Nvidia’s GeForce GTX 780 graphics card reviewed

Say Hello to the Titan’s younger brother
NVIDIA has been aggressive lately with its unveil of the GTX Titan a few months back, and now the progression onto the next generation of their GeForce GTX series. The GTX Titan raised many eyebrows considering its Rs. 80,000/- price tag and exceptional performance. The Titan was a the first of its kind ever released by Nvidia, and surely, there was going to be a trimmed down less expensive version of it. It was only a matter of time when Nvidia would announce it.

Come May 23, 2013, and Nvidia unveil their first card of the GTX 700 line-up, the GeForce GTX 780. The 780 is pretty much like the Titan, it is based on the same chipset the Titan uses i.e. the GK110, only without the bells and whistles the Titan offers along with the ultra-premuim price tag pinned to it.

Looking closely at the GTX 780 i.e. GK110
The GK110 represents the cream layer of Nvidia’s ageing Kepler lineup. Yes, its true, the Kepler lineup is still very much relevant, and will remain so, atleast in the GPU segment, for another year or so. The funny thing is, we’ve never really explained what the GK110 is all about, even when reviewing the GTX Titan. So the GTX 780 gives us an opportunity to do that. To top it off, our understanding on the subject is deeper, since the Titan and 780 are identical in so many ways. There is a reason why the GTX 780 is called the Titan’s ‘younger brother.’

First of all, what is the ‘GK110’. The GK110 is the chipset the GTX 780 is based on. What this means is, the GK110 chipset provides the GPU the underlying architecture and rules required for it to function. Now take a look at the diagram below (Image courtesy: Techreport). This is an oversimplified and massively shrunk down version of the GK110 GPU architecture. Keep in mind, we’re trying to show you the architecture of the most complex consumer semiconductor product in history. Blame that on the whopping 7.1 billion transistors used by the GK110. That is virtually super-computing level of power, now possible in your very homes.

Looking onto the diagram closely, we see that it is divided in five blocks. Each of these blocks is a ‘graphics processing cluster’ or GPC. Now, the GK110 has five of them, which is very significant, since one GPC is virtually a single GPU itself.  Why one GPC is like a sole-GPU in itself because each GPC has its own rasterizer engine and three separate shader multiprocessing engines (SMX units). Each SMX in turn has 192 shader ALUs, commonly known as ‘shader cores’, and 16 texture management units. Do the math and scale them up across 5 GPCs, and then you get the real horsepower of the GK110. So you could say that the GK110 is built of five standalone GPUs, and you won’t be wrong. That’s something to boast to your friends about, if you do get to own the GTX 780 or the Titan, eh?

Now, maintaining a chip with a size as large as this is very complicated. A single flaw in any one of the sectors can bring the whole ship down. And thus chip manufacturers disable parts of the chip that aren’t perfect during Quality Testing, to manage the massive processing pool. In the GTX Titan, the GK110 has one of its SMX units disabled, while in the GTX 780, three of its SMX units are locked.

The SMX units locked in the GTX 780 are not fixed. They can either be spread around the five GPCs, or they can be present in one single GPC, shutting the entire GPC down. In the case where all three SMX units disable belong to one GPC, you only have the four remaining GPCs working. And thus, the card will have four raster engines. In the other scenario where the locked SMX units belong to different GPCs, all the five GPCs will be active, thus giving you five raster engines. This is entirely random, and what configuration you get is based totally on luck. The performance difference between both the configurations is unknown, sadly. We have no way of testing it. You can give your ideas, if any, in the comments section.

Another difference between the Titan and the 780 are their capacities for double-precision floating-point math. The GK110 by principal was designed for Nvidia’s compute-focused Tesla lineup, which requires high precision. On the contrary, real-time graphics rendering for video games don’t require that level of precision, so that aspect is scaled down on the GK110 for the GTX 780, whereas for the Titan it is fully unlocked. The GTX 780 has precision 1/24th of that of the Titan, which is enough to maintain compatibility.

Apart from the GPCs, there is no difference whatsoever between the Titan and the GTX 780, you can verify that from the table give above. And physically, they both are almost exactly the same. So much for the double price tag the Titan commands, since the manufacturing costs of both the cards for Nvidia are the same, I suspect. In both the cards, all the ROP partitions and memory controllers are active, and so is its full 1536KB of L2 cache. Both cards share the 6Gbps memory transfer rate. The GTX 780 is available with 3GB of GDDR5 memory though, compared to the Titan’s 6GB.

It is worth mentioning that the 780’s clock frequencies are a tad higher than that of the Titan. The 680 has a base and Boost clock of 863Mhz and 900Mhz, compared to the Titan’s 836 and 867Mhz, respectively. All these ultra-fine maneuvers within the GK110 chipset leads to the following numbers:

Compared to the GTX 680, the 780 lashes ahead in every single key rate. Nvidia officially claims that the 780 is 70% faster than the GTX 580, and about 40% faster than the GTX 680. AMD’s single fastest GPU, the Radeon HD 7970 is also beaten left right and center by the GTX 780, both of them clashing only in their memory bandwidths.

The Card
Physically, the GTX 780 is quite similar to the GTX Titan. The dimensions of the 780 are exact to that of the Titan, which is 10.5 x 4.37 inches(L x W). And this is also the length of the Radeon HD 7970. So people with a cabinet which was able to fit in a GTX 570, GTX 580 or GTX 670, will find fitting the GTX 780 a success. It also has the same Aluminium body like the GTX Titan. This is the third GPU from Nvidia with an Aluminium body, the first one being the GTX 690.

New Fan with Adaptive Controller
The single cooler in the GTX 780 is superb. Its the same as the Titan, and we liked the cooler on the Titan. Built from premium materials, it feels nice and heavy to touch. Think of the feeling as a the one when you shut doors of a Japanese car versus a German car. That ‘thud’ feeling the German car gives, represents high quality build. The 780 has the same ‘vibe’ to it. Solidity.

Nvidia announced a new adaptive temperature controller for their GTX 700 lineup. What this means is a new fan controller that uses an adaptive temperature filter with an RPM and temperature targeted control algorithm to eliminate the unnecessary fan fluctuations that contribute to fan noise, providing a smoother acoustic experience.

We’ve spoken of all the features of the GTX 780, its physical architecture, its massive processing pool, all of this comes at a cost. And that cost, in India is Rs. 54,000/-. That about 2/3rd the price of the GTX Titan. In dollars, you could get a GTX Titan from America at that price. The Titan meanwhile in India costs a minimum of Rs. 80,000/-.