Nvidia's been getting a lot of mileage out of the same Kepler desktop chip that's used in both the GTX 670 and GTX 680 graphics card. The GTX 670 had one functional block disabled, but the new GTX 660 really only cuts out one key feature: one of the memory controller blocks. This reduces the data width to memory to 192 bits wide, versus the 256 bit pathways available on the GTX 680 and GTX 670. Since the ROP (raster operations) unit is part of that block, the number of ROP units are now 24 instead of 32. The clock rate is a little lower as well. See also: Nvidia GTX 660 Ti review – including SLI and 3-way SLI
eVGA GTX 660 Ti: Speeds and Feeds
Let's take a quick look at the speeds and feeds for the GTX 670 graphics card, and see just how it's different from its pricier cousin. Note that clock frequencies are for eVGA SC model, which is clocked slightly higher than Nvidia's reference card. Other shipping retail boards may differ in clock frequencies, depending on the design of the board and what the company shipping the board wants to support. The target price for the base GTX 660 Ti will be £249; the superclocked eVGA model we tested is priced at £264. See also: Group test: what's the best graphics card?
|
Feature |
eVGA GTX 660 |
GTX 670 |
GTX 680 |
|
Graphics Cores |
1,344 |
1,344 |
1,536 |
|
Texture Units |
112 |
112 |
128 |
|
ROPs |
24 |
32 |
32 |
|
Base Clock Frequency |
980 |
915 MHz |
1,006 MHz |
|
Boost Clock Frequency |
1059 |
980 MHz |
1,058 MHz |
|
Memory (Frequency) |
2GB (1,502MHz) |
2GB GDDR5 (1,502 MHz) |
2GB GDDR5 (1,502 MHz) |
|
Memory Interface |
192-bi |
256-bit |
256-bit |
|
Transistor Count |
3.5 billion |
3.5 billion |
3.5 billion |
|
Display Connectors |
2 x Dual Link DVI, 2 x HDMI 1.4a (Fast), 2 x DisplayPort 1.2 |
2 x Dual Link DVI, 2 x HDMI 1.4a (Fast), 2 x DisplayPort 1.2 |
2 x Dual Link DVI, 2 x HDMI 1.4a (Fast), 2 x DisplayPort 1.2 |
|
Power Connectors |
2 x 6-pin PCIe |
2 x 6-pin PCIe |
2 x 6-pin PCIe |
|
Thermal Design Power (TDP) |
150W |
170W |
195W |
That's the base specs in a nutshell. It's time to check out how the reference board actually performs.
eVGA GTX 660 Ti: Performance
Performance data was collected using FutureMark's 3DMark 2011 and four DirectX 11 games: Crysis 2, , DiRT3, Metro 2033 and Batman: Arkham City. Game tests were run at 1920 x 1200 pixel resolution with 4x multisampling anti-aliasing enabled. Also, system power under idle and load was measured using a Watts Up Pro power meter. The system used for testing consisted of a Core i7 3960X running at 3.3GHz with 12GB of DDR3 running at 1600MHz. Power supply, motherboard and other hardware remained constant during the entire run, as well as the operating system, Windows 7 Ultimate x64.
I compared performance to AMD's equivalently priced Radeon HD 7870 GHz Edition, in the guise of the Asus HD 7870 DirectCU II. I didn't have time to rerun all the benchmarks with the older GTX 560 Ti, but initial testing indicates that the 660 Ti is definitely a step up in performance.
The eVGA card easily bests wins this synthetic test. 3DMark 2011 isn't comparable to games, but does test most DirectX 11 features.
The GTX 660 wins out over the Radeon by about 10%, in Batman: Arkham Asyslum, which is a pretty substantial difference.
The difference in performance with Crysis 2 between the eVGA and Asus cards is smaller than in Arkham Asylum, but still worth noting.
Here, the roles are reversed slightly. The GTX 660 Ti is just a bit slower than the Radeon HD 7870 GHz Edition.
What gives? The Asus Radeon card pulls out a big win in this game over the GTX 660. In this case, however, it's not purely a graphics issue. DiRT Showdown uses the Microsoft's DirectCompute interface to perform some serious floating point calculations for two key feature: global illumination and advanced lighting. The AMD GPU is substantially better at pure computation than the Nvidia chip, which gives it an edge. Note that most games have yet to make heavy use of the GPU in general compute tasks, so it's unlikely you'll see big differences, and the GTX 660 tends to perform better in games which don't make heavy use of GPU compute.
It's impressive how idle power has improved on all the current generation cards. There's really little practical difference in power usage between these two cards.
