While Jen Hsun is almost done rehearsing his speech for GTC about Maxwell GM200, we are looking for clues about Big Maxwell’s performance.
With no surprise we found TITAN X in CompuBench database. It will probably be removed by the time most of you will see this post, so we took few screengrabs, and posted the full opencl info dump, just in case.
According to new information from CompuBench, TITAN X has 24 compute units, which basically means it has 3072 CUDA cores. The software also reports on maximum clock, in other words boost clock, of 1076 MHz.
TITAN X outperforms the competition in every benchmark. Just remember that CompuBench top list is using median value from all entries, so some cards can actually perform better (for instance, if someone is using overclocking).
I’m leaving the interpretation to you, OpenCL benchmark has no value for gamers, but some of you might still find it interesting.
| NVIDIA GeForce GTX TITAN X OpenCL | |
|---|---|
| Name | Value |
| address_bits | 32 |
| compiler_available | 1 |
| double_fp_config | CL_FP_DENORM CL_FP_INF_NAN CL_FP_ROUND_TO_NEAREST CL_FP_ROUND_TO_ZERO CL_FP_ROUND_TO_INF CL_FP_FMA |
| driver_version | 347.84 |
| endian_little | 1 |
| error_correction_support | 0 |
| execution_capabilities | CL_EXEC_KERNEL |
| extensions | cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll cl_nv_d3d9_sharing cl_nv_d3d10_sharing cl_khr_d3d10_sharing cl_nv_d3d11_sharing cl_nv_copy_opts cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_fp64 |
| global_mem_cache_size | 393216 |
| global_mem_cache_type | CL_READ_WRITE_CACHE |
| global_mem_cacheline_size | 128 |
| global_mem_size | 0 |
| host_unified_memory | 0 |
| image_support | 1 |
| image2d_max_height | 32768 |
| image2d_max_width | 32768 |
| image3d_max_depth | 4096 |
| image3d_max_height | 4096 |
| image3d_max_width | 4096 |
| local_mem_size | 49151 |
| local_mem_type | CL_LOCAL |
| max_clock_frequency | 1076 <- IMPORTANT |
| max_compute_units | 24 <- IMPORTANT (24 * 128 = 3072) |
 | |
| max_constant_args | 9 |
| max_constant_buffer_size | 65536 |
| max_mem_alloc_size | 3.22E+09 |
| max_parameter_size | 4352 |
| max_read_image_args | 256 |
| max_samplers | 32 |
| max_work_group_size | 1024 |
| max_work_item_dimensions | 3 |
| max_work_item_sizes | 1024 1024 64 |
| max_write_image_args | 16 |
| mem_base_addr_align | 4096 |
| min_data_type_align_size | 128 |
| name | GeForce GTX TITAN X |
| native_vector_width_char | 1 |
| native_vector_width_double | 1 |
| native_vector_width_float | 1 |
| native_vector_width_half | 0 |
| native_vector_width_int | 1 |
| native_vector_width_long | 1 |
| native_vector_width_short | 1 |
| opencl_c_version | OpenCL C 1.1 |
| preferred_vector_width_char | 1 |
| preferred_vector_width_double | 1 |
| preferred_vector_width_float | 1 |
| preferred_vector_width_half | 0 |
| preferred_vector_width_int | 1 |
| preferred_vector_width_long | 1 |
| preferred_vector_width_short | 1 |
| profile | FULL_PROFILE |
| profiling_timer_resolution | 1000 |
| queue_properties | CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE CL_QUEUE_PROFILING_ENABLE |
| single_fp_config | CL_FP_DENORM CL_FP_INF_NAN CL_FP_ROUND_TO_NEAREST CL_FP_ROUND_TO_ZERO CL_FP_ROUND_TO_INF CL_FP_FMA |
| type | CL_DEVICE_TYPE_GPU |
| vendor | NVIDIA Corporation |
| version | OpenCL 1.1 CUDA |
Source: CompuBench