NVIDIA GeForce GTX 660 Ti Round-Up Review

Introduction and Specifications

NVIDIA continues to flesh out their desktop GPU line-up today, with yet another high-performance graphics card based on the Kepler microarchitecture. To date, we’ve seen the GeForce GTX 670, GTX 680, and monstrous GTX 690 come to market, all sporting GK104 GPUs in one form or another. The new GeForce GTX 660 Ti we’ll be showing you here, although more affordable than its higher-end counterparts, continues the trend and also sports a GK104. As you probably expect though, certain portions of the GPU on the GeForce GTX 660 Ti are disabled or non-functional, so it’s not quite as capable as its more powerful brethren in the performance department. Its feature set, however, is identical.

One of the advantages of releasing multiple graphics cards built around the same GPU is that NVIDIA’s board partners have had lots of experience with the chip. And with experience comes customization. NVIDIA’s board partners were so comfortable with the GeForce GTX 660 Ti right out of the gate that every one of the retail-ready boards were received were custom options, which were all overclocked right from the factory. Although you will likely find many GeForce GTX 660 Ti cards based strictly on NVIDIA’s reference design in the market at somewhat lower price points, a multitude of customized options will be available as well, four of which we’ll be showing you here today.

There’s also been some activity on the AMD Radeon front this past week that’s caused quite a stir. AMD has made a firmware update available for its Radeon HD 7950 that increases its based GPU clock and adds support for Boost, which was first introduced on the Radeon HD 7970 GHz Edition. We have included numbers for the Radeon HD 7950 with and without Boost support on the pages ahead to provide as much performance data as possible and paint the clearest picture of the landscape, but be aware that our numbers come by way of a reference card flashed with the latest firmware and not one of the newer, customized cards due to hit retail shelves in the coming days.

Back on topic, we’ve got the main features of specifications of the GeForce GTX 660 Ti in the table below along with a few links to recent, related articles and some details on the changes to GK104 as it’s configured on the GTX 660 Ti. Later we’ll show you GeForce GTX 660 Ti cards from EVGA, MSI, Gigabyte, and ZOTAC and provide plenty of performance data for you speed freaks out there...


Custom GeForce GTX 660 Ti Cards: Different Coolers, Different PCBs, and Different Clocks. Same GPU.

NVIDIA GeForce GTX 660 Ti
Specifications & Features
Processing Units
Graphics Processing Clusters 4
SMXs 7
CUDA Cores 1344
Texture Units 112
ROP Units 24
Clock Speeds
Base Clock 915 MHz
Boost Clock 980 MHz
Memory Clock (Data Rate) 6008 MHz
L2 Cache Size 384KB
Memory
Total Video Memory 2048MB
Memory Interface 192-bit
Total Memory Bandwidth 144.2 GB/s
Texture Filtering Rate (Bilinear) 102.5 GigaTexels/sec
Physical & Thermal
Fabrication Process 28 nm
Transistor Count 3.54 Billion
Connectors 2 x Dual-Link DVI, 1 x mini-DP, 1 x HDMI
Form Factor Dual Slot
Power Connectors 2 x 6-pin
Recommended Power Supply 450 watts
Thermal Design Power (TDP) 150 watts
Thermal Threshold 98° C


From left to right: EVGA, MSI, Gigabyte, and ZOTAC

As the specifications in the table above show, the new GeForce GTX 660 Ti is based on the very same GK104 GPU that powers the higher-end GeForce GTX 690, GTX 680 and GTX 670 cards. On the GeForce GTX 680, the GK104 is fully-equipped, with all of its cores and memory partitions enabled (The GeForce GTX 680 has four Graphics Processing Clusters, with eight SMXes, 1536 CUDA cores, 128 texture units, and 32 ROPs). On the GeForce GTX 670, one of the GK104's SMX units is disabled, reducing the number of active cores to 1344 and bringing the number of texture units down to 112, but memory bus width and other features remain unchanged. With the GeForce GTX 660 Ti, NVIDIA takes things a step further and disables one of the GK104's memory partitions. As such, the GeForce GTX 660 Ti has the same core and texture unit configuration as the GeForce GTX 670, bit with a narrower 192-bit memory bus and only 24 ROPs. Memory capacity remains at 2GB on reference cards, but the narrower bus results in 144.2GB/s of peak memory bandwidth, down from 192.2GB/s on the GTX 670 and 680.


GK104 "Kepler" GPU Die Shot

Before we give you the rest of the scoop on the GeForce GTX 660 Ti, we want to direct your attention to a few other HotHardware articles related today's launch...

We've already covered the details of the GK104 GPU powering the GeForce GTX 660 Ti in our GeForce GTX 680 launch article, including things like GPU Boost, NVENC (NVIDIA Encoding Engine), and TXAA, so we won't go over them again here. If you'd like a deeper look into the technologies employed in the GK104, we'd definitely suggest taking a gander at that piece. The details of NVIDIA's brand new dual-GPU flagship are covered in our GeForce GTX 690 article, and our 3D Vision 2 and 3D Vision Surround articles cover those two NVIDIA technologies, which are obviously supported in a card like the GeForce GTX 660 Ti.
 


Related content