We present two e±cient Apriori implementations of Fre-
quent Itemset Mining (FIM) that utilize new-generation graph-
ics processing units (GPUs). Our implementations take ad-
vantage of the GPU's massively multi-threaded SIMD (Sin-
gle Instruction, Multiple Data) architecture. Both imple-
mentations employ a bitmap data structure to exploit the
GPU's SIMD parallelism and to accelerate the frequency
counting operation. One implementation runs entirely on
the GPU and eliminates intermediate data transfer between
the GPU memory and the CPU memory. The other im-
plementation employs both the GPU and the CPU for pro-
cessing. It represents itemsets in a trie, and uses the CPU
for trie traversing and incremental maintenance. Our pre-
liminary results show that both implementations achieve a
speedup of up to two orders of magnitude over optimized
CPU Apriori implementations on a PC with an NVIDIA
GTX 280 GPU and a quad-core CPU.