NVIDIA GeForce RTX 4070 Ti Bests An RTX 3090 Ti In Octanebench Leak
Otoy’s Octanerender is a plug-in renderer available for lots of software suites, including Blender, Cinema4D, Unity, and Unreal just to name a few. It’s a full-on cinematic-quality ray-tracer that runs entirely on the GPU without sacrificing quality. Naturally, that makes it an excellent GPU performance benchmark. Like a lot of benchmarks, Otoy maintains an “Octanebench” result database, and the first GeForce RTX 4070 Ti result has popped up there. Spoilers: it’s faster than every single other non-Ada part.
That includes, of course, all kinds of fancy professional visualization parts, like the RTX A6000 and the A100, but it also includes the ludicrously-expensive and power-thirsty GeForce RTX 3090 Ti. The leaked result, which scored 725 points on the OctaneBench test, is only one test against many, but it’s comfortably ahead of the average RTX 3090 Ti result and just a scant 5 points behind the highest-ever RTX 3090 Ti benchmark.
Going by the specifications of the two GPUs in their standard configurations, the RTX 4070 Ti (née RTX 4080 12GB) should put out 40.1 TFLOPS in single-precision FP32 compute, which puts it essentially on equal footing with the GeForce RTX 3090 Ti’s rated 40 TFLOPs. As a result, it shouldn’t really be any surprise that the newer card with improved efficiency pulls ahead of the older model, despite the deficit in terms of shader count—7680 versus 10,752.
We wouldn’t expect gaming results to necessarily go the same way. While the GeForce RTX 4080 has a comfortable lead on the GeForce RTX 3090 Ti in gaming, the GeForce RTX 4070 Ti sacrifices an additional 213 GB/sec of memory bandwidth. That puts it at exactly half the memory bandwidth of the last-gen flagship, and that’ll be a tough hill to climb for most games, especially in higher resolutions.
The relatively narrow memory buses of the Ada GPUs have been a point of concern for enthusiasts ever since the specifications first leaked for the new cards, but so far those concerns have been completely unwarranted. Still, you can tell NVIDIA itself thought it might be a pain point given that the GeForce RTX 4080 uses hot-clocked 22.4 Gbps GDDR6X memory, even faster than the 21 Gbps on the RTX 4090. We’ll see how the RTX 4070 Ti holds up when it comes to market; it’s rumored for a CES 2023 release—just ten days away.