radeon 7970 memtestCL errors

T

tscolin

n00b
Joined
Sep 12, 2005
Messages
13
Good morning gentlemen;

I'm running into issues involving very random hardlocks. I first thought it might be my mobo as when it occurs i cant restart the computer with the reset button (i have to hold the power button or flip the psu off and on). However, tests including long runs of linpack and prime95 as well as long runs of memtest86 have shown the cpu and ram to be fine.

I decided to test the video card using memtestCL (to test vram and board logic), which gave me tons of errors in the random blocks test.

memtestCL: http://folding.stanford.edu/English/DownloadUtils

This is the output ill recieve, its the same number of errors testing 128 megs or 3000.

Running memory bandwidth test over 20 iterations of 384 MB transfers...
Estimated bandwidth 6884.81 MB/s


-----------------------------------------------------------------------------------------------------------
Test iteration 1 on 3000 MiB of memory on device 0 (Tahiti): 0 errors so far
Moving Inversions (ones and zeros): 0 errors (4025 ms)
Moving Inversions (random): 0 errors (1076 ms)
Memtest86 Walking 8-bit: 0 errors (9080 ms)
True Walking zeros (8-bit): 0 errors (4508 ms)
True Walking ones (8-bit): 0 errors (4462 ms)
Memtest86 Walking zeros (32-bit): 0 errors (22526 ms)
Memtest86 Walking ones (32-bit): 0 errors (19594 ms)
Random blocks: 1036993 errors (780 ms)
Memtest86 Modulo-20: 0 errors (201491 ms)
Logic (one iteration): 0 errors (749 ms)
Logic (4 iterations): 0 errors (748 ms)
Logic (local memory, one iteration): 0 errors (593 ms)
Logic (local memory, 4 iterations): 0 errors (609 ms)

Test iteration 2 on 3000 MiB of memory on device 0 (Tahiti): 1036993 errors so far
-----------------------------------------------------------------------------------------------------------

so obviously this has me seriously irked. Could anyone else test their 7970's and see if they also get the same errors to rule out a false positive for me?

I still think my mobo is bad, but perhaps my video card is as well.

Thanks in advance!
 
That ammount of errors points to me to RAM problems.

I ran that test on my 6850, found under 2k errors at max ram overclock (1150MHz), dialed back the oc by 50MHz and found 0 errores on the next runs.

If you have so many errors at base clocks i would be wary indeed, as even though DDR5 has error correction, it can only go so far... Good luck!
 
Do you have your GPU overclocked?

Set it to stock and see what happens.

It doesn't surprise me that anything, any component, would malfunction when pushed too far.:D
 
yeah these are stock clocks.

magoo i see you have a 7970, can you try it on yours just to rule out false positives?
 
I ran the test OC'd and stock and I always get 50 random block errors on my asus 7970. Something is not right.:confused:

Just tried it running 700/1100 and get the same number of random block errors. I get no artifacting or freezing in any game or benchmark.
 
Last edited:
I have the Gigabyte reference 7970, and I am also getting the 50 random blocks error at stock clocks for both vram and gpu.
 
Stock clocks:
Random blocks: 50 failed iterations
(17995419 total incorrect bits)

1200/1600:
Random blocks: 50 failed iterations
(17331624 total incorrect bits)

I think something is wrong with this test...

and random blocks is the only error.
 
i tested only 128mb...

Test summary:
-----------------------------------------
50 iterations over 128 MiB of memory on device Tahiti
Moving inversions (ones and zeros): 0 failed iterations
(0 total incorrect bits)
Memtest86 walking 8-bit: 0 failed iterations
(0 total incorrect bits)
True walking zeros (8-bit): 0 failed iterations
(0 total incorrect bits)
True walking ones (8-bit): 0 failed iterations
(0 total incorrect bits)
Moving inversions (random): 0 failed iterations
(0 total incorrect bits)
True walking zeros (32-bit): 0 failed iterations
(0 total incorrect bits)
True walking ones (32-bit): 0 failed iterations
(0 total incorrect bits)
Random blocks: 50 failed iterations
(17331624 total incorrect bits)
Memtest86 Modulo-20: 0 failed iterations
(0 total incorrect bits)
Integer logic: 0 failed iterations
(0 total incorrect bits)
Integer logic (4 loops): 0 failed iterations
(0 total incorrect bits)
Integer logic (local memory): 0 failed iterations
(0 total incorrect bits)
Integer logic (4 loops, local memory): 0 failed iterations
(0 total incorrect bits)
Final error count: 50 test iterations with at least one error; 17331624 errors total
 
just as a double check/false positive i ran it on my 8800GT's got no errors. so it may be something to do with the app on the AMD side. might be worth having some one with a 6970 or another AMD card(other than a 7900 series) to test it to see if its purely related to the 7900 series and the new GCN architecture not being compatible with the current AMD CL code used in the test or all cards.
 
i can confirm all tests pass with 0 errors on my HD5870. So while i get no graphics errors on any test/benchmark with my 7970. it still errors in random blocks. The question i have concerning this issue is: Is it the card that has a basic logic/memory controller issue, or is it memtestCL which runs pretty simple code using openCL, of which AMD claims they fully support. I hope its the ladder.

