The GPU (integrated with the CPU here) is a low end power efficient GPU of sorts. In other words, since it's on the CPU die, it has lesser SIMD cores lecture 21, slide 30 than a dedicated GPU (e.g. the high end Nvidia GPUs). So, the integrated GPUs is used to handle lighter or less demanding workloads, such as rendering the desktop (as Prof. Kayvon mentioned in the class) or playback HD videos (I believe the current Intel integrated HD graphics are capable of this task). Essentially, these tasks are not very demanding computationally for a GPU, and can be performed by the integrated GPU with higher power efficiency.
The second GPU is a dedicated GPU, which has a large number of SIMD cores, and possibly graphic specific compute resources. In short, this GPU is used for demanding tasks such as playing high end video games (rendering frames at high rates) which are computationally intensive, or for high performance/scientific computing using parallelism. The AMD/Nvidia dedicated GPUs usually consume more power (in the range of 95W for a desktop I guess) compared to a CPU's power usage.
VP7
Another point would be that IGPs share the same L3 with the CPUs.
Also IGPs let does n't eataway a lot of your memory bandwidth unlike traditional GPUs.
msebek
Since power consumption is super important in laptops (even back in 2011, when 3 hour battery life was good enough doggone it), companies were looking to decrease power consumption when the computer was not under heavy graphics load.
The generation of "switchable graphics" was born, where normally the CPU-integrated graphics were powering the display (advantages of this are as mentioned as above by @ghawk), which would allow the discrete chip to be put into a low power state.
If things got too intense graphically, then the discrete chip would be brought out of its low power state, and would begin to power the display.
This approach allowed the computer to reap the performance of having a discrete card while also being able to achieve energy savings.
kayvonf
@msebek. Nice comment! Very clear.
jcarchi
It's interesting to note that before 2010, macbook pros with 2 GPUs did NOT automatically switch between the GPUs. The user would have to do it him/herself
Question: Why are there two GPUs???
The GPU (integrated with the CPU here) is a low end power efficient GPU of sorts. In other words, since it's on the CPU die, it has lesser SIMD cores lecture 21, slide 30 than a dedicated GPU (e.g. the high end Nvidia GPUs). So, the integrated GPUs is used to handle lighter or less demanding workloads, such as rendering the desktop (as Prof. Kayvon mentioned in the class) or playback HD videos (I believe the current Intel integrated HD graphics are capable of this task). Essentially, these tasks are not very demanding computationally for a GPU, and can be performed by the integrated GPU with higher power efficiency.
The second GPU is a dedicated GPU, which has a large number of SIMD cores, and possibly graphic specific compute resources. In short, this GPU is used for demanding tasks such as playing high end video games (rendering frames at high rates) which are computationally intensive, or for high performance/scientific computing using parallelism. The AMD/Nvidia dedicated GPUs usually consume more power (in the range of 95W for a desktop I guess) compared to a CPU's power usage.
Another point would be that IGPs share the same L3 with the CPUs. Also IGPs let does n't eataway a lot of your memory bandwidth unlike traditional GPUs.
Since power consumption is super important in laptops (even back in 2011, when 3 hour battery life was good enough doggone it), companies were looking to decrease power consumption when the computer was not under heavy graphics load.
The generation of "switchable graphics" was born, where normally the CPU-integrated graphics were powering the display (advantages of this are as mentioned as above by @ghawk), which would allow the discrete chip to be put into a low power state. If things got too intense graphically, then the discrete chip would be brought out of its low power state, and would begin to power the display.
This approach allowed the computer to reap the performance of having a discrete card while also being able to achieve energy savings.
@msebek. Nice comment! Very clear.
It's interesting to note that before 2010, macbook pros with 2 GPUs did NOT automatically switch between the GPUs. The user would have to do it him/herself