NVIDIA Launches Maxwell-Based GeForce GTX 980 and GeForce GTX 970 GPUs

MojoKid (1002251) writes NVIDIA has launched two new high-end graphics cards based on their latest Maxwell architecture. The GeForce GTX 980 and GTX 970 are based on Maxwell and replace NVIDIA's current high-end offerings, the GeForce GTX 780 Ti, GTX 780, and GTX 770. NVIDIA's GeForce GTX 980 and GTX 970 are somewhat similar as the cards share the same 4GB frame buffer and GM204 GPU, but the GTX 970's GPU is clocked a bit lower and features fewer active Streaming Multiprocessors and CUDA cores. The GeForce GTX 980's GM204 GPU has all of its functional blocks enabled. The fully-loaded GeForce GTX 980 GM204 GPU has a base clock of 1126MHz and a Boost clock of 1216MHz. The GTX 970 clocks in with a base clock of 1050MHz and Boost clock of 1178MHz. The 4GB of video memory on both cards is clocked at a blisteringly-fast 7GHz (effective GDDR5 data rate). NVIDIA was able to optimize the GM204's power efficiency, however, by tweaking virtually every part of the GPU. NVIDIA claims that Maxwell SMs (Streaming Multiprocessors) offer double the performance of GK104 and double the perf per watt as well. NVIDIA has also added support for new features, namely Dynamic Super Resolution (DSR), Multi-Frame Sampled Anti-Aliasing (MFAA), and Voxel Global Illumination (VXGI). Performance-wise, the GeForce GTX 980 is the fastest single-GPU powered graphics card ever tested. The GeForce GTX 970 isn't as dominant overall, but its performance was impressive nonetheless. The GeForce GTX 970 typically performed about on par with a GeForce GTX Titan and traded blows with the Radeon R9 290X.

Comments

Re:Only 4 displays, sticking to AMD.

[pushing-robot]

I know what you mean. I don't want to play games but am looking to carry sacks of grain through the Andes, and these cards lack the qualities of a trusty burro.

Still 28nm

[Solandri]

At this point I think it's safe to write off TSMC's 20nm fab process. It's not gonna happen, with signs pointing to development being shifted to 16nm instead.

A lot of what you see going on in the GPU and mobile front is being dictated by the failure of TSMC and other fabs to transition to 20nm for processors (memory is a lot easier and reached 16nm in 2013). Intel made the transition from 32nm to 22nm last year with Haswell and Bay Trail. The other fabs were supposed to leapfrog Intel by going from 28nm to 20nm this year. They haven't, which is what's allowed Intel to produce Atom SoCs with power usage approaching that of ARM SoCs. ARM has the lower power tech, but Intel's smaller lithography is mostly wiping out that advantage. If you see Intel successfully make the transition to 14nm in 2015 while the other fabs can't get 16nm to work, things are going to get really interesting on the mobile SoC front..

The GPU front is bleaker. Both nVidia and AMD use third party fabs like TSMC, so there's no competitive advantage to be had. We've just had to suffer with stagnating product lines and slow product releases because the expected lower power consumption in GPUs from 20nm didn't happen in 2014.