NVIDIA is doing the scientific community a great service by code-naming its GPU architectures after pioneering physicists. "Kepler" became synonymous with the GeForce GTX 680 launch back in 2012; and although it got its debut with the mid-range GeForce GTX 750 Ti earlier this year, "Maxwell", NVIDIA's latest GPU architecture, is really beginning to unravel with the new GeForce GTX 980 and GTX 970 we are reviewing today.
The development cycle of GPUs isn't too different from that of CPUs. A new GPU architecture is introduced every two or so years, facilitated by a new silicon fab process, which allows designers to cram in more transistors, to up performance. The common foundry partner for both NVIDIA and AMD, TSMC, threw a spanner into the workings of this cycle when news broke of delays in the implementation of its 20 nanometer manufacturing node, successor to the 28 nm node on which NVIDIA's GeForce Kepler and AMD's "Volcanic Islands" families of GPUs were built. NVIDIA probably had to redesign the performance-segment GPU based on its swanky new "Maxwell" GPU architecture for the existing 28 nm node. Its fruition is the GM204, the 5.2 billion transistor chip at the heart of the GeForce GTX 980 and GTX 970.
With the introduction of the GeForce GTX 980 and GTX 970, NVIDIA's product stack looks very similar to the way it did in 2012, in which the performance-segment GK104 went into making the GTX 680 and GTX 670, positioned as high-end parts because of their relatively higher performance to AMD's high-end chips at the time. Much like "Kepler" and GK104, "Maxwell" and GM204 boast leaps in performance-per-watt and overall performance. However, those two facets aren't Maxwell's only propositions. NVIDIA announced a bucket-list of innovations in consumer-graphics technologies, such as support for Microsoft's next-generation DirectX 12 API, Dynamic Super Resolution, Multi-Frame Sampled AA, Voxel Global Illumination, and VR Direct. Such is NVIDIA's confidence in the superiority of its "Maxwell" architecture, that it thinks a 28 nm chip with 165W TDP can outperform "Kepler" based chips on the same 28 nm process with a 50 percent higher TDP.
The GeForce GTX 980, hence, is NVIDIA's newest high-end single-GPU graphics card, by possible virtue of its performance. It is priced at US $549, which is $100 cheaper than the GeForce GTX 780 at launch, but $50 more than the GTX 680 at its launch. The GeForce GTX 970, on the other hand, is a high-performance offering priced at $329. With the introduction of these two, NVIDIA announced the discontinuation of the GeForce GTX 780 and GTX 770 from the product stack. The GTX 760 even sees a price-cut that puts it at $219. NVIDIA priced the GTX 980 to lock horns with AMD's R9 290X, and the GTX 970 is slotted to compete against the R9 290. Will they succeed?
GTX 780 Ti
|Shader Units||1536||2304||2560||1664||2816||2x 2048||2688||2880||2048||2x 1536||2x 2816|
|ROPs||32||48||64||64||64||2x 32||48||48||64||2x 32||2x 64|
|Graphics Processor||GK104||GK110||Hawaii||GM204||Hawaii||2x Tahiti||GK110||GK110||GM204||2x GK104||2x Hawaii|
|Transistors||3500M||7100M||6200M||5200M||6200M||2x 4310M||7100M||7100M||5200M||2x 3500M||2x 6200M|
|Memory Size||2048 MB||3072 MB||4096 MB||4096 MB||4096 MB||2x 3072 MB||6144 MB||3072 MB||4096 MB||2x 2048 MB||2x 4096 MB|
|Memory Bus Width||256 bit||384 bit||512 bit||256 bit||512 bit||2x 384 bit||384 bit||384 bit||256 bit||2x 256 bit||2x 512 bit|
|Core Clock||1006 MHz+||863 MHz+||947 MHz||1051 MHz+||1000 MHz||1000 MHz||837 MHz+||876 MHz+||1126 MHz+||915 MHz+||1018 MHz|
|Memory Clock||1502 MHz||1502 MHz||1250 MHz||1750 MHz||1250 MHz||1500 MHz||1502 MHz||1750 MHz||1750 MHz||1502 MHz||1250 MHz|