OmniTechnoMancer: sort-of yes (the Larrabee-style approach). more, "start by vectorising scalar PowerISA then incrementally add as many 3D instructions and capabilities as it takes to reach par Performance/Watt with Radeon/Nvidia/MALI/Vivante"
emeb_mac has joined #nmigen
i think the important part here is to pick a benchmark
hwacha was running circles around commercial GPUs years ago at linpack workloads, but most people use GPUs for other things