Intel vs. nVidia: where are they now?
Was just looking through the latest GfxBench submissions, and decided to compare Intel vs. nVidia across multiple tiers.
http://gfxbench.com/result.jsp?benchmark=gfx30
It really gives an interesting view. I'm eager to see how well the Iris 6100 scores. Given the power/performance improvement that's evident in Broadwell compared to Haswell, I'm hoping it gets something in the range of nVidia 830M.
Manhattan Offscreen
nVidia 850M (640 shader, 40W) 126.0
Intel Iris Pro 5200 (Haswell 320 shader, 45W) 78.2
nVidia 840M (384 shader, 30W) 67.1
nVidia X1 (512 shader, 15W) 63.4
nVidia 830M (256 shader, 25W) 55.7
Intel HD 4600 (Haswell 160 shader, 65W) 38.6
Intel HD 5500 (Broadwell 192 shader, 15W) 38.0
nVidia 820M (96 shader, 15W) 37.1
Intel Iris 5100 (Haswell 320 shader, 28W) 35.6
Intel HD 4400 (Haswell 160 shader, 15W) 31.6
nVidia K1 (192 shader, ~10W) 31.5
Intel HD 5300 (Broadwell 192 shader, 4.5W) 25.3
Intel Cherryview HD Graphics (128 shader, ~4W) 17.7
Intel Moorefield (128 shader, ~2.5W) 13.0
Intel Bay Trail (32 shader, ~4W) 9.7
P.S. the Iris 5100 score seems a little low to me. Several Intel submissions are unusually low, but there were enough submissions to find the ones with larger submission count (these scores are all median based). The lower scores probably reflect less mature drivers and/or optimizations. I would have expected something in the "40s". Iris Pro has the benefit of eDRAM and higher power, but that shouldn't account for more than 30-40% combined benefit.
P.P.S. the nVidia X1 is impressive... almost *outrageously* impressive. I'm thinking the double throughput FP16 capability really helped in this benchmark. It's just so far out of whack and impressive relative to the other Maxwell scores. Of course, it is on TSMC 20nm, too.
