If doubling both MIPS and FLOPS with AVX2 and FMA3 and possibly doubling the amount of EUs on the iGPU only increases TDP 2 watts it should cooler when any of those aren't in use, and will allow work to get completed faster when used so it will be idle sooner.
Quote:If doubling both MIPS and FLOPS with AVX2 and FMA3 and possibly doubling the amount of EUs on the iGPU only increases TDP 2 watts it should cooler when any of those aren't in use
Where did you get those figures from?
Why would you assume that using these instructions would significantly increase power consumption over different code?
That logic doesn't stack up. Especially when looking at previous Intel architectures that have significantly boosted max throughput via new instructions with small increases in power consumption.
Also the GT3 IGP with double the EUs will not be available on the desktop parts according to the chart.
We know that haswells idle power consumption is going to be MUCH better than previous generations but for high end workloads one would assume based on this data that power consumption should be the same or slightly higher in exchange for better performance. For most applications power consumption will likely be a bit lower due to fewer work cycles needed to complete the task, but this has nothing to do with whether AVX2 or FMA3 are used.
Quote:and will allow work to get completed faster when used so it will be idle sooner.
True.
AVX increased power consumption (
19w on 2500K) and my quick test with OCCT shows AVX adding 4-6° per core on my i5-3330. If you trust
Wiki SSE added 6watts (P2 450 27w P3 450 33w).
Fun fact AVX has a warm-up period allegedly to save power.
You're still assuming that the increased TDP is almost entirely based around the increase in MIPS/GFLOPs on AVX2 code, which makes little sense and has not been backed up with evidence. We'll have to wait and see.
Also it looks like the TDP increase over ivy bridge is 7 watts, not 2.
Edit: Ah, you're referring to the mobile cpus not the desktop cpus.