AMD R7 1800X Blender Benchmarks vs. i7-6900K, 7700K, 7600K, FX-8370
GN’s Andrew Coleman has several years of experience as a 3D animator, and so we are uniquely positioned to build our own rendering benchmarks and optimize render settings for the test. Blender does not work in its optimal state with out-of-box settings, and requires tuning tile size and compression settings for peak performance. CPUs, for instance, perform significantly better with smaller tile sizes (16x16), whereas GPUs tend to perform better in the 256x256 range, or thereabouts.
GN uses a custom-built Blender benchmark for its CPU rendering performance analysis. Here is a sample:
The benchmark mixes fur rendering (computationally intensive, particularly on GPUs), motion blur, ray tracing, multi-bounce light tracing, and various transparencies and mats.
In AMD’s marketing materials, the company used Blender for some of its initial tech day demonstrations that promised Intel’s 6900K – under some specific conditions – performing equally to AMD’s then-unnamed Ryzen CPU. AMD’s demonstration Blender benchmarks used different settings than what we would recommend. They were still deltas, so all is well in the world of comparisons, but it’s not an optimal test configuration for real Blender users. In its Blender testing, AMD executes renders using just 150 samples per pixel (what we consider to be “preview” quality), runs slightly unoptimized 32x32 tile sizes, and renders out at 800x800. In our benchmark, we render using 400 samples per pixel (release candidate quality), 16x16 tiles (faster for CPU rendering), and a 4K resolution. This means that our benchmarks are not comparable to AMD’s, but they are comparable against all the other CPUs we’ve tested. We also believe firmly that our benchmarks are a better representation of the real world.
Our Blender benchmark chart is below, and uses properly optimized benchmark settings (tile size, sample size, compression) for the test:
The AMD Ryzen R7 1800X completes the frame render in just under 28.73 minutes, with the Intel i7-6900K completing the render in 29.05 minutes. This puts AMD’s $500 Ryzen R7 1800X about on-par with the $1000 Intel i7-6900K for our Blender render task when both are in stock configuration. The R7 1800X is about 1-1.1% faster. Normally, this would be unsurprising or uninteresting, but the price makes AMD’s R7 platform worthy of further consideration.
Overclocked, we see the i7-6900K (~4.4GHz) reduce its render time to 25.35 minutes, while the Ryzen chip sits at 27.8 minutes. When overclocking is considered, AMD falls behind – but it’s still half the price, so consider both aspects.
As for other devices, the $340 i7-7700K (stock, HT1) completes the same render in 42.4 minutes, with the overclocked variant finishing in 38.3 minutes. We’re curious to see how the Ryzen R7 1700 holds up to the 7700K in this task.
For reference, the FX-8370 completes the render in 90.22 minutes, a far cry from today’s performance.
Blender is one of the few applications we tested where Ryzen shows its advantaged price-to-performance position, achieving mostly equal – slightly ahead, in some cases – render times to the $1000 6900K. This is because Blender is fully multithreaded, and tasks all threads to operate at full tilt until the render is done. Each tile of the scene is rendered by a single thread, and with 16 threads in flight, we see overall processing time reduce in a greater way than just increasing the frequency would do. This is demonstrated by looking between the 6900K and 7700K at 5.1GHz.
Of course, pushing it all to a GPU is normally much faster – a few times faster, in fact – making it tough to argue precisely when software accelerated rendering makes sense in Blender. Some specific elements of scenes will crunch better on a CPU, like heavy motion blur or (in previous versions of Blender) dense fur and hair. There’s potentially a valid use case in rendering from both ends of the animation and meeting in the middle, so to speak; that is, you might have the GPU render frame 1-1500, then have the CPU render from 1501-3000 (simultaneously).
Regardless, Ryzen holds a lead here, even if it does come down to a difference of 1.1% in an application that is normally (but not always) better GPU accelerated.
AMD R7 1800X Adobe Premiere Benchmark vs. i7-6900K, i7-7700K
Our video production workload is also a real-world test, pulling from files and export settings that we use every day for GN’s YouTube channel. The test is a render of our EVGA ICX review and includes dozens of audio, video, and image files loaded into it. Color correction, transforms, and other calculations (like Warp Stabilization) are included in the render, meaning that it is a precise representation of what we’d use Adobe Premiere CC for on a daily basis. This also means that the workload is computationally intensive, as the render is doing more than just encoding.
Our encoder settings are configured to 28Mbps target/max bitrate, with a “High” 5.1 profile. We are rendering at 1920x1080 and 60.0FPS (h.264).
Note: We only just added Adobe Premiere, so we’ve only got the three CPUs on the bench as of this review.
Configured to its stock clocks, the AMD R7 1800X completes the render task in 67.4 minutes, with the 6900K completing the video render in 68.5 minutes. If you prefer percentages, that’s a change of 1-2%. In rendering terms, you’d be saving about one minute for every hour of encoding. Ignoring all context, that’s not impressive, but context is what matters.
The context of Ryzen and the R7 1800X is that it’s a $500 CPU, and that high-end motherboards can be had for $255. Intel’s i7-6900K is $1000, and high-end motherboards cost $300 plus. Quad-channel memory is lost, but it’s also not all that beneficial in this specific test. The i7-7700K drags behind, completing the render nearly 40 minutes later than the 6900K and 1800X CPUs.
For Perspective: Adobe Premiere CUDA vs. CPU Benchmarks
That said, anyone running production renders knows that it’s the GPU that counts for most rendering tasks, despite some specific use cases where a high-end CPU can be beneficial.
For Premiere, we’d never use the CPU as the primary render workhorse. CUDA and OpenCL acceleration are significantly faster in supported applications.
This chart puts our render numbers into perspective, relying instead on the EVGA GTX 1080 FTW and CUDA for the render.
The result is expected: Our render times drop from around 60 minutes to 20 minutes. Rendering on the CPU takes 3x as long, and using a Ryzen CPU + GPU is not any faster than using a 6900K + GPU to render, as the heaviest work is still going to the GPU. Even aa GTX 1070 would be significantly faster.
AMD does deserve praise for accomplishing comparable render speeds to Intel’s high-end CPUs, but we must also frame that praise within the confines of reality: For the type of production rendering most people do, using a high-end GPU makes infinitely more sense than software rendering. That’s not to say that there aren’t use cases for CPU-based rendering, it’s just going to be on users in those situations to determine their needs.