It's interesting to note how well the bench scales with the amount of EUs for Intel in this graph, just as a speculation: 48 vs 20 gives about twice the fps (Haswell vs Broadwell), now take the Skylake GT2 score and suppose you have three times that plus eDRAM to remove bandwidth limits... ~30fps should be playable no?
It's not 48EUs versus 20EUs, but 48EUs + eDRAM(100GB/s equivalent bandwidth) versus 20EUs. You'll see that in most cases, the difference isn't even 2x:
http://www.anandtech.com/show/9320/intel-broadwell-review-i7-5775c-i5-5675c/7
You don't see a car going 4x the speed because the highway speed limits are increased by 2x and car maximum speeds are increased by 2x do you? You need both to enable the 2x speed. Same as with matching memory banwidth and GPU compute performance. You need 2x bandwidth for a 2x GPU to be 2x as fast as 1x GPU with 1x the bandwidth.
It's about 80%. Without the eDRAM, the performance difference between 48EU and 24EU parts lie in the 20-30% range. Pretty poor. Of course you have situations like Haswell's 15W GT3 where its so severely bound with TDP that the gain drops to 5-15%.
Anyways, we don't need to use wild speculations. Because some datapoints are already available for us. Intel claims 50% boost versus previous generation at same TDP.
72EU versus 48EUs. The scaling is about 60%, so you are getting 30% from that alone. The rest 20-25% comes from Gen 9 over Gen 8. At higher resolutions and settings, the texture compression technology and other enhancements will allow the gains to be greater than that, reducing the deficiency it has against Nvidia/AMD parts. Of course, that's against Geforce 750/Ti parts. This year, the pure gains of 16nm generation will beat the Iris Pro whatever to a pulp.
Intel seems to be a year late in both graphics and mobile. I wonder what their R&D investments are doing nowadays?
Also, i wouldnt say that 6200 outperformed the fastest Godaveri.
[URL]http://forums.anandtech.com/showthread.php?t=2470705[/URL]
The reason it varies a lot is because its Skylake that brings the architecture on par with the graphics players like AMD/Nvidia. You'll notice that in benchmarks. Broadwell is still deficient when it starts getting GPU limited. Fortunately for Intel, things like texture compression is a one-time low hanging fruit enhancement. So they won't be so far behind in the future.