ttnn.MatmulMultiCoreReuseMultiCastDRAMShardedProgramConfig

class ttnn.MatmulMultiCoreReuseMultiCastDRAMShardedProgramConfig

Bases: object

This program config is a specialized config for very narrow tensors stored in DRAM.

from_json

property fused_activation

Optional fused activation function to apply during computation.

If specified, the activation function is applied directly during the DRAM-sharded matmul operation. This can provide significant performance benefits by avoiding additional memory round-trips in DRAM-based operations.

property in0_block_w

Block width for both input tensors along the K dimension (shared inner dimension).

Determines the data granularity by specifying how many tiles wide each block is along the inner dimension for both input_tensor_a and input_tensor_b in DRAM-sharded operations. This parameter must be chosen to align with the DRAM sharding strategy and optimize memory bandwidth utilization for both tensors.

property num_workers_per_dram_bank

Number of Tensix reader/compute workers assigned to each DRAM bank.

The default of 1 preserves the established cross-architecture path. Values of 2 or 3 split each bank’s width shard evenly across multiple workers on Blackhole. All readers for one bank use NOC0 and the same allocator-selected DRAM endpoint. The per-bank shard width in tiles must equal this value times the reader width.

property per_core_M

Number of output tiles each core processes along the M dimension.

Determines how the M dimension is distributed across cores in DRAM-sharded scenarios. This must align with the DRAM sharding pattern to ensure optimal performance and avoid memory access conflicts.

property per_core_N

Number of output tiles each core processes along the N dimension.

Determines how the N dimension is distributed across cores in DRAM-sharded scenarios. This parameter affects the multicast efficiency and must be compatible with the DRAM sharding configuration.

to_json