Memory unlock on the 10 GB card - Revision history

ImportBot: The 10 GB card's memory ceiling: 40 GB from cmpunlocker, 48 GB on a lucky 6th stack, and why the disputed 80 GB is aliasing not memory. Discord Oct 2026 and the research repos.; sources: @_j3ns, @programmerboi, @djpandaman, @artistosu, @bigfoot1144, @_orrorin, @darkmaniac7, @rh3dai, @atsasin2126, @marklintonca, @tekkenz, @scallion_9883 (+6 more); GitHub ahnguyen17/ga100-fuse-re, amoghmunikote/cmpunlocker

2026-10-30T10:02:17Z

The 10 GB card's memory ceiling: 40 GB from cmpunlocker, 48 GB on a lucky 6th stack, and why the disputed 80 GB is aliasing not memory. Discord Oct 2026 and the research repos.; sources: @_j3ns, @programmerboi, @djpandaman, @artistosu, @bigfoot1144, @_orrorin, @darkmaniac7, @rh3dai, @atsasin2126, @marklintonca, @tekkenz, @scallion_9883 (+6 more); GitHub ahnguyen17/ga100-fuse-re, amoghmunikote/cmpunlocker

ImportBot: The 10 GB card's memory ceiling: 40 GB from cmpunlocker, 48 GB on a lucky 6th stack, and why the disputed 80 GB is aliasing not memory. Discord Oct 2026 and the research repos.; sources: @_j3ns, @programmerboi, @djpandaman, @artistosu, @bigfoot1144, @darkmaniac7, @rh3dai, @atsasin2126, @marklintonca, @tekkenz, @scallion_9883, @ahnguyen17 (+4 more); GitHub ahnguyen17/ga100-fuse-re, amoghmunikote/cmpunlocker

2026-10-30T08:49:22Z

The 10 GB card's memory ceiling: 40 GB from cmpunlocker, 48 GB on a lucky 6th stack, and why the disputed 80 GB is aliasing not memory. Discord Oct 2026 and the research repos.; sources: @_j3ns, @programmerboi, @djpandaman, @artistosu, @bigfoot1144, @darkmaniac7, @rh3dai, @atsasin2126, @marklintonca, @tekkenz, @scallion_9883, @ahnguyen17 (+4 more); GitHub ahnguyen17/ga100-fuse-re, amoghmunikote/cmpunlocker

← Older revision		Revision as of 08:49, 30 October 2026
Line 69:		Line 69:
	== Known issues ==		== Known issues ==

	* A dead region that returns 0 and stores nothing ~~can be worked around by pinning~~ a ~~dummy~~ allocation over ~~it; a longer~~		* A dead region that returns 0 and stores nothing: AnalyticETH had an agent write a script that pins a 2 MB throwaway
	~~write~~-~~up by @analyticeth exists; see Discord~~.		allocation over the bad address at startup so nothing important lands there, rather than discarding the card.
	* Cards with HBM errors: the driver ~~can mark~~ bad ~~physical~~ blocks ~~at boot~~ (~~4 x~~ 64 KB ~~blocks cost~~ 8 MB)~~. A longer write~~-up		{{Source\|analyticeth\|2026-10-08\|https://discord.com/channels/1518634379604136178/1518692084729511967/1557566171195383850}}
	~~by @zombielurker exists~~; ~~see~~ [[Troubleshooting]] ~~and Discord~~.		* Cards with HBM errors: Zombie's card threw about 5,000 single-bit and some multi-bit errors per 5-minute
			<code>memtest_vulkan</code> run, always at the same addresses (around 30 GB and 61 GB). He mapped the failing
			addresses to physical memory and used the driver's boot-time bad-block marking; only 4 blocks (64 KB each) were bad,
			costing 8 MB (63.06 to 63.05 GB), after which <code>memtest_vulkan</code> and <code>gpu-burn</code> passed with ECC
			off. The script must run every boot; the card refuses if there are too many errors (multiple GB). See
			[[Troubleshooting]]. {{Source\|zombielurker\|2026-10-05\|https://discord.com/channels/1518634379604136178/1518692084729511967/1556530076105838723}}

	== Status ==		== Status ==

ImportBot: The 10 GB card's memory ceiling: 40 GB from cmpunlocker, 48 GB on a lucky 6th stack, and why the disputed 80 GB is aliasing not memory. Discord Oct 2026 and the research repos.; sources: @_j3ns, @programmerboi, @djpandaman, @artistosu, @bigfoot1144, @darkmaniac7, @rh3dai, @atsasin2126, @marklintonca, @tekkenz, @scallion_9883, @ahnguyen17 (+2 more); GitHub ahnguyen17/ga100-fuse-re, amoghmunikote/cmpunlocker

