GTX 380: Vertex Shaders (512), ROP's (64), Pixel Shaders (512)
Core Speed (700, 1.6), Pixel Fill Rate (44,800), Texture Fill Rate (89,600)
Memory Bandwith/Type (512Mbit GDDR5), Memory Speed (1100Mhz), Memory Bandwith (281.6 GB/sec)
Just a note: the bandwidth for the 380 was calculated incorrectly, the Anandtech article clearly states it's going to have a 384 bit bus, not 512, so: (1100MHz*4)*(384bit/(8bit/byte))= 211,2 GB/s or 195,7 GiB/s, but i think it's usualy measured in GB. Over 200 GBps is still a lot though.
Nvidia's Shaders are faster than ATI's so ATI having a crap load more of them doesn't make that much of a difference, the 4890 has 800 of them while the GTX275 only has 240, but both cards perform similar, the GTX380 has more than doubled them (512 vs 240) while ati's have been exactly doubled (1600 vs 800), so it's already a sign that the GTX380's might be faster.
Yeah, the Nvidia's shaders are a lot faster, they work at A LOT higher clockspeed than the GPU core clock, in ATI cards they work at GPU core clockspeed. Also, ATI's shaders are grouped into packs of 5, with only one of those 5 shaders being able to perform all operations and the other four can only do simpler stuff. In Nvidia's cards, all shaders have full computational capabilities. ATI makes up for these differences by having a shitload more shaders.
Looks like nvidia has pretty much doubled everything up, the same as ati did with the 5 series. I wonder if all the architectural optimizations are going to make any difference in gaming performance. If not, the GTX 3xx vs HD5xxx performance differences should be around the same as GTX2xx vs HD4xxx. These new cards have the potential to be awesome, now we jus have to wait for the prices to see if it's worth it.