aboutsummaryrefslogtreecommitdiff
path: root/configs/GeForceGTX750Ti/gpgpusim.config
diff options
context:
space:
mode:
authortgrogers <[email protected]>2018-05-16 19:20:53 -0400
committertgrogers <[email protected]>2018-05-16 19:20:53 -0400
commit355c86895faf122deecb5de5ef17ff4ff9d654d9 (patch)
treedb13a006b2478e6bc603d0624b0bb39ee7e9dc1d /configs/GeForceGTX750Ti/gpgpusim.config
parent97fcd60a4fdef22d3ab89eb660ec4cdc07553f72 (diff)
More sane naming convention
Diffstat (limited to 'configs/GeForceGTX750Ti/gpgpusim.config')
-rw-r--r--configs/GeForceGTX750Ti/gpgpusim.config131
1 files changed, 0 insertions, 131 deletions
diff --git a/configs/GeForceGTX750Ti/gpgpusim.config b/configs/GeForceGTX750Ti/gpgpusim.config
deleted file mode 100644
index 9366f93..0000000
--- a/configs/GeForceGTX750Ti/gpgpusim.config
+++ /dev/null
@@ -1,131 +0,0 @@
-# functional simulator specification
--gpgpu_ptx_instruction_classification 0
--gpgpu_ptx_sim_mode 0
--gpgpu_ptx_force_max_capability 52
-
-
-# SASS execution (only supported with CUDA >= 4.0)
--gpgpu_ptx_convert_to_ptxplus 0
--gpgpu_ptx_save_converted_ptxplus 0
-
-# high level architecture configuration
--gpgpu_n_clusters 5
--gpgpu_n_cores_per_cluster 1
--gpgpu_n_mem 2
--gpgpu_n_sub_partition_per_mchannel 1
-
-# Maxwell clock domains
-#-gpgpu_clock_domains <Core Clock>:<Interconnect Clock>:<L2 Clock>:<DRAM Clock>
-# In Maxwell, each pipeline has 32 execution units, so the Core clock is just the reported number.
--gpgpu_clock_domains 1137.0:1137.0:1137.0:2700.0
-
-# shader core pipeline config
--gpgpu_shader_registers 65536
-
-# This implies a maximum of 64 warps/SM
--gpgpu_shader_core_pipeline 2048:32
--gpgpu_shader_cta 8
--gpgpu_simd_model 1
-
-# 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
--gpgpu_pipeline_widths 2,0,1,1,2,0,1,1,2
--gpgpu_num_sp_units 8
--gpgpu_num_sfu_units 32
--gpgpu_num_dp_units 0
-
-# Instruction latencies and initiation intervals
-# "ADD,MAX,MUL,MAD,DIV"
--ptx_opcode_latency_int 6,12,13,13,210
--ptx_opcode_initiation_int 1,1,1,1,4
--ptx_opcode_latency_fp 6,12,6,6,374
--ptx_opcode_initiation_fp 1,1,1,1,4
--ptx_opcode_latency_dp 8,19,8,8,330
--ptx_opcode_initiation_dp 8,16,8,8,130
-
--gpgpu_cache:dl1 N:32:128:4,L:L:m:N:H,A:32:8,8
--gmem_skip_L1D 1
--gpgpu_shmem_size 65536
-
-# The alternative configuration for fermi in case cudaFuncCachePreferL1 is selected
-#-gpgpu_cache:dl1 N:64:128:6,L:L:m:N:H,A:32:8,8
-#-gpgpu_shmem_size 16384
-
-# 1024 sets, each 128 bytes 8-way for each memory sub partition. This gives 2MB L2 cache
--gpgpu_cache:dl2 N:1024:128:8,L:B:m:W:L,A:32:4,4:0,32
--gpgpu_cache:dl2_texture_only 0
-
--gpgpu_cache:il1 N:4:128:4,L:R:f:N:L,A:2:32,4
--gpgpu_tex_cache:l1 N:16:128:32,L:R:m:N:L,F:128:4,128:2
--gpgpu_const_cache:l1 N:64:64:2,L:R:f:N:L,A:2:32,4
-
-# enable operand collector
--gpgpu_operand_collector_num_units_sp 6
--gpgpu_operand_collector_num_units_sfu 8
--gpgpu_operand_collector_num_in_ports_sp 2
--gpgpu_operand_collector_num_out_ports_sp 2
--gpgpu_num_reg_banks 16
-
-# shared memory bankconflict detection
--gpgpu_shmem_num_banks 32
--gpgpu_shmem_limited_broadcast 0
--gpgpu_shmem_warp_parts 1
-
--gpgpu_max_insn_issue_per_warp 2
-
-# interconnection
--network_mode 1
--inter_config_file config_fermi_islip.icnt
-
-# memory partition latency config
--rop_latency 150
--dram_latency 130
-
-# dram model config
--gpgpu_dram_scheduler 1
-# The DRAM return queue and the scheduler queue together should provide buffer
-# to sustain the memory level parallelism to tolerate DRAM latency
-# To allow 100% DRAM utility, there should at least be enough buffer to sustain
-# the minimum DRAM latency (100 core cycles). I.e.
-# Total buffer space required = 100 x 924MHz / 700MHz = 132
--gpgpu_frfcfs_dram_sched_queue_size 16
--gpgpu_dram_return_queue_size 300
-
-# for Maxwell, bus width is 128bits, this is 64 bytes (32 bytes at each DRAM chip) per memory partition
--gpgpu_n_mem_per_ctrlr 2
--gpgpu_dram_buswidth 32
--gpgpu_dram_burst_length 8
--dram_data_command_freq_ratio 4 # GDDR5 is QDR
--gpgpu_mem_address_mask 1
--gpgpu_mem_addr_mapping dramid@8;00000000.00000000.00000000.00000000.0000RRRR.RRRRRRRR.BBBCCCCB.CCSSSSSS
-
-# GDDR5 timing from hynix H5GQ1H24AFR
-# to disable bank groups, set nbkgrp to 1 and tCCDL and tRTPL to 0
--gpgpu_dram_timing_opt "nbk=16:CCD=2:RRD=6:RCD=12:RAS=28:RP=12:RC=40:
- CL=12:WL=4:CDLR=5:WR=12:nbkgrp=4:CCDL=3:RTPL=2"
-
-# Maxwell has four schedulers per core
--gpgpu_num_sched_per_core 4
-# Two Level Scheduler with active and pending pools
-#-gpgpu_scheduler two_level_active:6:0:1
-# Loose round robbin scheduler
-#-gpgpu_scheduler lrr
-# Greedy then oldest scheduler
--gpgpu_scheduler gto
-
-# stat collection
--gpgpu_memlatency_stat 14
--gpgpu_runtime_stat 500
--enable_ptx_file_line_stats 1
--visualizer_enabled 0
-
-# power model configs
-# power modeling is currently disabled for GTX750Ti. The gpuwattch_gtx750ti.xml file present is only a placeholder and has NOT been tested for accuracy.
-# proper modeling of Maxwell power is planned, but should be considered completely unsupported at present.
--power_simulation_enabled 0
--gpuwattch_xml_file gpuwattch_gtx750Ti.xml
-
-# tracing functionality
-#-trace_enabled 1
-#-trace_components WARP_SCHEDULER,SCOREBOARD
-#-trace_sampling_core 0