Unigine Engine Is Back To Slamming GPUs, Drivers

While Unigine Valley is finally being released this week as a beautiful OpenGL (and Direct3D for Windows users) tech demo, it's very demanding on the system's graphics card and drivers. The open-source Mesa/Gallium3D drivers are out of the question and even the proprietary graphics drivers drop to their knees.

As mentioned in the exclusive article this past weekend on Phoronix about the Unigine Heaven 4.0 and Unigine Valley 1.0 releases, they are coming to the public on Tuesday and Thursday, respectively. These releases incorporate many visual improvements to the Unigine Engine, which (again) makes the engine capable of taxing more GPUs and drivers.

As noted already, the latest engine work basically makes anything running atop it off-limits to the current open-source Linux GPU drivers. Unigine is now depending upon an OpenGL 3.2 Core Profile context and this GL support is just too much for Mesa/Gallium3D to handle correctly right now. Hopefully by the Mesa release at the end of 2013, the open-source drivers will be in better standing...

As shown with screenshots in the Sunday article, the AMD Catalyst Linux binary driver is also hiccuping on the latest OpenGL renderer advancements. Give AMD a couple of Catalyst releases and they'll hopefully have Unigine fixes/improvements ready, just as they had done when dealing with earlier Unigine releases.

As mentioned on Twitter, If you're using the binary NVIDIA Linux graphics driver, at least it's a pretty good story. Assuming you're on a high-end GeForce graphics card, the performance should still be decent while rendering everything correctly.

This week will be in-depth Phoronix benchmarks showing off the performance of Unigine Heaven 4.0 and Unigine Valley 1.0. The comparison will be done with AMD and NVIDIA drivers on a wide selection of Radeon and GeForce graphics hardware. Until then, here are some reference numbers tonight. Benchmarking was done from an AMD FX-8350 system with a NVIDIA GeForce GTX 680 Kepler graphics card.

As soon as Unigine Corp makes publicly available the binaries, it's simply a matter of running phoronix-test-suite benchmark unigine-heaven unigine-valley to run these benchmarks on your Linux system. There's already the OpenBenchmmarking.org test profiles for Unigine Heaven and Unigine Valley that are ready and complete, pending the URLs for the new binaries.

All of the benchmark results being shown on this page of the Unigine Engine were done using the Phoronix Test Suite and can be easily reproduced. The system details, system logs, and other information about this AMD FX-8350 + NVIDIA GeForce GTX 680 setup running Ubuntu 13.04 can be found within its OpenBenchmarking.org result file: 1302111-FO-UNIGINESC08.

If you wish to see all of the graphs condensed for concisely showing the impact of the resolution scaling, here's the modified version. These OpenBenchmarking.org results also show the performance results on this system when using the older Unigine tech demo releases: Unigine Heaven 3.0, Unigine Tropics, and Unigine Sanctuary (their oldest publicly available tech demo).

For the Unigine Heaven 4.0 benchmarking, the frame-rate was 117 FPS for the GeForce GTX 680 when at 800 x 600 while when hitting 2560 x 1600 on the 30-inch Samsung LCD display the frame-rate was just 22 FPS. The average FPS across the wide range of resolutions was 63 FPS. In comparison, the average Unigine Heaven 3.0 frame-rate was 116 when using the same configuration file/settings. The Unigine Heaven 3.0 results are in the aforelinked result file.

Unlike the Unigine Heaven 4.0 results and the data from the other Heaven/Tropics/Sanctuary demos, on the high-end AMD FX-8350 system it was this "Bulldozer 2" processor that was the main bottleneck and not the NVIDIA GeForce GTX 680 graphics card. The frame-rate was fairly consistent about 50 FPS through 1920 x 1080, but when reaching 2560 x 1600 is where it fell to just above 30 FPS.

Stay tuned for the rest of the Unigine Engine test results from other drivers/hardware this week.