Win 7, Video crashes in games

[charlie]

This is getting interesting!
Since Vista was first released, and continuing into Win 7, I've
experienced in game video crashing with various hardware combinations,
Including Nvidia and ATI video cards and chipsets. Common games having
the problem were/are Crysis and Far Cry 2.
Various driver and windows updates drastically reduced the number of
crashes, and extended play times before a crash occurred.
Common symptoms are crashed game application, with win 7 and vista.
Other cpu cores and windows are not usually crashed, although in earler
versions of the drivers,etc. the system memory was corrupted, so windows
might behave strangely, forcing a reboot to restore proper operation.
A contributing factor has to do with save game functions. In Far Cry 2,
a saved game that was saved during the session might be lost, unless the
game was saved, Far Cry was completely exited to windows, and restarted.
Perhaps the normal delayed write to the HDs comes into play.
When error processing is implemented at the default install settings,
error messages related to the video card or driver may be logged, and
usually the failed to respond type error is logged, along with a video
driver module.

What was changed (win 7) to improve things - -
Win 7 DX updates, various win 7 updates, some of which (recent) were for
"compatibility" with earlier win versions.
The video driver updates seem to have solved some of the problems,
judging from the increased play times before crashing occurs, and in
some cases, providing a more controlled exit to the desk top, instead of
to a black screen.

Due to the way windows works, the chain of various modules involved
makes it difficult to pin the problems down to specific ones. A possible
simple example might be that a driver or other module in the chain
passed an unexpected value or result to another module. It in turn may
have accepted the value, processed it in some way and passed the result
on. Down the chain, another module received an unexpected input, and
caused the error to occur.

One of the more frustrating details seems to be that "golden" video
cards may not exhibit the problems that "production" cards have.
And occasionally, just the other way around!

I may not have any hair left by the time someone finds the gory details,
and solutions for them!

Rant for the week, sorry!

 

[Paul]

This is getting interesting!
Since Vista was first released, and continuing into Win 7, I've
experienced in game video crashing with various hardware combinations,
Including Nvidia and ATI video cards and chipsets. Common games having
the problem were/are Crysis and Far Cry 2.
Various driver and windows updates drastically reduced the number of
crashes, and extended play times before a crash occurred.
Common symptoms are crashed game application, with win 7 and vista.
Other cpu cores and windows are not usually crashed, although in earler
versions of the drivers,etc. the system memory was corrupted, so windows
might behave strangely, forcing a reboot to restore proper operation.
A contributing factor has to do with save game functions. In Far Cry 2,
a saved game that was saved during the session might be lost, unless the
game was saved, Far Cry was completely exited to windows, and restarted.
Perhaps the normal delayed write to the HDs comes into play.
When error processing is implemented at the default install settings,
error messages related to the video card or driver may be logged, and
usually the failed to respond type error is logged, along with a video
driver module.

What was changed (win 7) to improve things - -
Win 7 DX updates, various win 7 updates, some of which (recent) were for
"compatibility" with earlier win versions.
The video driver updates seem to have solved some of the problems,
judging from the increased play times before crashing occurs, and in
some cases, providing a more controlled exit to the desk top, instead of
to a black screen.

Due to the way windows works, the chain of various modules involved
makes it difficult to pin the problems down to specific ones. A possible
simple example might be that a driver or other module in the chain
passed an unexpected value or result to another module. It in turn may
have accepted the value, processed it in some way and passed the result
on. Down the chain, another module received an unexpected input, and
caused the error to occur.

One of the more frustrating details seems to be that "golden" video
cards may not exhibit the problems that "production" cards have.
And occasionally, just the other way around!

I may not have any hair left by the time someone finds the gory details,
and solutions for them!

Rant for the week, sorry!

Test your system memory.

A basic memory test is memtest86+ from memtest.org . It is good at
static faults (stuck-at faults, where a bit in RAM can only have
one value and not the other).

Transient faults are harder to find, and for that, you can use
the Prime95 stress tester.

http://www.mersenne.org/freesoft

When you run the program (it runs in Windows or Linux), it will ask
you whether you want to "Join GIMPS". Say "No, Just Testing". The program
runs a test thread per CPU core. The program should run for four hours,
without an error being detected, or a thread stopping and "turning red".
If the program stops quite quickly, your system could have transient
memory errors. Prime95 is testing CPU-Northbridge-Memory, and any one
of those could have a problem.

I rate memory, according to how many games exhibit unexplainable
behavior. What I've found is, an improvement in my two DDR2 based
systems, when compared to my two DDR systems (with PC3200 in them).
While the PC3200 system is clean, according to the above testing
methods, in terms of "unexplained game behavior events per month",
the DDR2 systems are outperforming the DDR systems. The implication
is, that game testing is even more stressful than Prime95. And over
a long baseline (many hours of gaming over a period of a month),
is able to tell you a bit more about your computer. I'm not patient
enough, to test with Prime95 for an equivalent number of hours. Four
hours is about all I can stand.

Paul

 

[Char Jackson]

Test your system memory.

A basic memory test is memtest86+ from memtest.org . It is good at
static faults (stuck-at faults, where a bit in RAM can only have
one value and not the other).

Transient faults are harder to find, and for that, you can use
the Prime95 stress tester.

http://www.mersenne.org/freesoft

When you run the program (it runs in Windows or Linux), it will ask
you whether you want to "Join GIMPS". Say "No, Just Testing". The program
runs a test thread per CPU core. The program should run for four hours,
without an error being detected, or a thread stopping and "turning red".
If the program stops quite quickly, your system could have transient
memory errors. Prime95 is testing CPU-Northbridge-Memory, and any one
of those could have a problem.

I rate memory, according to how many games exhibit unexplainable
behavior. What I've found is, an improvement in my two DDR2 based
systems, when compared to my two DDR systems (with PC3200 in them).
While the PC3200 system is clean, according to the above testing
methods, in terms of "unexplained game behavior events per month",
the DDR2 systems are outperforming the DDR systems. The implication
is, that game testing is even more stressful than Prime95. And over
a long baseline (many hours of gaming over a period of a month),
is able to tell you a bit more about your computer. I'm not patient
enough, to test with Prime95 for an equivalent number of hours. Four
hours is about all I can stand.

I agree with Paul's analysis, especially the part about some games
being more of a stress test than a stress tester is, and would also
keep an eye on CPU temps during your testing. Some CPU's simply shut
down when they get too hot, while others do 'funny things' before
going ballistic.

 

[charlie]

I agree with Paul's analysis, especially the part about some games
being more of a stress test than a stress tester is, and would also
keep an eye on CPU temps during your testing. Some CPU's simply shut
down when they get too hot, while others do 'funny things' before
going ballistic.

Memory testing and complete system testing with Burn In Pro Passed with
flying colors. Temperature sensors showed that all temps were well
within reason. A caveat is that localized heating inside a GPU may not
show up on the GPUs temp sensor.

The Video cards each have 1G of DDR5 memory. When running in crossfire,
only one of the two cards video memory is (so I'm told) utilized in
crossfire. I tried swapping cards, to see if that made any difference.
The system has 4G of DDR2 RAM in two sockets (two channel), running at
1T, and 400Mhz (800Mhz effective, conservative), as it's rated to 533Mhz
at 2T)
5-5-5-15-24
Both cards have passed every video card stress test that I've thrown at
them.

The system has passed every test sequence, including such things as
Mem86, and even specific video card memory testing.

Recent (yesterday, today) win 7 updates have also improved things,
based upon extended game play times in the wee hours.
How much is yet to be determined.