11 posts.
An interconnect routes each transfer to a slave by its address, using the memory map. Address decoding worked through, with arbitration and bridges to APB.
A system on chip puts processors, memory, peripherals and accelerators on one die, joined by an on-chip interconnect. The parts, and why they share one chip.
How a cache splits an address into tag, index and offset, why associativity removes conflict misses, LRU replacement, write policies and average access time.
A DRAM bit is charge on a tiny capacitor. How a read shares that charge with a bitline, why every row must be refreshed, and what rows and banks do to latency.
A warp's 32 loads are served in 32-byte sectors. Neighboring threads reading neighboring words need 4 sectors; a large stride needs 32, wasting 87.5%.
In SIMT, the 32 threads of a warp issue one instruction together. When they branch different ways, the warp runs each path in turn with lanes masked off.
When the work is parallel, does enough arithmetic per byte, and is big enough to pay for the data transfer. A roofline comparison, worked through.
A GPU splits work into thread blocks, runs each on a streaming multiprocessor in warps of 32, and hides memory latency by switching warps.
B and J immediates look shuffled, but each immediate bit lives in at most a few fixed instruction bits across all formats, which keeps the decoder small.
A hazard is when the next instruction can't safely enter its next stage. Data, control and structural hazards, forwarding, and why a load costs one stall.
RISC-V is an open instruction set. How its 32 registers and six instruction formats work, worked through for addi x1, x0, 5, which encodes as 0x00500093.