2026-10-30T08:04:43Z

The 10 GB card's memory ceiling: 40 GB from cmpunlocker, 48 GB on a lucky 6th stack, and why the disputed 80 GB is aliasing not memory. Discord Oct 2026 and the research repos.; sources: @_j3ns, @programmerboi, @djpandaman, @artistosu, @bigfoot1144, @darkmaniac7, @rh3dai, @atsasin2126, @marklintonca, @tekkenz, @scallion_9883, @ahnguyen17 (+2 more); GitHub ahnguyen17/ga100-fuse-re, amoghmunikote/cmpunlocker

New page

{{Unlock
|unlocks=Memory on the 10 GB card (40 GB, 48 GB on a lucky card)
|variant=10 GB
|driver_version=cmpunlocker-patched driver (610+)
|unlock_build=cmpunlocker (40 GB) + RIR software override (6th stack)
|steps_page=Memory unlock on the 10 GB card
|status=Reported
|tested_by=_j3ns
|source=https://github.com/amoghmunikote/cmpunlocker
}}

The 8 GB card unlocks to a flat '''64 GB''' (see [[Driver unlock]]). The 10 GB card is different and the ceiling is a
lottery. {{Status|warning}} Do not expect 80 GB.

== 40 GB (the baseline) ==

[[Driver unlock|cmpunlocker]] sets the 10 GB card to '''40 GB''' (40960 MiB): CFG1 <code>0x009a0204 = 0x02669000</code>,
LMR <code>0x00100ce0 = 0x0000028A</code>. That is 5 HBM stacks of 8 GB. This is {{Status|Proven}} and reproduced by
everyone running the unlock. {{Source|programmerboi|2026-10-25|https://discord.com/channels/1518634379604136178/1534370835220004885/1563704621388529725}}

== 48 GB (the 6th stack, if it is good) ==

48 GB comes from enabling a disabled sixth stack (48 GB = 6 x 8 GB). It needs the [[RIR fuse burn]] (software fuse
remask) and the card must have a working 6th stack, which not all do; estimates put the chance around 20% or less.
{{Source|djpandaman|2026-10-11|https://discord.com/channels/1518634379604136178/1518692084729511967/1558942410724085843}}
{{Source|artistosu|2026-10-18|https://discord.com/channels/1518634379604136178/1518695328008638516/1561898669966364697}}
On most sampled cards the last stack is marked '''defective''', not merely disabled (bigfoot: all 6 of his; orrorin:
both of his), and defective-marked stacks really are defective. A "pristine" card with the 6th stack only disabled
(gpcprobe <code>OPT_FBP_DISABLE 0x840</code>, all DEFECTIVE masks 0) is the one that can reach 48 GB.
{{Source|bigfoot1144|2026-10-19|https://discord.com/channels/1518634379604136178/1518695328008638516/1561796223017681007}}

== HBM vendor ==

{{Status|Reported}} The 10 GB card carries '''Samsung HBM2e'''; the 8 GB card carries '''SK hynix'''. Checked by Ivan
on his own cards. {{Source|darkmaniac7|2026-10-30|note=lab notes}}

Earlier Discord posts on the question, kept for reference:
* RH-3D: 10 GB = Samsung HBM2, 8 GB = Hynix HBM2e 16 GB stacks.
{{Source|rh3dai|2026-10-17|https://discord.com/channels/1518634379604136178/1518692084729511967/1561039191343632425}}
* atsasin: the 10 GB card never had HBM2e, only the 8 GB.
{{Source|atsasin2126|2026-10-16|https://discord.com/channels/1518634379604136178/1518692084729511967/1560781758667169864}}
* mrpandagod: the evidence says the stacks are binned 16 GB HBM2e.
{{Source|djpandaman|2026-10-17|https://discord.com/channels/1518634379604136178/1518692084729511967/1560892095739469847}}
* Mark Linton's read-only dumps of two 8 GB cards show the SK Hynix strap <code>0x004906D1</code>.
{{Source|marklintonca|2026-10-10|https://discord.com/channels/1518634379604136178/1518695328008638516/1558556618818457612}}

