import: hbmmon, @matstatman, @zefrench6961, @phoenixone7, @bigfoot1144; repos matstatman/hbmmon

2026-10-30T08:38:26Z

Import: hbmmon, @matstatman, @zefrench6961, @phoenixone7, @bigfoot1144; repos matstatman/hbmmon

← Older revision		Revision as of 08:38, 30 October 2026
Line 1:		Line 1:
			{{DISPLAYTITLE:hbmmon}}
	{{Status\|Reported}} {{Cards\|8 GB}}		{{Status\|Reported}} {{Cards\|8 GB}}

ImportBot: Import: hbmmon, @matstatman, @zefrench6961, @phoenixone7, @bigfoot1144; repos matstatman/hbmmon

2026-10-30T08:04:31Z

Import: hbmmon, @matstatman, @zefrench6961, @phoenixone7, @bigfoot1144; repos matstatman/hbmmon

New page

{{Status|Reported}} {{Cards|8 GB}}

'''hbmmon''' is an nvtop-style terminal monitor for the CMP 170HX that shows each HBM stack separately: its temperature, vendor, density, date code and which memory partitions are live, harvested or defective, next to the GPC fuse map, PCIe link and power. It reads these from the card rather than from tables. Author: @matstatman (GitHub matstatman). Repository: https://github.com/matstatman/hbmmon {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288|repo=https://github.com/matstatman/hbmmon}}.

== Requirements ==
* Root, kernel headers, and <code><nowiki>nvidia-smi</nowiki></code> on the PATH. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288|repo=https://github.com/matstatman/hbmmon}}.
* For the full view (stack identity), a driver whose GSP booter opens the HBM debug-bridge registers. The author's box runs a custom 610.43.02 build that has them. On a stock driver these writes are dropped and every capture reports DROPPED; use <code><nowiki>--read-only</nowiki></code> there. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288|repo=https://github.com/matstatman/hbmmon}}.
* It ships its own copy of the [[gpcprobe]] kernel module with one extra, restricted write call. The two modules are not interchangeable. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288|repo=https://github.com/matstatman/hbmmon}}.

== Install and usage ==
<syntaxhighlight lang="bash">
git clone https://github.com/matstatman/hbmmon && cd hbmmon
make # builds gpcprobe.ko against the running kernel
sudo insmod gpcprobe.ko # creates /dev/gpcprobe (root only)
sudo python3 ./hbmmon.py --read-only # BAR0 reads only, no sweep
sudo python3 ./hbmmon.py # startup identity sweep, then a live view every 2 s
sudo python3 ./hbmmon.py --once # sweep and one frame
sudo python3 ./hbmmon.py --ids # per-stack identity table, then exit
sudo python3 ./hbmmon.py --mock # synthetic card, no hardware
</syntaxhighlight>
Other options: <code><nowiki>--live N</nowiki></code> (refresh seconds), <code><nowiki>--card N</nowiki></code>, <code><nowiki>--no-color</nowiki></code>. <code><nowiki>eccprobe.py</nowiki></code> in the same repository shows the state of the SRAM ECC units. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288|repo=https://github.com/matstatman/hbmmon}}.

== Safety ==
{{Status|Warning}} The default run writes to the HBM debug bridge. At startup it arms one read-type instruction per live memory partition (4 register writes each), and after each partition it compares two sentinel registers and the kernel log against a baseline and aborts on any change; if it aborts, stop and reboot the GPU. The module only allows writes to the two bridge control registers. The author excluded the data latch because writing it wedged the bridge on the whole card until a reboot in his 2026-09-25 tests. <code><nowiki>--read-only</nowiki></code> and <code><nowiki>--mock</nowiki></code> write nothing. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288|repo=https://github.com/matstatman/hbmmon}}.

== Discord ==
* @matstatman: "a 170hx has like 10 temp sensors". He wants per-HBM-stack temperature graphs when people debug dropouts. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556770474695069718}} Most owners had only seen GPU, hotspot and memory temperatures. {{Source|handle=zefrench6961|date=2026-10-06|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1557175100824879127}} {{Source|handle=phoenixone7|date=2026-10-07|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1557190829087395901}}
* He points people to the repository and offers settings to anyone who writes a control panel or CLI for features real NVIDIA products lack, such as per-unit SRAM ECC control. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288}} {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556742488063090779}} {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556748200084504688}}
* @catcat0458's hbmmon screenshot of an 8 GB card unlocked to 64 GB showed 70 SMs; the current unlocker gives 74, so he was on an older unlocker. {{Source|handle=matstatman|date=2026-10-08|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1557782065959669861}} {{Source|handle=bigfoot1144|date=2026-10-08|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1557782487688544361}} {{Source|handle=matstatman|date=2026-10-08|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1557783127814971503}}

== Caveats ==
* The measured stack identities cover the author's 8 GB cards only (SK Hynix, 16 Gb 8-high). {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288|repo=https://github.com/matstatman/hbmmon}}.
* It depends on a custom driver and on research notes that are not in the repository. No licence file; the kernel module is GPL-2.0. {{Source|handle=matstatman|date=2026-10-05|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1556758870565593288|repo=https://github.com/matstatman/hbmmon}}.

== See also ==
* [[Monitoring]]
* [[gpcprobe]]
* [[Tools]]

{{ImportFooter}}
[[Category:Tools]]
[[Category:8 GB]]

Hbmmon - Revision history

ImportBot: Import: hbmmon, @matstatman, @zefrench6961, @phoenixone7, @bigfoot1144; repos matstatman/hbmmon

ImportBot: Import: hbmmon, @matstatman, @zefrench6961, @phoenixone7, @bigfoot1144; repos matstatman/hbmmon