Page 1: Test: NVIDIA GeForce GTX 660 Ti
NVIDIA has complemented their top-class graphics card portfolio with the GeForce GTX 680 and 690, rounding up the upper middle class segment with the GeForce GTX 670. But the middle price range continues to be dominated by AMD's Radeon HD 7800-series. This is where NVIDIA wants to get a foothold with the GeForce GTX 660 Ti. Once again, the Kepler architecture is the manufacturing basis, wich already turned out to be very effective with the first card, the GeForce GTX 680. NVIDIA succeeded in providing a high level of gaming performance coupled with the least possible power consumption. Together with GPU Boost, this efficiency is the most important aspect in NVIDIA's Kepler generation. The GeForce GTX 660 Ti is meant to be equal to the bigger models. ASUS, EVGA and Zotac have provided us with their models of the GeForce GTX 660 Ti, and we will be taking a close look at their performance.
NVIDIA does not provide a reference model of the GeForce GTX 660 Ti. This is why the three manufacturers supplied us with their respective cards. In the coming days and weeks, we expect to see more from these and other manufacturers, and we will confront them in a larger comparison.
NVIDIA has declared two goals for the GeForce GTX 660 Ti: to beat the Radeon HD 7870, and to at least attack the Radeon HD 7950's position. They want to do this with a 150 watts GPU, and with a price similar to the previous GeForce GTX 560, but it is definately meant to beat the former high-end model GeForce GTX 580.
Here is a summarizing table with the technical data:
|GeForce GTX 660 Ti model overview|
|Model||NVIDIA GeForce GTX 660 Ti||ASUS GeForce GTX 660 Ti DirectCU II TOP||EVGA GeForce GTX 660 Ti SuperClocked||Zotac GeForce GTX 660 Ti AMP! Edition|
|GPU||GK104 (GK104-300-KD-A2)||GK104 (GK104-300-KD-A2)||GK104 (GK104-300-KD-A2)||GK104 (GK104-300-KD-A2)|
|Manufacturing Process||28 nm||28 nm||28 nm||28 nm|
|Transistors||3,54 Billion||3,54 Billion||3,54 Billion||3,54 Billion|
|GPU Clock||915 MHz (Boost: 980 MHz)||1059 MHz (Boost: 1137 MHz)||980 MHz (Boost: 1059 MHz)||1033 MHz (Boost: 1111 MHz)|
|RAM Clock||1502 MHz||1502 MHz||1502 MHz||1652 MHz|
|RAM Size||2048 MB||2048 MB||2048 MB||2048 MB|
|RAM Interface||192 Bit||192 Bit||192 Bit||192 Bit|
|RAM Transfer Rate||144,2 GB/sec.||144,2 GB/sec.||144,2 GB/sec.||158,6 GB/sec.|
|Shader Units||1344 (1D)||1344 (1D)||1344 (1D)||1344 (1D)|
|Pixel Filling Rate||22 GPixel/Sec.||25,4 GPixel/Sec.||23,5 GPixel/Sec.||24,8 GPixel/Sec.|
|TDP||150 watts||150 watts||150 watts||150 watts|
To date, all new graphics cards of the GTX 600 series are using the GK104-GPU with Kepler architecture. The only exceptions are some low-end models in which the small Fermi models were simply renamed, and the GeForce GT 640, based on GK107. So we have a GK104 - manufactured in 28 nm and also outfitted with 3.54 billion transistors - operating the GeForce GTX 690 (twice in this one), the GeForce GTX 680, the GeForce GTX 670, and now the GeForce GTX 660 Ti, too. It's only when you look at the label more closely that the differences become obvious: the GeForce GTX 660 Ti has a version named GK104-300-KD-A2.
Technically, this GPU is closely related to the one operating the GeForce GTX 670. Both have 1344 shader processors, constituted by four Graphics Processing Clusters, which in turn are divided into seven SMX-cluster. We detailed the new architecture with the SMX clusters in our launch article on the GeForce GTX 680 (german test). This setup allows for 112 texture units and 24 ROPs. ROP's were cut down by NVIDIA in comparison to the GeForce GTX 670, which featured 32. The GPU clock amounts to 915 MHz, just like the GeForce GTX 660 Ti and the GeForce GTX 670. With GPU Boost, the GPU can operate at least at 980 MHz, like the model above it. These are NVIDIA's specifications; the vast majority of retail models will be more or less heavily overclocked. The GDDR5 memory operates at 1502 MHz, and comes with a 192 bit wide memory interface to achieve a bandwidth of 144.2 GB/sec. This is the second aspect with which NVIDIA wants to set the GeForce GTX 660 Ti apart from the GeForce GTX 670, since the latter has a 256 bit wide memory interface. The memory size ammounts to 2048 MB.
NVIDIA has a TDP of (Thermal Design Power) of 150 watts for the GTX 660 Ti, which is 20 watts lower than the GeForce GTX 670. So you need at least one 6 pin connector to supply the card with power. However, NVIDIA has decided to make two connections mandatory, in order to leave a little margin.
The first retail cards will not be very different from the reference version, only in GPU and memory clock speed, depending on the manufacturer. The GPU on the ASUS GeForce GTX 660 Ti DirectCU II TOP operates at a whopping 1072 MHz, an overclocking of more than 150MHz. With GPU Boost it even reaches 1137 MHz. Memory clock was left unchanged at the usual 1502 MHz. EVGA is more conservative with its GeForce GTX 660 Superclocked Ti: the GPU clock at 980 MHz, and GPU Boost up to 1059 MHz, again, RAM clock unchanged. Zotac on the other hand has a slightly different approach, and decided to overclock the GPU to 1033 MHz, and 1111 Mhz with GPU Boost. Unlike the first two manufacturers, Zotac decided to also overclock the RAM memory to 1652 MHz.
We will see these overclocking figures' repercussions on the benchmarks later on. But first we will have a closer look at the three competing models.