The boinc client detects the GPU as 'Caicos' series with 2048MB VRAM. If you look at the third entry on this page, there is such a GPU with the 2048MB size as one of the available options. On a quick look through those massive tables of GPU variants, that was the only entry I saw that showed a 2048MB option. I didn't do an exhaustive search so could have missed others.
That GPU was from a very old architecture (TeraScale 2) and first released in early 2011. It doesn't have hardware FP64 so fiddling with drivers isn't going to make a difference. The 'error' message about FP64 is just the Einstein app trying to detect that capability. It does this by trying to compile a little test routine which just fails if there is no FP64. That requirement can be provided by the CPU so it isn't a show stopper - it just makes the followup stage rather lengthy when it has to be done on the CPU.
The real error message is right near the end of the output on the example I looked at:-
Error in computing index of fft input array, i:-2139062144 pair:2605055
fft is Fast Fourier Transform and this involves setting up huge arrays of data in memory. My guess is that your hardware has some sort of transient problem with handling these huge arrays. Your tasks list shows just a couple of successful completions and around 40 failures. The successful tasks took around 8 hours. This is crazy. If you picked up a slightly less ancient GPU from the HD7000 series - eg HD7750 or 7770 - it should cost peanuts and give you crunch times around the 1 hour mark. You would just need to confirm space in your case and power needs.
All the errored tasks are
)
All the errored tasks are complaining of no FP64 support in the driver when compiling the tasks.
I would reload your drivers.
The boinc client detects the
)
The boinc client detects the GPU as 'Caicos' series with 2048MB VRAM. If you look at the third entry on this page, there is such a GPU with the 2048MB size as one of the available options. On a quick look through those massive tables of GPU variants, that was the only entry I saw that showed a 2048MB option. I didn't do an exhaustive search so could have missed others.
That GPU was from a very old architecture (TeraScale 2) and first released in early 2011. It doesn't have hardware FP64 so fiddling with drivers isn't going to make a difference. The 'error' message about FP64 is just the Einstein app trying to detect that capability. It does this by trying to compile a little test routine which just fails if there is no FP64. That requirement can be provided by the CPU so it isn't a show stopper - it just makes the followup stage rather lengthy when it has to be done on the CPU.
The real error message is right near the end of the output on the example I looked at:-
Error in computing index of fft input array, i:-2139062144 pair:2605055
fft is Fast Fourier Transform and this involves setting up huge arrays of data in memory. My guess is that your hardware has some sort of transient problem with handling these huge arrays. Your tasks list shows just a couple of successful completions and around 40 failures. The successful tasks took around 8 hours. This is crazy. If you picked up a slightly less ancient GPU from the HD7000 series - eg HD7750 or 7770 - it should cost peanuts and give you crunch times around the 1 hour mark. You would just need to confirm space in your case and power needs.
Cheers,
Gary.