2 ms·
QDR-IV has lower-latency than L3 cache from CPUs. And you can achieve higher-bandwidth arbitrarily by simply talking to more QDR-IV chips. The harder the memor
by dragontamer 9y ago
QDR-IV has lower-latency than L3 cache from CPUs. And you can achieve higher-bandwidth arbitrarily by simply talking to more QDR-IV chips.
The harder the memory-problem becomes, the better it is for QDR-IV SRAM compared to CPUs or GPUs. An ASIC or FPGA designer can use any RAM that exists in the world, but CPU / GPU users are limited to DDR3, DDR4, the L1/L2/L3 caches, GDDR5, and maybe HBM2 (AMD Vega / NVidia V100)
Any memory-hard problem are basically solved by these exotic low-latency / high-bandwidth RAM chips + an FPGA or ASIC interfacing with them.