We'll begin with a series of synthetic tests aimed at exposing the true, delivered throughput of the GPUs. In each instance, we've included a table with the relevant theoretical rates for each solution, for reference.
|Radeon HD 5870||27||68/34||154|
|Radeon HD 6970||28||85/43||176|
|Radeon HD 7970||30||118/59||264|
|Radeon R9 280X||32||128/64||288|
|Radeon R9 290X||64||176/88||320|
|GeForce GTX 770||35||139/139||224|
|GeForce GTX 780||43||173/173||288|
|GeForce GTX Titan||42||196/196||288|
Although the 290X has, in theory, much higher fill capacity than the Titan, this test tends to be limited more by memory bandwidth than anything else. None of the GPUs achieve anything close to their peak theoretical rates. The 290X's additional ROP power will more likely show up in games using multisampled anti-aliasing.
The back-and-forth here is kind of intriguing. 3DMark's texture fill test isn't filtered, so it's just measuring pure texture sample rates, and the Titan manages to outperform the 290X in that test. The results from the Beyond3D test tool are bilinearly filtered, and in the first of these, the 290X takes the top spot.
Once we get into higher-precision texture formats, a major architectural difference comes into play. Hawaii and the other Radeons can only filter FP16 texture formats at half the usual rate. Even the GK104-based GTX 770 is faster than the 290X with FP16 and FP32 filtering.
In all cases, though, the 290X offers a nice increase over the Radeon R9 280X—which is just a re-branded Radeon HD 7970 GHz Edition, essentially.
Tessellation and geometry throughput
|Radeon HD 5870||0.9||154|
|Radeon HD 6970||1.8||176|
|Radeon HD 7970||1.9||264|
|Radeon R9 280X||2.0||288|
|Radeon R9 290X||4.0||320|
|GeForce GTX 770||4.3||224|
|GeForce GTX 780||3.6 or 4.5||288|
|GeForce GTX Titan||4.4||288|
I'm not sure what to make of these results. I expected to see some nice gains out of the 290X thanks to its higher rasterization rates, but the benefits are only evident in TessMark's x16 subdivision mode and with our low-res/extreme tessellation scenario in Unigine Heaven.
A couple of potential explanations come to mind. One, TessMark uses OpenGL, and it's possible AMD hasn't updated its OpenGL drivers to take full advantage of Hawaii's quad geometry engines. Two, the drivers could be fine, and we could be seeing an architectural limitation of the Hawaii chip. As I noted earlier, large amounts of geometry amplification tend to cause data flow problems. It's possible the 290X is hitting some internal bandwidth barrier at the x32 and x64 tessellation levels that's common to GCN-based architectures. I've asked AMD to comment on these results but haven't heard back yet. I'll update this text if I find out more.
|Radeon HD 5870||2.7||154|
|Radeon HD 6970||2.7||176|
|Radeon HD 7970||3.8||264|
|Radeon R9 280X||4.1||288|
|Radeon R9 290X||5.6||320|
|GeForce GTX 770||3.3||224|
|GeForce GTX 780||4.2||288|
|GeForce GTX Titan||4.7||288|
Welp. This one's unambiguous. That massive GCN shader array is not to be denied. The 290X wins each and every shader test, sometimes by wide margins.
Now, let's see how these things translate into in-game performance.