We've covered several of nVidia's video card announcements this year, including the 780 + ShadowPlay, 770, and 760. With the launch of the GTX 760, NVidia's final slide stated that -- from memory -- there would be no further cards launched in a SKU lower than the 760 this year. At GN, we saw it as an indicator that no GTX 750 or 750 Ti would be released this year, but higher cards were still likely (at the time, the GTX 760 Ti seemed feasible, but never happened, and now we have the GTX 780 Ti).
Update: The 780 Ti is now posted on Newegg.
The 750 & 750 Ti models might launch next year, but we have no solid evidence of the existence of either card.
Regardless, nVidia's GTX 780 Ti now stands between the 780 and TITAN as a high-end video card. Where Titan aims to be used in professional computationally-intensive applications (with double-point precision), the 780 Ti is meant to be more "gaming-grade." In this post, we'll cover the nVidia GTX 780 Ti specs, some benchmarks vs. AMD's R9-290X, and the release date and MSRP.
NVIDIA GTX 780 Ti Specifications & 780 Ti Price
| GPU Spec | GTX 780 Ti | Titan | GTX 780 | GTX 770 | GTX 760 | AMD R9 290X |
| GPU Core | GK-110 | GK110-400 | GK110-300 | GK104-425 | GK104-225 | Hawaii XT |
| Transistor Count | 7.1B | 7.1B | 7.1B | 3.5B | 3.5B | 6.2B |
| Fab Process | 28nm | 28nm | 28nm | 28nm | 28nm | 28nm |
| Stream Processors | 2880 | 2668 | 2304 | 1536 | 1152 | 2816 |
| Core Clock | 875MHz | 837MHz | 863MHz | 1146MHz | 980MHz | 800MHz |
| Boost Clock | 928MHz | 876MHz | 902MHz | 1085MHz | 1033MHz | 1000MHz |
| Memory Clock | 1752MHz | 1502MHz | 1502MHz | 1753MHz | 1502MHz | 1250MHz |
| Memory Bandwidth | 336GB/s | 288.4GB/s | 288.4GB/s | 224.4GB/s | 192GB/s | 320GB/s |
| Memory Configuration | 3GB | 6GB | 3GB | 2/4GB | 2/4GB | 4GB |
| Memory Interface | 384-bit | 384-bit | 384-bit | 256-bit | 256-bit | 512-bit |
| TDP | 250W | 250W | 250W | 230W | 170W | 290W |
| ROPs | 48 | 48 | 48 | 32 | 32 | 48 |
| TMUs | 240 | 224 | 192 | 128 | 96 | 176 |
| MSRP | ~$700 | ~$1300 | ~$650 | ~$400 | $250 | ~$550 |
| Release Date | 11/7/13 | Available | Available | Available | June 25, 13 | Available |
 |  |
With a suggested retail price of $700, the GTX 780 Ti sits between nVidia's Titan ($1000), AMD's R9-290X (~$580), and nVidia's GTX 780 (now at $500).
GTX 780 Ti Analysis - The 780 Ti Does Not Linearly Compare to TITAN
It's easy to think the 780 Ti would outperform Titan based purely on the core count differential, but it's not a linear comparison, and I wish those reporting on the cards would stop perpetuating this conception.
The fact is that the 780 Ti is not meant to replace Titan. Titan was, technically, never meant to be a gaming card -- it's built for high-end developers, CUDA programmers, scientists, physicists, and anyone rendering highly-complex simulated models. This is evidenced by Titan's presence in the namesake Titan Super Computer.
Because the Titan is utilized in scientific research, it is required to perform its parallel processing with highly-accurate results. This is done by using double-precision floating point values on what we call GPGPUs, or General-Purpose Computing on Graphics Processing Units; because video games are exceedingly trivial in comparison to, say, a simulation of the known universe, video cards built for gaming use single-precision floating point values. The technology required to implement double-precision and higher computational accuracy is what makes the Titan expensive, and because it's not needed for gaming, it also makes the Titan a bad buy for pure gaming machines. Actually, the Titan's double-precision execution of floating point values will often slow games down, and nVidia's board partners generally disable double-precision mode by default for this reason (running the Titan in single-precision until manually toggled). Basically, don't directly compare/contrast the two. Different marketed purposes separate them.