== 80 / 96 GB: not real on these dies ==

{{Status|contested}} Several people got a 10 GB card to '''report''' 80 GB, but the memory was unstable above 40 GB.
{{Source|tekkenz|2026-10-12|https://discord.com/channels/1518634379604136178/1534370835220004885/1559048743947341955}}
Humvee55, a co-author of the paper that used a 10 GB card as the test vehicle, now believes the apparent extra capacity
is '''aliasing in L2''', not real memory. {{Source|scallion_9883|2026-10-16|https://discord.com/channels/1518634379604136178/1518692084729511967/1560782124049760389}}
A read-only fuse RE of a Samsung 10 GB card found the 80 GiB aperture (LMR <code>0x28B</code>) '''folds''' at about
16 KB above the expected address (address folding confirmed die-side), i.e. the upper half is aliased, not addressable.
{{Source|ahnguyen17|2026-10-03|https://github.com/ahnguyen17/ga100-fuse-re}}
Stacks are physically 16 GB parts that Samsung binned; unlocking the upper half is "basically impossible" with the
information available, likely needing internal Samsung codes. {{Source|djpandaman|2026-10-11|https://discord.com/channels/1518634379604136178/1518692084729511967/1559096192036642817}}

A Chinese write-up "170Hx 10G 解锁80G探索" (10G unlock to 80G exploration) is at blog.kkk.rs/archives/71; no one reported a
stable 80 GB card from it. A screenshot claiming 130104 MB was debunked as two 64 GB-unlocked 8 GB cards reported
together (65024 MiB x 2). {{Source|kio063218|2026-10-15|https://discord.com/channels/1518634379604136178/1534858283138158697/1560165388883468288}}

== How the state is reported ==

Community agreement as of 2026-10-23: a 10 GB card now equals "an A100 40G PCIe with more SM"; after the RIR tool went
open source, except for the 40 to 80/96 GB VRAM-initialisation issue, most unlock problems are solved. NVLink is still
not public. {{Source|ob88594|2026-10-23|https://discord.com/channels/1518634379604136178/1518695328008638516/1563057728887328779}}

== Known issues ==

* A dead region that returns 0 and stores nothing can be worked around by pinning a dummy allocation over it; a longer
write-up by @analyticeth exists; see Discord.
* Cards with HBM errors: the driver can mark bad physical blocks at boot (4 x 64 KB blocks cost 8 MB). A longer write-up
by @zombielurker exists; see [[Troubleshooting]] and Discord.

== Status ==

40 GB is {{Status|Proven}}. 48 GB is {{Status|Reported}} and card-dependent. 80/96 GB is {{Status|Contested}} and most
evidence says it is aliasing, not memory.

== Related ==

[[RIR fuse burn]], [[Full SM unlock]], [[Driver unlock]], [[Which card do I have]], [[10 GB card]], [[Troubleshooting]],
[[VBIOS and firmware]].

{{ImportFooter}}
[[Category:Unlocks]]
[[Category:10 GB]]

← Older revision		Revision as of 10:02, 30 October 2026
Line 24:		Line 24:
	remask) and the card must have a working 6th stack, which not all do; estimates put the chance around 20% or less.		remask) and the card must have a working 6th stack, which not all do; estimates put the chance around 20% or less.
	{{Source\|djpandaman\|2026-10-11\|https://discord.com/channels/1518634379604136178/1518692084729511967/1558942410724085843}}		{{Source\|djpandaman\|2026-10-11\|https://discord.com/channels/1518634379604136178/1518692084729511967/1558942410724085843}}
	{{Source\|artistosu\|2026-10-18\|https://discord.com/channels/1518634379604136178/1518695328008638516/~~1561898669966364697~~}}		{{Source\|artistosu\|2026-10-26\|https://discord.com/channels/1518634379604136178/1518695328008638516/1564127408305479803}}
	On most sampled cards the last stack is marked '''defective''', not merely disabled (bigfoot: all 6 of his; orrorin:		On most sampled cards the last stack is marked '''defective''', not merely disabled (bigfoot: all 6 of his; orrorin:
	both of his), and defective-marked stacks really are defective. A "pristine" card with the 6th stack only disabled		both of his), and defective-marked stacks really are defective. A "pristine" card with the 6th stack only disabled
	(gpcprobe <code>OPT_FBP_DISABLE 0x840</code>, all DEFECTIVE masks 0) is the one that can reach 48 GB.		(gpcprobe <code>OPT_FBP_DISABLE 0x840</code>, all DEFECTIVE masks 0) is the one that can reach 48 GB.
	{{Source\|bigfoot1144\|2026-10-19\|https://discord.com/channels/1518634379604136178/1518695328008638516/1561796223017681007}}		{{Source\|bigfoot1144\|2026-10-19\|https://discord.com/channels/1518634379604136178/1518695328008638516/1561796223017681007}}<br />
			{{Source\|handle=_orrorin\|date=2026-10-20\|link=https://discord.com/channels/1518634379604136178/1518695328008638516/1562165405924069639}}

