I got my hands on some official documents from NVIDIA thanks to my Asian friends, thus I decided to share my knowledge with you.
NVIDIA GeForce GTX 780 Ti “the best gaming GPU on the planet”
GeForce GTX 780 Ti in a nutshell:
- 25% more cores than GTX 780 — 2880
- GPU Boost 2.0 — Max Clocks
- Power Balancing — Max OC
- Blazing Fast Memory — 7 Gbps
GeForce GTX 780 Ti specifications
The new flagship was not designed as TITAN replacement, it will serve a different purpose, which is strictly gaming-oriented. TITAN is still better for CUDA developers, researches or scientists, since it provides good double-precision performance and 6GB frame buffer. The GTX 780 Ti on the other hand, will (for now) only ship with 3GB memory. Official document does not mention any 6GB or 12GB variants.
NVIDIA GeForce GTX 780 Ti is equipped with 2880 CUDA cores. The most interesting part is that GTX 780 Ti still has high number of 64-bit CUDA cores (960), this is even more than GTX TITAN. So it seems that the only benefit of having a TITAN right now is 6GB frame buffer.
GTX 780 Ti is using GK110 GPU with 5 Graphics Processing Clusters, 15 Streaming Multiprocessors, 2880 FP32 CUDA cores, 960 FP64 CUDA cores, 240 Texture Mapping Units and and 48 Raster Operating Units.
Card is equipped with 3GB GDDR5 memory clocked at 7 GHz, across 384-bit interface. This equals to 336 GB/s bandwidth (16GB/s more than R9 290X). The GTX 780 Ti has higher base clock, although AMD does not officially confirm 800 MHz clock for 290X. On the other hand 780 Ti has lower boost clock (928 MHz vs 1000 MHz), but AMD is using peak performance clock, which is far from real-world scenario.
Moving on to power characteristics. GTX 780 Ti has much lower TDP than 290X, which is 250 watts. This is actually the same number as for TITAN and 780.
| GeForce GTX 780 Ti Official Specifications | ||||
|---|---|---|---|---|
| GeForce GTX 780 | GeForce GTX TITAN | Radeon R9 290X | GeForce GTX 780 Ti | |
| Picture |  |  |  |  |
| GPU | 28nm GK110 | 28nm GK110 | 28nm Hawaii XT | 28nm GK110 |
| Cores | 2304 | 2688 | 2816 | 2880 |
| TMUs | 192 | 224 | 176 | 240 |
| ROPs | 48 | 48 | 64 | 48 |
| Memory | 3GB GDDR5 | 6GB GDDR5 | 4GB GDDR5 | 3GB GDDR5 |
| Memory Bus | 384-bit | 384-bit | 512-bit | 384-bit |
| Base Clock | 863 MHz | 837 MHz | 800 MHz | 875 MHz |
| Typical Boost Clock | 900 MHz | 876 MHz | 1000 MHz | 928 MHz |
| Memory Clock | 1502 MHz | 1502 MHz | 1250 MHz | 1750 MHz |
| Effective Clock | 6008 MHz | 6008 MHz | 5000 MHz | 7000 MHz |
| Bandwidth | 288 GB/s | 288 GB/s | 320 GB/s | 336 GB/s |
| Power Connectors | 8+6 Pin | 8+6 Pin | 8+6 Pin | 8+6 Pin |
| TDP | 250 W | 250 W | 290 W | 250 W |
| Launch Price | $649 | $999 | $549 | $699 |
GeForce GTX 780 Ti Power Balancing for overclockers
NVIDIA equipped its GTX 780 Ti with a new feature called Power Balancing. Without this feature power drawn from: 6-pin power connector, 8-pin power connector and PCI Express interface would be balanced across these three sources respectively depending on the current load. However if user overclocks the card power delivery becomes unbalanced, thus card draws more power from one source than the others. To fix this problem NVIDIA came up with an idea of Power Balancing. You probably see where this is going. With this feature enabled GPU can steer the power from one input to another. This will improve overclocking capabilities in comparison to GTX 780 or the TITAN.
Raw horsepower
Chart was remade from the original
NVIDIA says that its new flagship has 35% higher texture fillrate than R9 290X. It also boasts 24% more geometry throughput.
| RAW horsepower | ||
|---|---|---|
| NVIDIA GeForce GTX 780 Ti @ 993 MHz | AMD Radeon R9 290X @ 1000 MHz | |
| Geometry | 5 billion triangles/s | 4 billion triangles/s |
| Texture fillrate | 238 GTexel/s | 176 GTexel/s |
| GFLOPS | 5720 | 5632 |
| Memory Bandwidth | 336 GB/s | 320 GB/s |
Bigger die equals better heat dissipation
NVIDIA clearly targets Radeon R9 290X in their comparisons. The green team is well aware of AMD problems with the noise and the temperature of their new flagship. NVIDIA’s theory is that since R9 290X is using 455mm2 die and GTX 780 Ti is based on 533 mm2 die, it equals to lower thermal density, thus less power condensed into smaller area. Long story short, NVIDIA’s GPU will generate less thermal density per square millimeter, so it is easier to dissipate.
NVIDIA made a test 20 minute Crysis 3 run with R9 290X and 780 TI. According to their data after 2 minutes R9 290X drops to 720 MHz, while GTX 780 Ti sustains 940 MHz clock. The average clock speeds are 799 MHz and 968 MHz for 290X and 780 Ti respectively.
NVIDIA:
Low thermal density translates directly into better cooling, acoustics and performance
| Thermal characteristics | ||
|---|---|---|
| NVIDIA GeForce GTX 780 Ti | AMD Radeon R9 290X | |
| Die Size | 533 mm2 | 455 mm2 |
| Power | 250 W | 290 W |
| W/mm2 | 0.47 | 0.64 |
| Temperature | 83 C | 95 C |
GeForce GTX 780 Ti gaming performance
Check this post for gaming performance.
GeForce GTX 780 Ti price
NVIDIA already announced that their newest, and fastest graphics card will be available for $699, which is obviously much more than R9 290X’s launch price ($549), but also much less than TITAN’s ($999). Official document does not mention any special edition of the card.
