http://gpu.userbenchmark.com/Compare/Nvidia-GTX-1060-6GB-vs-Nvidia-GTX-1070/3639vs3609
Why does 1070 have 50% more cores/ROPs/transistors than 1060/6, but is only 40% more powerful is synthetics and just 22% more powerful in games?
Because the number of transistors doesn't directly correlate to performance
>>60248949
Because their clocks and bandwidth are different, their frontend isn't 50% bigger on the 1070 to feed the extra shit.
tl;dr hardware isn't 2+2
>>60248949
1070 doesn't just have 50% more transistors, it has 50% CUDA cores two. Theoretically, wouldn't 1.5x cuda cores perform 50% faster than x cuda cores at the same clocks?
>>60249012
>but cores
What do you think cores are made of? Pixie dust? It's still transistors.
>>60248886
gtx 1060 has newer aluminum gpu armor which shields it from heat which means better performance than expected
>>60249069
1070 and 1080 have the same amount of transistors but different number of cores.
>>60249094
That's because the 1080 and 1070 are the exact same chip with the 1070's transistors/cores being fused off.
This is basic, it has been done for 2 decades.
>>60249120
I know, this isn't really relevant to the question. What exactly limits the 150% of cores of the same architecture clocked the same perform 50% better in real world applications?
>>60249145
Bandwidth and frontend performance.
It's the same with Fury X and Hawaii, the former had 50% more cores but the same frontend as Hawaii but it only was 15-20% faster
>>60249169
what exactly is front end? vram and voltage control?
>>60249199
Geometry units, prefetchers, decoders, etc basically the initial stage of the pipeline
>>60249226
The command scheduler should go into the frontend as well