Hi,
While compiling a shared version of magma on linux with the tesla architecture, it was missing:
- magmablas_dsymv2 (missing #define magmablas_dsymv2_130 magmablas_dsymv2 in dsymv_tesla.cu file)
- magmablas_ssymv2 (missing #define magmablas_ssymv2_130 magmablas_ssymv2 in ssymv_tesla.cu file)
- magmablas_chemv2 (not implemented? in chemv_tesla.ca, or use magmablas_chemv_130)
Thanks
magma 1.4 on linux/tesla
Re: magma 1.4 on linux/tesla
By Tesla architecture, do you mean CUDA compute capability 1.x? These are fairly old cards now -- like the C1060 (circa 2009). Current cards are Kepler (compute capability 3.x). Please check the list at
https://developer.nvidia.com/cuda-gpus
There is confusion because NVIDIA uses "Tesla" in two different ways:
1) to mean the high-performance line: Tesla C1060 which 1.x, Tesla C2050 which is Fermi (2.x), Tesla K20 which is Kepler (3.x).
2) to mean the compute capability 1.x
We use Tesla in the second meaning, to mean the old 1.x cards. From the NVIDIA C programming guide:
"Devices with the same major revision number are of the same core architecture. The major revision number is 3 for devices based on the Kepler architecture, 2 for devices based on the Fermi architecture, and 1 for devices based on the Tesla architecture."
-mark
https://developer.nvidia.com/cuda-gpus
There is confusion because NVIDIA uses "Tesla" in two different ways:
1) to mean the high-performance line: Tesla C1060 which 1.x, Tesla C2050 which is Fermi (2.x), Tesla K20 which is Kepler (3.x).
2) to mean the compute capability 1.x
We use Tesla in the second meaning, to mean the old 1.x cards. From the NVIDIA C programming guide:
"Devices with the same major revision number are of the same core architecture. The major revision number is 3 for devices based on the Kepler architecture, 2 for devices based on the Fermi architecture, and 1 for devices based on the Tesla architecture."
-mark