OpenGL/OpenCL Performance & Perf-Per-Watt From NVIDIA's GeForce 9800GTX To GTX 1080

Written by Michael Larabel in Graphics Cards on 5 June 2016. Page 1 of 7. 7 Comments

Now that my initial GeForce GTX 1080 Linux review is out the door, I spent this weekend working on a "fun" comparison out of curiosity to see how the raw OpenGL and OpenCL performance has improved over the generations going back to the once-powerful GeForce 9800GTX plus including the top-end cards of the GeForce 600/700/900 Kepler and Maxwell series too.

Eight years ago, the GeForce 9800GTX (G92) was a beast with its 65nm GPU and 754 million transistors. The 9800GTX boasted a core clock of 675MHz and 2200MHz for its GDDR3 memory. The 9800GTX was rated for 648 single-precision GFLOPS and had a 140 Watt TDP. Fast forward to today, the GeForce GTX 1080 is fabbed on a 16nm process and has more than 7.2 billion transistors. The base core clock is 1607MHz (and 1733MHz boost) while having GDDR5X for its video memory. The GTX 1080 has a 180 Watt TDP and its single-precision rating is 8228~8873 GFLOPS. Quite impressive seeing how much more advanced Pascal is over hardware from eight years ago and is still being used by some.

The GeForce cards I used for this performance/perf-per-Watt comparison to complement yesterday's complete Linux review was the GeForce 9800GTX, GTX 460, GTX 680, GTX 780 Ti, GTX 980, GTX 980 Ti, GTX TITAN X, and GTX 1080. I was aiming for the high-end cards of each generation, based upon the cards I had available, which turned out fine except for the GTX 460 was the highest-end Fermi card I have available. All of the tests were done on the same Ubuntu 16.04 LTS system. The 9800GTX was tested with the 340.96 driver as it was the latest driver for supporting pre-Fermi hardware while the newer graphics cards were on the 367.18 beta driver that provides Pascal support.

The GeForce 9800GTX supports OpenGL 3.3 and OpenCL 1.1, which were thus the requirements for the games/benchmarks being tested today. Beyond that, they had to be on the lighter side too due to the 9800GTX having only 512MB of video memory. The tests ran on these new and old graphics cards under Linux were Insurgency, Dota 2, Team Fortress 2, Xonotic, Tesseract, OpenArena, Unigine Valley, and Unigine Heaven. For some OpenCL benchmarking, the compute tests still running on the 9800GTX were SHOC and JuliaGPU. The tests were done at 1920 x 1080 and 2560 x 1600 using a Samsung SyncMaster 30-inch dual-link DVI panel.

GeForce 9800GTX vs. GTX 1080

All of these benchmarks were done in a fully-automated and repeatable manner using the open-source Phoronix Test Suite software. The Phoronix Test Suite was monitoring the AC system power consumption using a WattsUp Pro USB power meter. Let's see how the performance and performance-per-Watt has advanced since the GeForce 9 days...

Related Articles