*edit* id love it if others with 7xxx cards could keep chiming in on this.
 
How do you modify for more memory?? I have a 6950 unlocked I want to test...
 
This app has not been updated since 2010 so all these "errors" are most likely the result of obsolete code incompatible with the new architecture.

Updated code paths are needed for the new architecture. The best example of this is bitcoin mining. At the 7970s launch, there were no miners capable of working with the card's new architecture until someone coded a path for it and since then performance has been pretty stellar.
 
sorry for this noobish question, but.
I just got my first AMD card for ages(7970), and downloaded the latest driver when i installed it, also the "amd APP SDK"
why is there still no openCL support, what am i missing?
 
I found the problem. the driver package did not install it, but there is an installer for it in the unpacked folder, once i ran that everything is good.

and yes, 50 "random blocks" errors.
 
Last edited:
IM also getting random block errors....

Test summary:
-----------------------------------------
50 iterations over 128 MiB of memory on device Tahiti
Moving inversions (ones and zeros): 0 failed iterations
(0 total incorrect bits)
Memtest86 walking 8-bit: 0 failed iterations
(0 total incorrect bits)
True walking zeros (8-bit): 0 failed iterations
(0 total incorrect bits)
True walking ones (8-bit): 0 failed iterations
(0 total incorrect bits)
Moving inversions (random): 0 failed iterations
(0 total incorrect bits)
True walking zeros (32-bit): 0 failed iterations
(0 total incorrect bits)
True walking ones (32-bit): 0 failed iterations
(0 total incorrect bits)
Random blocks: 50 failed iterations
(17160445 total incorrect bits) Memtest86 Modulo-20: 0 failed iterations
(0 total incorrect bits)
Integer logic: 0 failed iterations
(0 total incorrect bits)
Integer logic (4 loops): 0 failed iterations
(0 total incorrect bits)
Integer logic (local memory): 0 failed iterations
(0 total incorrect bits)
Integer logic (4 loops, local memory): 0 failed iterations
(0 total incorrect bits)
Final error count: 50 test iterations with at least one error; 17160445 errors t
otal
 
Lorien said:
This app has not been updated since 2010 so all these "errors" are most likely the result of obsolete code incompatible with the new architecture.

Updated code paths are needed for the new architecture. The best example of this is bitcoin mining. At the 7970s launch, there were no miners capable of working with the card's new architecture until someone coded a path for it and since then performance has been pretty stellar.
Click to expand...

pretty much.. the memCL test is actually designed for testing the core16 F@H openCL code which is pretty mediocre patch job coding to begin with, so it doesn't surprise me. but the fact that its just random block errors and so far only seen with the 7900 series cards its probably the program. plus its to consistent to be an actual problem with the cards.

and yeah its the same way with F@H, the 7900 series works but it runs like crap because the openCL core has no clue what to do with the GCN architecture.


tscolin said:
sure would be nice if this app was maintained, such a useful lil app.
Click to expand...

the app was never really meant to be used long term. it was created to test PG's openCL client code for F@H. but i agree, it would be nice to see PG release the code so some one else can take over maintaining it since i'm sure some one would be willing to.
 
Bump. I also am getting the 50 errors. Two 7970's did it. One was really bad, this one may not be though.


This should probably be a sticky in the main video card forum.
 
Why is anyone even using this? There are far better ways to check stability.
 
CleanSlate said:
Don't just say that, list them out so people will follow your advice
Click to expand...

Are you using this to test video card stability? Unigine Heaven 3.0 is probably the quickest way to check stability. Let it loop at the highest detail settings for 30 minutes, if it errors out, its not stable. If it keeps running, its probably stable. You can also use Metro 2033 benchmark if you own the game, I run about 30 loops of it also at the highest settings. From there, you can play some Crysis 2 and Battlefield 3 with the highest settings for 30-60 minutes and see if it errors. If not, it will most certainly be stable at anything else.
 
On my 7950 I'm getting the same "Random blocks: 50 failed iterations" ( 128 MiB ).

No matter if stock ( 1250Mhz ) or overclocked ( 1575 ) Memory.
 
NiColaoS said:
On my 7950 I'm getting the same "Random blocks: 50 failed iterations" ( 128 MiB ).

No matter if stock ( 1250Mhz ) or overclocked ( 1575 ) Memory.
Click to expand...

That should tell you right there that test is not valid then.
 
You must log in or register to reply here.
Back
Top