Why exactly is the GTX 970 faster than the GTX 770?

[fragoutcarpet]

The question might sound dumb, but comparing the specifications of these two cards (along with that of the GTX 780 to put things in perspective), we find

 

GTX 970:

 

CUDA cores: 1664

Texture Fill Rate: 109 GigaTexels/sec

Memory Bus Width: 256-bit

Memory Bandwidth: 224 GigaBytes/sec

 

GTX 770:

 

CUDA cores: 1536

Texture Fill Rate: 134 GigaTexels/sec

Memory Bus Width: 256-bit

Memory Bandwidth: 224.3 GigaBytes/sec

 

GTX 780:

 

CUDA cores: 2304

Texture Fill Rate: 160.5 GigaTexels/sec

Memory Bus Width: 384-bit

Memory Bandwidth: 288.4 GigaBytes/sec

 

As, I've been comparing video card performance using these stats in the past, it's kinda unnerving to watch the entire theory you've built up in your head, be blown to smithereens. Although the 780 beats out the other cards, in terms of specs, by a landslide, the 970 performs almost as well, if not better, than the 780. Specs-wise, there not a whole lot separating the 970 and the 770.

 

Please, don't hesitate to get technical.

[Pugs501]

Don't look at the specs for determine if you want it or not.

[Thebman712]

maxwell cores are more efficient and use less power and it has more vram

[That Norwegian Guy]

Each CUDA core, in layman's terms, got more potent/ complex.

[FloRolf]

You can only compare these if you have two cards with the same architecture. But here we have Kepler vs. Maxwell.

[alby800]

The actual chip is different I believe, the "Architecture"

 

Like why is the i7 2700k slower than the i7 4790k if they are running at the same speeds?

[iAsuno]

The 970 is based on the new architecture "Maxwell" while the 770 is based on the older "Kepler" architecture.

[RevoltTrain]

Because science and stuff.

[Tacitus]

Dunno.

[EMENCII]

Never look at specs to see what's faster...

Kepler vs Maxwell is like amd vs nvidia, specs not comparable, are just two different videocards...

[Enderman]

Different architectures. With the newer architectures, the manufacturing process gets smaller. This means they can fit more transistors in a cude core while taking up the same or less space than the previous architecture. Less power consumption, heat, and more vram don't really affect the fps performance, but more transistors on a die do. So basically the cuda cores despite being less, got more powerful.

[Real_PhillBert]

There are a few more cores but the cores are ~40% more powerful than in the 770.

[Briggsy]

Nvidia has been dropping considerable money into the SOC market space with their tegra chip, and digging their way into learning all the secrets of arm processor technology, and learning how and why other mobile processors are so efficient. although Nvidia are losing a buttload of money in that market, the secrets they have gleaned over the years are paying in dividends, and Nvidia are porting that technology into their GPUs. per core performance and power efficiency, upscaled to as big as they need to to slightly outperform the competition for now. 

 

An interesting side note is that AMD have been sticking it to intel in just the same fashion, but in reverse. the knowledge they gained from buying ATI allowed them to have a massive iGPU gain over intel. AMD could make their iGPU more powerful by a factor of ten, but they don't need to, plus they dropped the ball on the cpu end so they aren't gaining as much ground as they could be.

 

apples to oranges yes, but not surprising that collaboration into other areas is allowing each of the companies to break new ground. too bad its money that drives these companies, and not hippy love.

[Toggle]

The 970 is more efficient and each cuda core is able to do more than those of the last generations of cards. The new ones are basically a small improvement to efficiency and processing power. It's a very similar idea to processor generations. An increase in efficiency while being able to push better performance at a lower cost.

[fragoutcarpet]

I'm lead to believe that both Kepler and Maxwell are based on the a 28nm manufacturing process. What else could be causing the disparity? If someone could go into the specifics, it would be much appreciated.

[fragoutcarpet]

Like, if the Cuda 5.0 cores are performing better, what has changed in these cores that has brought out such a tremendous increase in performance and efficiency

[Cl2G]

Maxwell cores & color compression on the memory bus