Testing Part One: Stock Clocks

Testing will be done in two parts as usual, first with stock GPU clocks and then another go-round overclocked. For both parts I have set my 3770k CPU to a standardized 4GHz. The follow equipment was used for testing:

It's a solid system, well suited for pushing the GTX670. For benchmarks I'll be using the following:

• 3DMark03, a fairly old benchmark still in competition use.
• 3DMark Vantage, a DX10 benchmark that stresses GPUs at 1280x1024.
• 3DMark11 Performance, a DX11 benchmark that stresses GPUs harshly at 720p.
• 3DMark11 Extreme, a DX11 benchmark that puts a massive load on the GPU at 1080p.
• HWBot Unigine Heaven DX9, a DX9 benchmark with a brutal GPU load.
• HWBot Unigine Heaven DX11, A DX11 benchmark at 1680x1050 that places a very heavy load on the GPU.
• S.T.A.L.K.E.R. Call of Pripyat, one of the nastier game benchmarks, also free.
• Battlefield 3 on ULTRA settings at 1080p, fast paced and brutal.

First up, the CPUz and GPUz windows

The GPUz window gives us a full rundown of the card, though the "Boost" clocks aren't actually the peak boost clocks. This is rather higher than the reference design's 915MHz base and 980MHz boost clocks, though not as huge a jump as the 660Ti OC had. Hopefully this GTX670 will make good use of the wider memory bus and stomp its smaller sibling into the ground.

3DMark03

105k isn't a world record, nor is it huge for this generation of cards. It does beat the 660Ti easily. Kepler based cards (the GTX600 series) aren't very good for 3D03, it uses a very old DirectX that they don't know what to do with.

3DMark Vantage

32.7k in vantage at stock is tasty, also significantly higher than the GTX660Ti and in the same ballpark as the 7970. Note that this was done with PhysX turned off.

3DMark11 Performance

Clearly the extra memory bandwidth helps a lot here. This is also significantly higher than some 7970s. Do note that drivers can make a big difference in 3DMark11, AMD's recent drivers gave a pretty massive performance boost somehow, combined with the GHz edition cards they are running at this speed or higher.

3DMark11 Extreme

This may well be the most brutal of the benchmarks, note the average FPS in the four GTs (GameTests)! There is a lot of detail and a lot of tessellation in this benchmark.

HWBot Unigine Heaven DX9 and DX11

In DX9 and 1280x1024, this card brutalizes the benchmark. It's way overpowered for that resolution.

In DX11 at 1680x1050 the GTX670 still cruises right along. For reference the GTX660Ti got 1624 here. What a difference memory bandwidth makes! It's also on par with or faster than most 7970s.

S.T.A.L.K.E.R. Call of Pripyat

1080p, ULTRA settings.

This can't be compared vs the GTX660 as I didn't have 1080p capabilities at that point. That said, the GTX670 wins at the higher resolution anyway in everything but night.

Battlefield 3

1080p, ULTRA settings. I ran through the first campaign level with FRAPS as it makes a nice reproducible test ground. Large multiplayer maps can be a higher load, but the load is extremely variable, making it a dubious test ground in my opinion.

This can't be compared vs the 660Ti results either, though the minimum FPS is higher despite the higher resolution. 57FPS as a minimum on ULTRA is quite nice, plenty for gaming. The GTX670 is on par with or a bit slower than a 7970, depending on the factory OC of the 7970. Similarly priced 7970s lose to the GTX670 OC for the most part.

Let's see how far this GPU will overclock. The 660Ti OC'd like mad, I'm hoping this GTX670 will do similarly. After that, I'll sum up the performance.

Newer

Older