Failed in 2D Graphics test and shows video memory corruption

[GI Wan]

Hi ,

When we run burn-in test 6.0 build 1014 , if we just select 2D test only, no error happens and if we select multiple test items with 2D test, we will find 5% 2D error "video memory corruption". Can you provide some advice about how to fix this error?

Failure Rate : 5% (5/100)

Configuration :
CPU Type: Intel(R) Core(TM) CPU Q 720 @ 1.60GHz
MEMORY : Total Physical Memory: 3005MB
Available Physical Memory: 2496MB
Virtual memory: C:\pagefile.sys (allocated base size 3304MB)
GRAPHICS
NVIDIA NVS 5100M
DISK VOLUMES
C: Local drive, NTFS, (29.30GB total, 11.92GB free)
D: Local drive, KAQDOS, FAT32, (203.54GB total, 203.50GB free)
DISK DRIVES
Disk drive: Model ST9250410AS ATA Device (Size: 232.88GB)
Disk drive: Model ST9250410AS ATA Device (Size: 232.88GB)
OPTICAL DRIVES
NETWORK
Intel(R) 82577LM Gigabit Network Connection
Bluetooth Device (Personal Area Network)
PORTS
Parallel port: LPT1
Intel(R) 5 Series/3400 Series Chipset Family USB Enhanced Host Controller - 3B3C
Intel(R) 5 Series/3400 Series Chipset Family USB Enhanced Host Controller - 3B34
NEC Electronics USB 3.0 Host Controller

**************
RESULT SUMMARY
**************
Test Start time: Wed Oct 21 20:49:04 2009
Test Stop time: Wed Oct 21 21:04:11 2009
Test Duration: 000h 15m 07s
Test Name Cycles Operations Result Errors Last Error
CPU 147 1.247 Trillion PASS 0 No errors
Memory (RAM) 1 7.349 Billion PASS 0 No errors
2D Graphics 5 5169 FAIL 3 Video memory corruption
3D Graphics 45 91259 PASS 0 No errors
TEST RUN FAILED

******************
DETAILED EVENT LOG
******************
LOG NOTE: 2009-10-21 20:49:04, Status, Main Tests started
LOG NOTE: 2009-10-21 20:49:04, CPU, 3DNow! SSE4a test(s) selected, but CPU feature not detected
LOG NOTE: 2009-10-21 20:49:05, 3D Graphics, 3D test starting: Duty cycle = 50. Test on primary monitor (0). (x,y) = (310,0). WxH = 300x200.
SERIOUS: 2009-10-21 20:59:28, 2D Graphics, Video memory corruption
LOG NOTE: 2009-10-21 20:59:28, 2D Graphics, 32 Bit ==> Address 0x486B21F0: Expected 0xaaaaaaaa: Read 0xaaaa80
SERIOUS: 2009-10-21 21:02:25, 2D Graphics, Video memory corruption
LOG NOTE: 2009-10-21 21:02:25, 2D Graphics, 32 Bit ==> Address 0x4883C998: Expected 0xff: Read 0x0
SERIOUS: 2009-10-21 21:03:40, 2D Graphics, Video memory corruption
LOG NOTE: 2009-10-21 21:03:40, 2D Graphics, 32 Bit ==> Address 0x46EAE7E4: Expected 0xaaaaaaaa: Read 0xaaaa00
LOG NOTE: 2009-10-21 21:04:16, Status, Test run stopped

[Ian (PassMark)]

The 2D Video memory test basically creates a number of 2D surface textures, writes these textures to video memory via DirectX DirectDraw, then reads the surfaces back and verifies that the test data written matches the test data read. A video memory corruption error means that the data written to video memory does not match the data read.

This problem can be caused by faulty video RAM, and video RAM settings (such as video RAM frequency and voltages) or a driver issue bug.

We did some work on a 2D Video memory test “video memory corruption” error earlier this year with integrated Intel graphics chipsets. Eventually Intel agreed it was a driver problem. They made a change to their driver, unfortunately our customers that reported the problem advised us that it only changed the behavior of the problem. Instead of getting a video memory corruption error, with the new Intel driver the 2D graphics tests hangs (at what looks to be the same problem), you get a No operations error.

If you only get the problem when running other tests with the2D test, then your problem only occurs when there is some other load on the system, this could be related to a number of issues: Such as memory usage, heat, current drawn from the power supply....

We do recommend that the 2D video memory test be set to test the local video RAM and then use the Memory test to test system RAM. That is, we recommend that you change the 2D Video test configuration from “All available video memory” to “Local video memory only”.

Edit: Updated with other instances:
ATI investigated a similar issue for Acer around mid 2004, we aren't sure about the detail but believe it was corrected with a video card driver update.

NVIDIA also looked a this issue in 2005 for some of their cards. Their feedback was that the cause was, "P4 cache coherency rules (and ordering rules) with regards to Write Combining".

AMD found an issue in 2019. Their feedback was. "Our [video card] command buffers are reusable and there was possibility that the command buffer could be clobbered by CPU while the contents of the chunks were still used by GFX pipeline. We put a wait for pipeline done there and solved the problem". So another instance of the problem being a driver issue.

Low DDR RAM and video card AGP voltages caused a problem in once instance.