Any word in next gen iGPU?

[sonicrobby]

Are there any rumors or facts floating around about the possible graphics for the next gen rMBPs?

What will replace the Iris and Iris Pro, and how much of an improvement will they be?

 

[iKrivetko]

They will be replaced by Iris and Iris Pro. The rumours are generally reporting a 20-40 percent boost.

 

[Joelist]

They will be replaced by Iris and Iris Pro. The rumours are generally reporting a 20-40 percent boost.

True. I expect the same branding but the die shrink plus adding more pipes will boost performance. If they also do what is rumored and ramp up the bandwidth on the pipe between the iGPU itself and Crystalwell we could see even better than the 20-40% gains expected.

 

[Harinoorvirk28]

True. I expect the same branding but the die shrink plus adding more pipes will boost performance. If they also do what is rumored and ramp up the bandwidth on the pipe between the iGPU itself and Crystalwell we could see even better than the 20-40% gains expected.

so it will be almost as powerful as the 650m? the base model next gen I'm talking about

 

[dusk007]

The leaked specs all mention about a 20% increase in execution units. So it is basically 20% bigger GPU. It is also Gen 8 architecture which should again offer some extra performance. Intel a long time ago talked of about 40% of aimed for improvement. I guess clock rate should stay very similar, any gains from the new process will more likely go into power savings rather than the highest possible clocks.
The new architecture could be better than 20% more though. Given what room there is for improvement. Looking at what the smartphone GPUs can pull of on performance per Watt or Nvidia with Maxwell. Depends on how good Gen 8 turns out to be. No real news on that end.
I would say 40% faster is more like the minimum that can be expected.

 

[Joelist]

so it will be almost as powerful as the 650m? the base model next gen I'm talking about

The next gen from nVidia is the 850m. Iris Pro currently is already in the same league as a 650m at a fraction of the TDP. Remember too that Iris Pro does not need to be equal or better than current gen dGPUs to be a big winner. All it needs to do is be similar to the last gen. So for Broadwell this would be the 750m. And in some categories (like Open CL) Iris Pro blows all the dGPUs away already.

----------

The leaked specs all mention about a 20% increase in execution units. So it is basically 20% bigger GPU. It is also Gen 8 architecture which should again offer some extra performance. Intel a long time ago talked of about 40% of aimed for improvement. I guess clock rate should stay very similar, any gains from the new process will more likely go into power savings rather than the highest possible clocks.
The new architecture could be better than 20% more though. Given what room there is for improvement. Looking at what the smartphone GPUs can pull of on performance per Watt or Nvidia with Maxwell. Depends on how good Gen 8 turns out to be. No real news on that end.
I would say 40% faster is more like the minimum that can be expected.

They were also rumored to be working on pumping up the thruput between Crystalwell and the GOU - if that comes off then performance will really jump.

 

[dusk007]

They were also rumored to be working on pumping up the thruput between Crystalwell and the GOU - if that comes off then performance will really jump.

I don't think there is a big issue there. The bandwidth is quite good. Look at what a Maxwell 850M with DDR3 can do. Then Intel also has the LLC aka L3 cache that the GPU driver can use which offers at least as much cache as Maxwell has overall. There is like 300GB/s bandwidth to that cache. I don't think Crystalwell is too slow. It is the rest of the architecture in general with the whole pipeline that just isn't yet as efficient as it needs to be.
It is the first time Intel actually built a decent and big GPU wie a lot of die space dedicated to it but the architecture can keep up with AMD, Nvidia pulled ahead and there is still plenty room on 14nm. I doubt Crystalwell is a significant bottleneck for a while. The slower end will eventually get faster with DDR4 but the 50GB/s two way should do until the GPU gets a lot faster.
Obviously the non Crystalwell GPUs should get a boost if they get it now, like would assume the 13" CPU should get.