Performance Testing Part 1: Stock Clock Speeds
The test system is my 24/7 use computer, composed of the following bits at the moment:
| Motherboard: | Gigabyte Z77X-UD5H |
| CPU: | Intel Core i7 3770k |
| RAM: | G.Skill RipjawsX 2133MHz 7-10-7 |
| GPU: | Gigabyte GTX660Ti EVGA GTX580 (for comparison) |
| Storage: | OCZ Vertex 3 240GB MaxIOPS |
| PSU: | Antec HCP-850 |
| Case: | Thermaltake Armor Revo |
| CPU Cooler: | Phanteks PH-TC14PE |
Not a large card, the cooler (the fan, really) just barely extends past the motherboard.
For this section the ram is run with XMP enabled and CAS set to 8, everything else is on defaults/auto. Just to reiterate a bit, here is the GPUz window at stock:
We’re going to start with some competition benchmarks and then move into game tests. I’ll include GTX580 stock results on the synthetics for comparison purposes.
First up is the oldest benchmark:
3DMark03
3dMark03 is old, but despite that it makes full use of modern GPUs and isn’t nearly as CPU limited as many more modern benchmarks.
You can see the age in the FPS numbers, 654-1324! Mostly this is included for the competitive benchmarking crowd. We’re the only group that buys GPUs based on their performance in benchmarks from 2003! The GTX580 scores 94691. Chaulk one up for a massive improvement from Fermi to Kepler!
3DMark Vantage
Vantage is much more recent, it tests DX10 performance, though at a fairly low resolution of 1280×1024. It can use CUDA for Physx calculations as well. The first shot is with Physx turned on, the second shot it is disabled.
The Physx bits take over from the CPU, note the difference in CPU score!
The 580 gets 29830 with Physx, loosing to even the non-Physx GTX660 result. I gets 26277 without Physx enabled, again the GTX660 eats it for lunch.
3DMark11
3DMark11 is very new, it tests DX11 performance with fairly heavy tesselation.
This isn’t that far short of the GTX670, really. Closer than I expected. The GTX580 gets murdered, only making 6790
HWBot’s Unigine Heaven
HWBot made a wrapper for Unigine Heaven to force certain presets and detect tampering, it makes it nice standard. Heaven can test both DX9 performance and DX11 performance, so I did both. DX11 is first:
This is getting depressing for GTX580 owners, the 580 gets 1457 and 3116. It’s not even close to the GTX660 really.
Now for the game tests.
Skyrim
Skyrim turned out to be a dubious benchmark for this card. If you go over 60FPS there are serious graphics and physics issues with the game, it does not appreciate it at all. Hence it caps the frame rate at 60. Even with everything set to ultra and a ton of action going on the FPS never budged from 60FPS. The GTX580 does well too.
Including the screenshots would be pointless, suffice to say that this card is plenty for Skyrim at 1680×1050!
S.T.A.L.K.E.R. Call of Pripyat
I ran this benchmark on ultra details with everything turned up absolutely as high as it could go. My monitors max resolution is 1680×1050 and that is where I ran it.
Call of Pripyat isn’t the newest game out there, but they thought ahead with the graphics as you can see. It loads things down nicely, but the GTX 660 Ti chugs though.
Battlefield3
1680×1050 resolution, all graphical settings at maximum. I played a 32man deathmatch and used Fraps to record minimum, maximum and average frame rates. Here are the results:
55FPS is just fine, it falls slightly short of the sometimes-considered-a-standard 60FPS, but I certainly can’t tell the difference. The average of 84FPS is more than most monitors can display. All together this is quite good.
During testing the highest core temperature I recorded was a laughable 64c, the fan speed hit ~50% and the fans were very quiet. This is a nicely overpowered cooler for this card!
