NVIDIA GeForce GTX 980 Review – GIGABYTE G1 Gaming
We already tasted the highly efficient Maxwell architecture when we reviewed the Geforce GTX 750 Ti, based on the GM107 GPU. It was pretty obvious since that time that we will see this architecture scaled up, because the potential was pretty big.
The GM204 core that powers up the new Geforce GTX 980 / 970 is based on the same architecture but is equipped now with 16 SMMs, totaling 2048 CUDA cores with 16 geometry units and 128 texture units. To compete with GTX 780 and R9 290 class graphic cards, the GM204 has some extensive memory architecture optimizations to reduce the bandwidth needed. This allowed NVIDIA to further reduce costs and use only a 256 bit memory subsystem, but running at a hefty frequency of 7.0GHz.
Another big improvement in terms of performance is the doubling of ROPs from 32 to 64 units, in the past this was reserved only for 512 bit cards. Don’t forget that the GTX 980 is the succesor of the GTX 680 and we have SMs with twice the performance per watt and 40% better performance per CUDA core. The big picture shows us that the GM204 has twice the performance of the GK104 (GTX 680 / GTX 770).
The GM204 itself is more complex than its succesor having 5.2 billion transistor, compared to just 3.54 billion for the GK104. PCB complexity remains about the same, while the TDP went down from 195W to 165W although the GPU clocks went up with more than 100MHz. Great engineering masterclass from the NVIDIA team in terms of performance per watt if you ask me. Can’t wait to see what the Maxwell based replacement for GTX 780 Ti will look like and also the AMD response to this…