An uncorrectable error led the card to queue a row swap, and that swap has waited for a GPU reset across seven reads and 76 days. The host never reset it, so the faulty row is still in use.
Damage ladder
Bit flips fixednot seen
Row swapped for a sparerecorded
Waiting for a resetrecorded
Data lostrecorded
Bank out of sparesnot seen
Memory banks
192 memory banks · one has used a spare row
Identity
Serial
···0230
Built
Sep 2023
Board part
900-2G133-0080-000
VBIOS
95.02.66.00.02
InfoROM
G133.0242.00.03
Driver
565.57.01
Signed reads
7
First read
Jul 16, 2026
Last read
Sep 30, 2026
Span
76 days
Record
Your GPUs come with a record.
A GPU counts every memory error it has seen and every repair it has made. Voltry reads those counts, signs them,
and files them where anyone can check.
16 signals in one read
Memory errors, repairs, spare rows, heat, power, links. Read straight from the card.
Signed the moment it is read
Change one byte and the signature breaks. Anyone can check it, free.
Filed under the card itself
The record stays with the GPU when it changes hands.
$1 per GB of the GPU's memory. An H100 80 GB is $80.
Buying?
Ask your seller. Copy this line into your request.
Please include a Voltry condition record for each GPU. Records are checked free at voltry.io/verify.
Counters the card keeps about itself, in four groups. The reader only looks, and nothing on the card changes.
Line drawing, not to scale. Pick a group to light the parts it is read from.
Memory7 signals
Counted on the card itself and kept for its whole life, across every owner and reboot. This is where wear shows first.
Lifetime correctable errors
Bit flips the card caught and fixed on the spot.
Lifetime uncorrectable errors
Errors the card could not fix, where data was lost.
Rows repaired after correctable errors
Rows retired for good after repeated fixed errors, each replaced by a spare.
Rows repaired after uncorrectable errors
Rows retired for good after an error that lost data, each replaced by a spare.
Repair pending
A row swap that only takes effect after the GPU is reset.
Repair failure
A swap the card could not make because the bank had no spare rows left.
Spare reserve, bank by bank
How many spare rows each memory bank still holds, as the card reports them.
Between reads3 signalscomes with tracking
Lifetime counts only go up. Reading the same card again shows whether anything moved.
Movement between reads
Any rise in a lifetime count between two signed reads, the clearest sign of wear in progress.
GPU resets between reads
A drop in the card's energy counter between reads marks a reset, which a pending repair needs.
Counts never going down
A lifetime count that drops would mean wiped history, a swapped card or a bad read.
Board and power4 signals
A card can have spotless memory and still be held back by its board or the server around it.
Throttle reasons
Why the card is holding its clocks down at the moment of the read.
Temperature at idle
GPU temperature against the point where the card slows itself.
Power limit
Idle draw and the enforced limit against the factory default, so a capped card shows.
Clock-limited time
Time spent clock-limited since the driver loaded, recorded but not graded.
Links5 signals
PCIe connects the card to the server. NVLink connects GPUs to each other on SXM boards.
PCIe width
Lanes in use against what the card supports.
PCIe generation
Link speed against the card's maximum.
PCIe resent packets
Packets the link had to send again, which rise as a connection degrades.
NVLink links up
GPU to GPU links active out of the total the card has.
NVLink resends and recoveries
Transfers sent again and link resets between GPUs.
NVLink signals apply to cards that have NVLink.
Where wear shows first.
GPU memory holds data as tiny charges in rows of cells, split into banks. Each bank keeps a few spare rows of its own, and the card counts every step as damage builds.
Normal wear and tearCard can no longer heal itself
1 of 8
Scroll, use Next and Back, or the arrow keys.
The stack
A memory stack.
On HBM cards (A100, H100, H200, B200) memory sits in stacks beside the GPU. GDDR6 cards (A10, A40, L4, L40S) use separate chips that work the same way.
The banks
Split into banks.
Memory is divided into banks: 192 on an A10, 640 on an A100, 2,560 on an H100 SXM.
The spare rows
Each bank keeps its own spare rows.
When a row keeps failing, the card retires it and a spare row takes its place. The spares are held back below the dashed line.
Step 1 of 5
Bit flips, fixed.
Lifetime correctable errors
One bit flipped and error-correcting code fixed it on the spot. A few over years is normal; thousands, or a climbing count, mark a weak spot.
Step 2 of 5
A row swapped for a spare.
Rows repaired, after correctable or uncorrectable errors
A row that keeps failing is retired for good and a spare takes its place. Real damage, fixed.
Step 3 of 5
Swap waiting for a reset.
Repair pending
The swap only takes effect after the GPU is reset. Until then the bad row is still in use.
Step 4 of 5
Data lost.
Lifetime uncorrectable errors
Two or more bits flipped together and could not be fixed. Data was lost and the job running usually crashed.
Step 5 of 5
A bank out of spares.
Repair failure, spare reserve by bank
The bank holding the bad spot has no spare rows left, so the card can no longer repair it.
A real record
This is what the buyer sees.
Voltry condition record
NVIDIA L40S45 GB GDDR6
Serious
An uncorrectable error led the card to queue a row swap, and that swap has waited for a GPU reset across seven reads and 76 days. The host never reset it, so the faulty row is still in use.
Damage ladder
Bit flips fixednot seen
Row swapped for a sparerecorded
Waiting for a resetrecorded
Data lostrecorded
Bank out of sparesnot seen
Memory banks
192 memory banks · one has used a spare row
Readouts at the last read
lifetime correctableno bit flips on record
0
lifetime uncorrectabledata the memory could not fix
2
rows repairedafter correctable · after uncorrectable
0 · 1
repair waiting for a resetthe bad row is still in use
yes
temperature at idleread while the card had nothing to do
36°C of 95°C
power limitfactory default
350 W of 350 W
PCIe linkidle links step down to save power
x16 · gen 1
Between reads
7 signed reads between Jul 16 and Sep 30.
Jul 16Sep 30
repair waiting for a reset on every read · 76 days
no GPU reset seen between reads · driver 565.57.01
Read log
read (UTC)
correctable
uncorrectable
rows repaired
repair pending
reset before
Jul 16 17:41
0
2
1
yes
no
Jul 21 08:20
0
2
1
yes
no
Jul 24 12:19
0
2
1
yes
no
Jul 26 20:20
0
2
1
yes
no
Aug 17 04:19
0
2
1
yes
no
Sep 26 08:32
0
2
1
yes
no
Sep 30 08:41
0
2
1
yes
no
Identity
Serial
···0230
Built
Sep 2023
Board part
900-2G133-0080-000
VBIOS
95.02.66.00.02
InfoROM
G133.0242.00.03
Driver
565.57.01
Signed reads
7
First read
Jul 16, 2026
Last read
Sep 30, 2026
Span
76 days
Reads are taken at rest and cover condition, not speed under load. Provider withheld.
Who relies on it
One record, four people who stop guessing.
The buyer sees the card's own history before paying, not the seller's word for it.
The lender holds a signed record of the collateral, checkable without calling anyone.
The insurer prices from the card's condition on the day cover starts.
Warranty and returns settle on facts: what the card had recorded before it shipped.