diff options
| author | Mahmoud <[email protected]> | 2018-11-05 20:01:42 -0500 |
|---|---|---|
| committer | Mahmoud <[email protected]> | 2018-11-05 20:01:42 -0500 |
| commit | f799cd5485a36de48abce286b2f69cee75d2be38 (patch) | |
| tree | 0cfd4f8baf37d2d692e1951fb7338e3a50f06fd2 /configs/tested-cfgs/SM7_TITANV | |
| parent | 9a57a802d1fc13ae47bfba42325e368b43fa96c3 (diff) | |
adding Volta sub_core model and double L1 BW
Diffstat (limited to 'configs/tested-cfgs/SM7_TITANV')
| -rw-r--r-- | configs/tested-cfgs/SM7_TITANV/gpgpusim.config | 29 |
1 files changed, 13 insertions, 16 deletions
diff --git a/configs/tested-cfgs/SM7_TITANV/gpgpusim.config b/configs/tested-cfgs/SM7_TITANV/gpgpusim.config index 03ef5b9..2093cc0 100644 --- a/configs/tested-cfgs/SM7_TITANV/gpgpusim.config +++ b/configs/tested-cfgs/SM7_TITANV/gpgpusim.config @@ -42,9 +42,9 @@ # Pipeline widths and number of FUs # ID_OC_SP,ID_OC_DP,ID_OC_SFU,ID_OC_MEM,OC_EX_SP,OC_EX_DP,OC_EX_SFU,OC_EX_MEM,EX_WB -## Volta GV100 has 2 SP SIMD units, 2 SFU units, 2 DP units per core +## Volta GV100 has 4 SP SIMD units, 4SFU units, 4 DP units per core ## we need to scale the number of pipeline registers to be equal to the number of SP units --gpgpu_pipeline_widths 4,4,4,4,4,4,4,4,12 +-gpgpu_pipeline_widths 4,4,4,4,4,4,4,4,8 -gpgpu_num_sp_units 4 -gpgpu_num_sfu_units 4 -gpgpu_num_dp_units 4 @@ -72,6 +72,8 @@ # For more info, see https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#shared-memory-7-x # disable this mode in case of multi kernels/apps execution -adpative_volta_cache_config 1 +# Volta unified cache has four ports +-mem_unit_ports 4 -gpgpu_cache:dl1 S:4:128:64,L:L:s:N:L,A:256:8,16:0,32 -gpgpu_shmem_size 98304 -gmem_skip_L1D 0 @@ -96,20 +98,15 @@ # 64 KB Const -gpgpu_const_cache:l1 N:128:64:8,L:R:f:N:L,S:2:64,4 -# enable operand collector --gpgpu_operand_collector_num_units_sp 12 --gpgpu_operand_collector_num_units_sfu 6 --gpgpu_operand_collector_num_units_mem 8 --gpgpu_operand_collector_num_units_dp 6 --gpgpu_operand_collector_num_in_ports_sp 4 --gpgpu_operand_collector_num_out_ports_sp 4 --gpgpu_operand_collector_num_in_ports_sfu 1 --gpgpu_operand_collector_num_out_ports_sfu 1 --gpgpu_operand_collector_num_in_ports_mem 1 --gpgpu_operand_collector_num_out_ports_mem 1 --gpgpu_operand_collector_num_in_ports_dp 1 --gpgpu_operand_collector_num_out_ports_dp 1 -# two banks per scheduler +# Volta has sub core model, in which each scheduler has its own reisiter file and EUs +# i.e. schedulers are isolated +-sub_core_model 1 +# disable specialized operand collectors and use generic operand collectors instead +-enable_specialized_operand_collector 0 +-gpgpu_operand_collector_num_units_gen 8 +-gpgpu_operand_collector_num_in_ports_gen 8 +-gpgpu_operand_collector_num_out_ports_gen 8 +# volta has 8 banks, 4 schedulers, two banks per scheduler -gpgpu_num_reg_banks 8 # shared memory bankconflict detection |