ASUS GeForce GTX 780 Review
Published by Marc Büchel on 23.05.13Page:
Technical Data / Specifications
With the ASUS GeForce GTX 780 you get a card which has been built entirely according to the reference design. The ASIC quality measured on our sample was 70.8 % which is quite a bit below average considering it being at 73 % on the cards we had in our hands.
About ASIC quality:
NVIDIAs latest flagship GPU is based on their massive Kepler GK110-300 chip, which is manufactured by TSMC using the latest 28 nanometer process technology. With the GeForce GTX 780 you get twelve SMX units which results in 2'304 CUDA cores. Following the specs further, there are 192 TMUs, 48 ROPs, a 384 bit wide memory interface and 3 Gigabyte of GDDR5 memory, which is clocked at 1'502 MHz (effective 6'008 MHz). In the end there is a memory bandwidth of 288 Gigabyte per second. Concerning the GPU's clock speeds it runs at a base clock of 863 MHz and a boost clock of 900 MHz. During our tests the highest boost clock we measured was 992 MHz and the average boost clock was at 940 MHz.
The GTX 780 also supports GPU Boost 2.0. The very first version of GPU Boost stopped overclocking the GPU when a certain power target was hit. This new second version of GPU Boost stops overclocking the card, when a certain temperature is being reached. This makes sense since the temperature is the bigger inhibitor than the power target in most cases. In this case the temperature target is 80 degrees Celsius. In other words, as long as the GPU runs at less than 80°C it will keep overclocking until the maximum frequency has been reached. This is why in our case the average boost clock was 40 MHz higher than the typical boost clock NVIDIA advertises. Things will get even more interesting if you equip the GTX 780 with a watercooler and then increase the voltage a little bit. In this case frequencies way above 1'000 MHz should be absolutely no problem at all. NVIDIA even leaves it up to the user where he wants to set the temperature target. Therefore it's no problem to set the temperature target to for example 85°C to get an even higher boost frequency.
| nVidia GeForce |
GeForce GTX Titan |
ASUS GTX 780 |
GeForce GTX 680 |
| Chip | GK110 | GK110 | GK104 |
| Process | 28 nm | 28 nm | 28 nm |
| Transistors | 7.10 Billion | 7.10 billion | 3.54 billion |
| GPU clock | 837 MHz | 863 MHz | 1'006 MHz |
| GPU Boost clock | 876 MHz | 900 MHz | 1'059 MHz |
| Memory | 6'144 MB GDDR5 | 3'072 MB GDDR5 | 2'048 MB GDDR5 |
| Memory clock | 1'502 MHz (6'008 MHz) | 1'502 MHz (6'008 MHz) | 1'502 MHz (6'008 MHz) |
| Memory interface | 384 Bit | 384 Bit | 256 Bit |
| Memory bandwidth | 288'400 MB/s | 288'400 MB/s | 192'300 MB/s |
| TMUs | 224 | 192 | 128 |
| Shader Cores | 2'688 (14 SMX) | 2'304 (12 SMX) | 1'536 (8 SMX) |
| ROPs | 48 | 48 | 32 |
| Maximum board power | 250 Watt | 250 Watt | 195 Watt |
| PCB Type | Reference Design | Reference Design | Reference Design |
| Lenght (PCB - Total) | 27.0 - 27.0 cm | 27.0 - 27.0 cm | 25.6 - 25.6 cm |
| Cooler | NVIDIA Reference | NVIDIA Reference | NVIDIA Reference |
| MSRP | $999 | $649 | $499 |
NVIDIA decided to equip the GTX 780 with the same cooler they already equipped their GeForce GTX Titan with. In this case you get a so called vapor chamber, quite a large heatsink and an advanced fan regulation system. Disassembling the GeForce GTX 780 really takes some time, since there is an astonishing number of screws and NVIDIA is using lots of different types. Once the cooler has been removed you can see thermal pads on the bottom side. NVIDIA decided to provide the GPU as well as the memory chips and the VRM area with cooling. The actual cooling is being handled by a blower type fan, which NVIDIA places in a metallic frame. Talking about the "advanced fan regulation system" would mean that we repeat the story with the power target. Therefore at this point we refer to the the above paragraph.
A closer look at the PCB shows that NVIDIA equipped this card with an analogue eight phase power design. Basically NVIDIA is using the exact same PCB like you get with the more expensive GeForce GTX Titan. The GPU in this case can rely on six phases and the memory gets an additional two phases. In our opinion this power design isn't too strong. We're quite curious to see the first custom designs from the NVIDIAs board partners. We're quite sure, that there will be some very interesting solutions.
The voltage regulation chips are identical too. Therefore we find a OnSemi NCP4206 from ON Semiconductur for the GPU and an unidentified Richtek 2-phase PWM labelled 0T=FA R1W chip for the memory.
The memory chips used are made by Samsung and carry the model number K4G20325FD-FC03. They are specified to run at 1'500 MHz (6'000 MHz effective).
Discuss this article in the forums
| Navigate through the articles | |
ASUS GTX 670 DirectCU Mini OC Review |
ASUS GeForce GTX 770 DirectCU II OC Review
 |
|
|