Page 1: Test: NVIDIA GeForce GTX 660
With the GeForce GTX 660 Ti (Hardwareluxx Review), NVIDIA recently entered the middle class segment with its new "Kepler" generation GPUs. Now they follow up with their next hit: this time they want to address budget-conscious buyers, and release the GeForce GTX 650 and 660. The GeForce GTX 660 is rumored to have enough gaming capability. We take a closer look at two of the card's variants.
Graphics cards manufacturers such as AMD and NVIDIA are interested only in certain price ranges: only the zealous will spend 400 euros on the spur of the moment. But the quantities bought in this price range are so small that a certain sales volume must be reached before break-even. These "sweat spots" are price ranges where buyers will go for it more easily. One of these sweat spots can be found between 250 to 300 euros; this range is already covered by the GeForce GTX 660 Ti. With the GeForce GTX 660, NVIDIA wants to enter the region of 200 euros and below.
NVIDIA provides a reference version of the card, though most manufacturers will merely take over the design of PCB. Our two test candidates have an alternative cooling system and are overclocked ex factory.
We summarised the technical data in this table:
|GeForce GTX 660 Series Overview|
|Model||NVIDIA GeForce GTX 650||NVIDIA GeForce GTX 660||NVIDIA GeForce GTX 660 Ti|
|Street Price||-||as from 225 Euro||as from 299 Euro|
|GPU||GK106 (GK106-400-A1)||GK106 (GK106-400-A1)||GK104 (GK104-300-KD-A2)|
|Manufacturing Process||28 nm||28 nm||28 nm|
|Transistors||2,54 Billion||2,54 Billion||3,54 Billion|
|GPU Clock||1058 MHz||980 MHz (Boost: 1033 MHz)||915 MHz (Boost: 980 MHz)|
|RAM Clock||1250 MHz||1502 MHz||1502 MHz|
|RAM Size||1024 / 2048 MB||2048 MB||2048 MB|
|RAM Interface||128 Bit||192 Bit||192 Bit|
|RAM Bandwidth||80 GB/Sek.||144,2 GB/s||144,2 GB/s|
|Shader Units||384 (1D)||960 (1D)||1344 (1D)|
|Pixel Filling Rate||22 GPixel/s|
|TDP||64 watts||140 watts||150 watts|
The GeForce GTX 650 and 660 are the first two cards based on the smallest yet "Kepler" design, the GK106. It's manufactured in 28 nm, but due to the cutbacks in some areas of the card it only provides 2.54 billion transistors. In the case of the GeForce GTX 650, the GPU provides a Graphics Processing Cluster with two SMX clusters, resulting in 384 CUDA cores, 32 texture units and 16 ROPs. RAM (1024 or 2048 MB GDDR5) memory is connected with a 128 bit interface. NVIDIA specifies a base clock of 1058 MHz; this version of "Kepler" has no GPU Boost. Memory is being addressed at 1250 MHz. The potential performance of the GeForce GTX 650 can be estimated by taking the thermal design power of 64 watts into consideration. It is obviously an entry-level card with no major ambitions.
The GeForce GTX 660 is also based on the GK106 GPU - the number of the transistors is identical to the smaller model, but it uses three Graphics Processing Clusters, for a total of five SMX clusters. The number of CUDA cores (960) is significantly higher than the GeForce GTX 650's, as is the number of texture units, of which the GeForce GTX 660 has 80. The number of ROPs (24) is identical to the 660 GTX Ti, and they are connected with the same 192bit memory interface. NVIDIA specifies a base clock of 980 MHz, with GPU Boost you get at least 1033 Mhz. Memory runs at 1502 MHz, a rather high clock speed. NVIDIA specifies a total of 2048 MB of GDDR5 RAM. With a TDP of 140 watts, the GeForce GTX 660 is very close to the Ti model.
In the following table we compare the two cards, that provided by the Board Partners.
|GeForce GTX 660 Model Overview|
|Model||ASUS GeForce GTX 660 DirectCU II TOP||Zotac GeForce GTX 660 AMP! Edition|
|Street Price||-||249 Euro|
|Manufacturing Process||28 nm||28 nm|
|Transistors||2,54 Billion||2,54 Billion|
|GPU Clock||1072 MHz (Boost: 1137 MHz)||1111 MHz (Boost: 1176 MHz)|
|RAM Clock||1527 MHz||1652 MHz|
|RAM Size||2048 MB||2048 MB|
|RAM Interface||192 Bit||192 Bit|
|RAM Bandwidth||146,6 GB/s||158,6 GB/s|
|Shader Units||960 (1D)||960 (1D)|
|Pixel Filling Rate||25,7 GPixel/s||26,7 GPixel/s|
|TDP||140 watts||140 watts|
In addition to the alternative cooling, which we'll detail later on, both manufacturers provide an ex factory overclock. Of the original 980 MHz, the ASUS GeForce GTX 660 DirectCU II TOP goes up to 1072 MHz. GPU Boost even allows for 1137 MHz and more. ASUS has the memory overclocked by 25 MHz to 1527 MHz. The remaining specifications are in sync with architectural requirements by NVIDIA, or have been adjusted by the clock speed.
The Zotac GeForce GTX 660 AMP! Edition uses a slightly higher clock: 1111 MHz, a value well above NVIDIA's specifications of 980 MHz. Using GPU Boost, the GPU runs with at least 1176 MHz under load. The GDDR5 memory modules run at a higher clock rate, too: 1652 MHz, an increase of 150 MHz. The values for memory bandwidth and pixel filling rate have been adjusted accordingly. For the rest we find the same GK106 GPU specifications of the GeForce GTX 660.
We will see how is this reflected in the overclocking performance during the benchmarks. First things first, however: let's look at the two variants in more detail.