So, that quick lesson on Titan's intended uses aside, we can dive more into the 780 Ti. As far as nVidia's product line-up goes, the 780 Ti is really where the buck stops for high-end gaming. The new Ti card keeps nVidia's high-end GK110 (28nm fabrication process) GPU, but has scaled up to 2880 CUDA cores -- outclassing the regular 780's 2304 cores. Further, the Ti-edition of the 780 is fitted with 3GB of 7GHz (adjusted) GDDR5 memory on a 384-bit memory interface. The extrapolated memory bandwidth of the GTX 780 Ti is 336GB/s. For comparison, the 290X has 320GB/s of memory bandwidth and the Titan/780 both rest at 288GB/s.
GTX 780 Ti - Overclocking & Thermal Concerns
NVidia also claims to have upped the ante on overclocking with the 780 Ti over its 780 and Titan predecessors. Because overvolting and overclocking the GPU's clockrate imbalances power draw from the 8-pin, 6-pin, and PCI-e power sources, OCing the 780 and Titan has proven less stable than we'd like. The 780 Ti will use "Power Balancing" to better normalize the current/wattage drawn from each source of power, theoretically improving stability and reducing voltage drops to the GPU. In turn, this increases the overall overclocking stability and cap of the 780 Ti, meaning that we should see higher deltas between OC'd and non-OC'd cards with the 780 Ti.
Back when Fermi launched, nVidia became the center of industry-wide criticism for its high-thermal, high-TDP, high-noise GTX 480; since then, the company has worked to mitigate thermals and power draw as much as reasonably possible. In continued effort to do this, the GTX 780 Ti's GK110 uses a 533mm^2 silicon die, approximately 78mm^2 larger than AMD's R9-290X, theoretically allowing better thermal dissipation over a larger surface area. This is because the thermal density (the amount of power per square millimeter, or W/mm^2 of the die) of the 780 Ti is significantly lower than the 290X, which has a smaller overall die area and a higher overall power-draw (290W TDP vs. 250W TDP).
The calculated thermal density of each product's GPU is 0.47W/mm^2 (780 Ti) and 0.64W/mm^2 (290X).
And, actually, that was a large part of nVidia's presentation: A direct battle against AMD's R9-290X, which has recently made major waves in the gaming GPU market. To interject my opinion here, I don't really think it's a fair comparison to make: The 290X might be AMD's current flagship device, but the 780 Ti is in an entirely different price bracket from the 290X. Take all these comparisons under that consideration. NVidia put together this "raw horsepower" chart, showing direct comparisons to the 290X on a specifications-level (note: Load performance calculated at 1080p; the 290X will perform closer at higher resolutions):
Although we haven't personally tested the 780 Ti yet, nVidia's internal benchmarks can be found below; take these with a grain of salt, but do keep in mind that the benchmarks conducted by either company (AMD & NVidia) are generally pretty fair. Then again, the 290X is significantly cheaper than the 780 Ti, so it's not exactly an immediately linear comparison between price-points.
NVidia GTX 780 Ti vs. AMD R9-290X Benchmarks & Performance
NVidia noted that the 780 Ti runs Assassin's Creed 4 on maximum settings @ 4K resolutions with 32FPS, and the new Batman game runs the same max/4K settings with a 46FPS framerate.
In addition to these graphics, nVidia claims that the 780 Ti outperforms the 290X by 26% in Battlefield 4 (47FPS vs. 38FPS, respectively), which AMD has closely supported in development. The caveat here, though, is that nVidia ran this test with the 290X on "quiet" mode, so it's tough to get an idea of the performance with the 290X at full bore. At worst, they'll perform about equally. At best, the 780 Ti should outperform the 290X marginally. The BF4 tests were run at 2560x1600, 4xAA/16xAF with Dx11 and "Ultra Quality" settings.
If you have any questions about modern GPU choices, drop a comment below or post on our forums!
- Steve "Lelldorianx" Burke.