**And again, I get "Compiling entry function...sm_10, sm_11, sm_12, sm_13, sm_21, and sm_20", but nothing for sm_30 or sm_35

Maxwell Compatability Guide says to use

/usr/local/cuda/bin/nvcc
  -gencode=arch=compute_20,code=sm_20
  -gencode=arch=compute_30,code=sm_30
  -gencode=arch=compute_35,code=sm_35
  -gencode=arch=compute_35,code=compute_35
  -O2 -o mykernel.o -c mykernel.cu

Note the omission of the 1xs -- on compiling my hologram application with CUDA 6 on Windows (haven't upgraded Linux yet) I got warnings that they are deprecated and soon to be removed. OTOH, after putting all the recommended options into the Windows compile my GTX 750 Ti still ran my reconstructions at exactly the same speed.

---

How about Maxwell being compute 50 ...

-gencode=arch=compute_50,code=sm_50
-gencode=arch=compute_50,code=compute_50

Just a guess.

---

To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones

How about Maxwell being compute 50 ...

-gencode=arch=compute_50,code=sm_50
-gencode=arch=compute_50,code=compute_50

Just a guess.

Now that's a puzzle. Yes, my results are saying "computeCap 5.0". I'll have to check tomorrow whether the URL I quoted matches the documentation I have on my Windows systems.

[Edit]
Actually, due to the wonders of broadband, ssh, and cygwin:

admD405@W7-SE-D304-01 /cygdrive/d/ProgramData/HoloMov/HoloMovie
$ grep sm HoloMovie_vs2010.vcxproj
      <CodeGeneration>compute_10,sm_10;compute_20,sm_20</CodeGeneration>
      <CodeGeneration>compute_10,sm_10;compute_20,sm_20</CodeGeneration>
      <CodeGeneration>compute_10,sm_10;compute_20,sm_20</CodeGeneration>
 <CodeGeneration>compute_20,sm_20;compute_30,sm_30;compute_35,sm_35;compute_50,sm_50;compute_50,compute_50</CodeGeneration>

so yes, my latest docs did specify sm_50 -- but I still need to check that I actually applied it to x64 Release version...
[/Edit]

---