GPU_MEMTEST_ERROR for GTX 660 Ti

E

Elledan

[H]ard|DCer of the Month - April 2010
Joined
Oct 18, 2001
Messages
15,913
With the latest v7 client on Windows 7 Ultimate x64 and latest NVidia drivers the simulation stops with the error GPU_MEMTEST_ERROR. This even after following this guide: http://www.evga.com/forums/tm.aspx?&m=1526339&mpage=1

Things tried so far:
- Set the client-type to 'beta'.
- Updated GPU.txt to the version on here: http://folding.typepad.com/news/201...gpus-such-as-kepler-in-the-v7-fah-client.html
- Reinstalled GPU drivers (clean install)
- tested GPU RAM with MemtestG80 (no errors)

I can see that a 'beta' folder is created under 'cores' in the data folder and that it fetches an a15 core from the fermi/beta folder on the server.

Any ideas? So far it seems like F@H on the GTX6xx cards is impossible for everyone but a lucky few for whom the planets aligned correctly. It feels like a shame to let these new (3 GTX660 Ti) cards go to waste for a few months while the Pande team gets its stuff together for these new GPUs...
 
Had FAH v7.2.9 working fine with my roomies' GTX670... got 8057 units too. Did about 109k-120k ppd when I got those units (stock clocked evga card). Didn't do anything fancy driver wise... those were installed back during like July.
 
Log file is pretty basic: it starts the core, returns 'mdrun returns 59' after a while and the GPU_MEMTEST_ERROR.

What is the right core it should be using for Kepler anyway?
 
Did you test the maximum quantity of memory of your gtx660ti with memtestG80 ?

For exemple , with my gtx660 , i did some tests of 1250 mb with this command line :

MemtestG80 1250 100

Instructions of MemtestG80 :

" By default, MemtestG80 will test 128 megabytes of memory on your first video
card, running 50 iterations of its tests. On typical machines, each iteration
will complete in under 10 seconds with these parameters (the speed will vary
both with the speed of the card tested and the amount of memory tested). The
amount of memory tested and number of test iterations can be modified by adding
command line parameters as follows:

MemtestG80 [amount of RAM in megabytes] [number of test iterations]

For example, to run MemtestG80 over 256 megabytes of RAM, with 100 test
iterations, execute the following command:

MemtestG80 256 100 "
 
@Jeanjean I tested the first 128 MB with MemtestG80.

@Xavier Zepherious I checked, but FAHControl was not installed as a service.
 
Elledan,

Check your core version - I've had it on a 570 with core 2.25, but not with 2.22.

H.
 
Overclocked? How big is your psu? I had a client that was doing this...more volts.
 
i tried boosting my 660ti without boosting the voltage and it immediately dumped the wu (at 80%0. Then it died trying to start 2 more, (one was a 8057). Finally figured out what to do. My 460 was more forgiving and a lot less sensitive.
 
With my GTX 660 , i had to reduce the factory memory frequency by 100 mhz in order to be full stable and although memtestG80 did not detect any error .

fortunately , i did not have such problem with the GPU frequency .
 
elledan - if that latest core is 2.25, that's the one that was causing me problems. Going back to the 2.22 fixed the problem for the WU I was on, but some WU's will re-download the newer core. I've put together a little batch script that uses the GPU console client, that reverts back to the 2.22 core after each WU.

H.
 
Elledan said:
@sc0tty8 - It's a 550 Watt Gigabyte PSU. Not a stellar unit, but serviceable. It's coupled with a Q6600, 6 GB RAM and two HDDs.

It's also a factory overclock, so I assume it's stable. It's this unit: http://www.techpowerup.com/reviews/MSI/GTX_660_Ti_Power_Edition/

@Haitch - it's using the latest beta core, I assume, from beta/fermi/.
Click to expand...

Haitch said:
elledan - if that latest core is 2.25, that's the one that was causing me problems. Going back to the 2.22 fixed the problem for the WU I was on, but some WU's will re-download the newer core. I've put together a little batch script that uses the GPU console client, that reverts back to the 2.22 core after each WU.

H.
Click to expand...

Fermi won't fold on 2.22.
 
Elledan said:
@sc0tty8 - It's a 550 Watt Gigabyte PSU. Not a stellar unit, but serviceable. It's coupled with a Q6600, 6 GB RAM and two HDDs.

It's also a factory overclock, so I assume it's stable. It's this unit: http://www.techpowerup.com/reviews/MSI/GTX_660_Ti_Power_Edition/

@Haitch - it's using the latest beta core, I assume, from beta/fermi/.
Click to expand...

Sorry for prev post iPad is a pos...

Try another psu, I had a box with a pair of 560's that would toss the Gould memtest error, new psu and it's happy....
 
sc0tty,

Fermi does fold with 2.22, and for better PPD. It's the Keplers that won't fold on 2.22, and only now just noticed lledan is running 6xx not 5xx - so my suggestion is moot.

H.
 
Haitch said:
sc0tty,

Fermi does fold with 2.22, and for better PPD. It's the Keplers that won't fold on 2.22, and only now just noticed lledan is running 6xx not 5xx - so my suggestion is moot.

H.
Click to expand...

Whoops, that is what I ment :blushes:
 
I wonder if a WU that couldn't run on Keplar was downloaded before all the settings were correct. Might have needed to delete it to get things working properly. Posting the log may have helped.
 
@ChelseaOilman - I did delete all the GPU WUs and cores as per the instructions in the thread I gave earlier, so it should have started off clean every time.
 
MemtestCL returned zero errors after 50 iterations. The virus scanner present was removed.

I also installed a 660 Ti into my main system and that one worked without any issues for F@H. They're both very comparable systems, with both a Q6600 and Win7 x64 Ultimate . Only difference in hardware are the PSUs, mainboards and the presence of also a 9500 GT card in the other system.
 
as noted earlier mixed cards will cause this issue
the EVGA post I mentioned earlier had this issue - removing the older card fixed it

if you plan mixed system then you have to do manual installs of the GPU slots in V7 to make sure that v7 doesn't mix them up.

Im sure your team has a guide on mixed card installs just like EVGA
 
Thanks. I did manual installs of both GPU slots (0 - 9500, 1 660 Ti), but to no avail. I'd be very interested in a good guide on mixed card installs.
 
I had similar problem with mixed install of 470 and 8800gt. you have to manually set the GPU-index, Opencl-index, and Cuda-index when adding the slot. I am sure someone else has a better explanation on how, but I just fiddled with setting one card to 0 and other other to 1 till they both eventually worked.

EDIT: if you go into the system info tab, it will tell you which card the v7 client thinks is GPU0 and GPU1
 
Ah yes but what the client thinsk are GPU 0 and 1 might not be what Cuda thinks are 0 and 1, download cuda-z and see what its telling you regarding the cards' numerical order, you may find them reversed
 
In the end the issue was indeed the mixed 660 Ti and 9500 GT setup. Upon removing the 9500 GT the 660 Ti started folding right away :)
 
You must log in or register to reply here.
Back
Top