You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
The BOINC client decodes the 'Peak Flops' value for NVidia GPUs according to the architecture used in each succeeding card generation. The calculation is done in COPROC_NVIDIA::set_peak_flops() at
The architecture is assessed from the Compute Capability, and we currently have values for up to CC 5.
We didn't add any extra handlers for the CC 6.1 reported by the GTX 10x0 range.
Today (20 Sep 2018) was supposed to be the launch day for the new RTX 2080 and RTX 2080 Ti cards. I have seen a screen shot implying that they have CC 7 (or possibly 7.5). I'm putting out an urgent call for the cores_per_proc value associated with these new generations, and will report back anything I hear or can find.
We have the opportunity to update the detection for v7.14 if we move fast.
Provisional information from Wikipedia (usual caveats apply)
We are missing:
CC 6 (Pascal architecture GTX 10x0 series)
CC 7.0 (Volta architecture - Titan V)
CC 7.5 (Turing architecture - the new RTX 20x0 series)
Pascal seems to use cores_per_proc = 128, so our default is good for now.
Volta and Turing seem - very, very, provisionally - to use cores_per_proc = 64, so we would currently be overstating the power of TITAN cards twofold, and this would continue with Turing cards.
The text was updated successfully, but these errors were encountered:
The BOINC client decodes the 'Peak Flops' value for NVidia GPUs according to the architecture used in each succeeding card generation. The calculation is done in COPROC_NVIDIA::set_peak_flops() at
boinc/lib/coproc.cpp
Line 525 in f14d96d
The architecture is assessed from the Compute Capability, and we currently have values for up to CC 5.
We didn't add any extra handlers for the CC 6.1 reported by the GTX 10x0 range.
Today (20 Sep 2018) was supposed to be the launch day for the new RTX 2080 and RTX 2080 Ti cards. I have seen a screen shot implying that they have CC 7 (or possibly 7.5). I'm putting out an urgent call for the cores_per_proc value associated with these new generations, and will report back anything I hear or can find.
We have the opportunity to update the detection for v7.14 if we move fast.
Provisional information from Wikipedia (usual caveats apply)
We are missing:
CC 6 (Pascal architecture GTX 10x0 series)
CC 7.0 (Volta architecture - Titan V)
CC 7.5 (Turing architecture - the new RTX 20x0 series)
Pascal seems to use cores_per_proc = 128, so our default is good for now.
Volta and Turing seem - very, very, provisionally - to use cores_per_proc = 64, so we would currently be overstating the power of TITAN cards twofold, and this would continue with Turing cards.
The text was updated successfully, but these errors were encountered: