Maxon Releases New R20 Version of Cinebench Benchmark

Maxon’s Cinebench 4D benchmark has been a mainstay of CPU benchmarking for nearly two decades. While it’s never been the only rendering benchmark that we run at ET, it’s a simple, easy way for gamers and enthusiasts to check CPU performance in a free application tied back to a real-world rendering engine. With that said, the previous version of the test, CB15, had been getting more than a little long in the tooth. Today, the company dropped a new version of the test, CB20, to address these issues.

The two tests are not directly comparable to one another; Maxon notes that “Cinebench R20 uses a much larger and more complex test scene than R15, requiring about 8x the computational power needed to render it. The test also requires about 4x the memory. Therefore, R15 and R20 results cannot be compared.” It’ll be interesting to see how Cinebench 20 impacts the relative standing of Intel versus AMD, and we’ll be including both tests for a little while as we bring the new benchmark up.

The new benchmark no longer includes a GPU test (no loss, no one used it), but it does take advantage of Intel’s new Embree ray tracing technology. Embree isn’t something we’ve covered before. It’s a set of new ray tracing kernels developed by Intel, intended to speed execution of these workloads on x86 CPUs by using SSE, AVX, and AVX-512 more efficiently. The project is actively maintained on Github and a presentation on the tech has been uploaded to SlideShare.

Any time a company announces that it’s implementing an Intel-designed technology, there will be concerns from AMD users about whether test treats CPUs fairly. We’ve reached out to AMD to see if the company has any comments, but I want to stress that we’ve never seen any evidence that Cinebench has been gamed in any fashion.

I’m not just saying that. I once tested every version of Cinebench ever released, from the 2000 or 2001 variant I had archived (I don’t recall the exact year), up to and including CB15. There’s a Python script, developed by Mark Mackey, that can be used to confirm whether an application was compiled in a way that artificially favors Intel over AMD by refusing to run SIMD code on an AMD chip that should otherwise support it. There’s never been any evidence of shenanigans in the Cinebench benchmark code, even during the exact period when we know some of these strategies were being used in other contexts.

I’ll update this article with any response from AMD, but there’s no reason to think this benchmark has been artificially gamed that I’m aware of. It’s also worth noting that the performance comparisons for CB20 echo the standings we would’ve expected in CB15, with mild (but not huge) variation in test results. With new CPU architectures on the way from both AMD and Intel in the next 12 months, we’re glad to see benchmark authors updating their own software platforms. For now, the test is only available from the Windows Store and Mac App Store.

Continue reading

Apple’s M1 Continues to Impress in Cinebench R23, Affinity Photo

New Cinebench R23 benchmarks paint AMD in a more competitive light against the M1, but Apple's SoC still acquits itself impressively. The Affinity Photo benchmark, however, is a major M1 win.