	== HBM vendor ==		== HBM vendor ==
Line 50:		Line 51:
	{{Source\|tekkenz\|2026-10-12\|https://discord.com/channels/1518634379604136178/1534370835220004885/1559048743947341955}}		{{Source\|tekkenz\|2026-10-12\|https://discord.com/channels/1518634379604136178/1534370835220004885/1559048743947341955}}
	Humvee55, a co-author of the paper that used a 10 GB card as the test vehicle, now believes the apparent extra capacity		Humvee55, a co-author of the paper that used a 10 GB card as the test vehicle, now believes the apparent extra capacity
	is '''aliasing in L2''', not real memory. {{Source\|scallion_9883\|2026-10-16\|https://discord.com/channels/1518634379604136178/1518692084729511967/~~1560782124049760389~~}}		is '''aliasing in L2''', not real memory. {{Source\|handle=scallion_9883\|date=2026-10-16\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560782230777888841}}<br />
			{{Source\|handle=scallion_9883\|date=2026-10-16\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1560782273652195479}}
	A read-only fuse RE of a Samsung 10 GB card found the 80 GiB aperture (LMR <code>0x28B</code>) '''folds''' at about		A read-only fuse RE of a Samsung 10 GB card found the 80 GiB aperture (LMR <code>0x28B</code>) '''folds''' at about
	16 KB above the expected address (address folding confirmed die-side), i.e. the upper half is aliased, not addressable.		16 KB above the expected address (address folding confirmed die-side), i.e. the upper half is aliased, not addressable.
	{{Source\|ahnguyen17\|2026-10-03\|https://github.com/ahnguyen17/ga100-fuse-re}}		{{Source\|ahnguyen17\|2026-10-03\|https://github.com/ahnguyen17/ga100-fuse-re}}
	Stacks are physically 16 GB parts that Samsung binned; unlocking the upper half is "basically impossible" with the		Stacks are physically 16 GB parts that Samsung binned; unlocking the upper half is "basically impossible" with the
	information available, likely needing internal Samsung codes. {{Source\|djpandaman\|2026-10-11\|https://discord.com/channels/1518634379604136178/1518692084729511967/1559096192036642817}}		information available, likely needing internal Samsung codes. {{Source\|handle=djpandaman\|date=2026-10-12\|link=https://discord.com/channels/1518634379604136178/1518692084729511967/1559096192036642817}}
			On 2026-10-20 Mountains put the current memory maximum for a 10 GB card at 48 GB, and bigfoot saw "no indication that
			80 or 96 gb is achievable", with 48 GB possible "if your last 2 stacks are lucky".
			{{Source\|handle=mountains1605\|date=2026-10-20\|link=https://discord.com/channels/1518634379604136178/1518695328008638516/1561898669966364697}}<br />
			{{Source\|handle=bigfoot1144\|date=2026-10-20\|link=https://discord.com/channels/1518634379604136178/1518695328008638516/1562173508094533753}}

	A Chinese write-up "170Hx 10G 解锁80G探索" (10G unlock to 80G exploration) is at blog.kkk.rs/archives/71; no one reported a		A Chinese write-up "170Hx 10G 解锁80G探索" (10G unlock to 80G exploration) is at blog.kkk.rs/archives/71; no one reported a
	stable 80 GB card from it. A screenshot claiming 130104 MB was debunked as two 64 GB-unlocked 8 GB cards reported		stable 80 GB card from it. A screenshot claiming 130104 MB was debunked as two 64 GB-unlocked 8 GB cards reported
	together (65024 MiB x 2). {{Source\|kio063218\|2026-10-15\|https://discord.com/channels/1518634379604136178/1534858283138158697/~~1560165388883468288~~}}		together (65024 MiB x 2). {{Source\|handle=kio063218\|date=2026-10-16\|link=https://discord.com/channels/1518634379604136178/1534858283138158697/1560618333542293526}}

	== How the state is reported ==		== How the state is reported ==