Techgage logo

Best CPU for Rendering & Video Encoding: Spring 2021

AMD Ryzen and Intel Core Product Packaging

Date: March 10, 2021
Author(s): Rob Williams

We’re taking a fresh look at workstation performance revolving (almost) entirely around the CPU. With many encoding and rendering tests in-hand, we’re exploring different performance angles with the help of our CPU collection which includes models ranging from six- to 64-cores. What matters more; core-count or clock speeds?



Introduction

It’s been a little while since we last dove into CPU workstation testing in a meaningful way, so we’re going to get things up to speed here, and see where things stand in early 2021. With Intel’s 11th-gen Core processors set to land soon, we’ll follow-up on this article after that series’ release to see where Intel’s latest fits into the mix.

In this article, we’re going to take an in-depth look at CPU performance revolving around many different workloads, with encoding- and rendering-type tests representing the bulk. If you’ve read through our CPU performance evaluations before, you’ll recognize many of the tests, but as always, all of our software was updated to the latest versions before getting down to retesting.

For those interested in CPU performance in Linux, we’re planning to follow-up on this article in the near-future with comprehensive tests there. As this article focuses on workstation performance, gamers can check out our semi-recent look at performance on four platforms. We’ve been feeling an itch lately to do more comprehensive game tests, so we plan to revisit that soon, as well.

AMD and Intel Processor Boxes

The tables below give a quick run-down of both AMD’s and Intel’s current line-ups. Bear in mind that not all models are listed in every case, but the most crucial ones are.

AMD’s Ryzen & Ryzen Threadripper Lineup
Cores Clock (Turbo) Cache Memory IGP TDP Price
Ryzen Threadripper
3990X 64 (128T) 2.9 GHz (4.3) 288MB Quad No 280W $3990
3970X 32 (64T) 3.7 GHz (4.5) 144MB Quad No 280W $1999
3960X 24 (48T) 3.8 GHz (4.5) 140MB Quad No 280W $1399
Ryzen 9
R9 5950X 16 (32T) 3.4 GHz (4.9) 72MB Dual No 105W $799
R9 5900X 12 (24T) 3.7 GHz (4.8) 70MB Dual No 105W $549
Ryzen 7
R7 5800X 8 (16T) 3.8 GHz (4.7) 36MB Dual No 105W $449
Ryzen 5
R5 5600X 6 (12T) 3.7 GHz (4.6) 35MB Dual No 65W $299
Ryzen 3
R3 3300X 4 (8T) 3.8 GHz (4.3) 18MB Dual No 65W $120
R3 3100 4 (8T) 3.6 GHz (3.9) 18MB Dual No 65W $99
Ryzen w/ Radeon Vega Graphics
R5 3400G 4 (8T) 3.7 GHz (4.2) 0.5+4MB Dual Yes 65W $149
R3 3200G 4 (4T) 3.6 GHz (4.0) 0.5+4MB Dual Yes 65W $99
Intel Core & Core X Processor Lineup
Cores Clock (Turbo) Cache Memory IGP TDP Price
Core X-Series
i9-10980XE 18 (36T) 3.0 GHz (4.8) 24.75MB Quad No 165W $979
i9-10940X 14 (28T) 3.3 GHz (4.8) 19.25MB Quad No 165W $784
i9-10920X 12 (24T) 3.5 GHz (4.8) 19.25MB Quad No 165W $689
i9-10900X 10 (20T) 3.7 GHz (4.7) 19.25MB Quad No 165W $590
Core Series
i9-10900K 10 (20T) 3.7 GHz (5.3) 20MB Dual Yes 125W $488
i9-10900 10 (20T) 2.8 GHz (5.2) 20MB Dual Yes 65W $439
i7-10700K 8 (16T) 3.8 GHz (5.1) 16MB Dual Yes 125W $374
i7-10700 8 (16T) 2.9 GHz (4.8) 16MB Dual Yes 65W $323
i5-10600K 6 (12T) 4.1 GHz (4.8) 12MB Dual Yes 125W $262
i5-10600 6 (12T) 3.3 GHz (4.8) 12MB Dual Yes 65W $213
i5-10500 6 (12T) 3.1 GHz (4.5) 12MB Dual Yes 65W $192
i5-10400 6 (12T) 2.9 GHz (4.3) 12MB Dual Yes 65W $182
i3-10320 4 (8T) 3.8 GHz (4.6) 8MB Dual Yes 65W $154
i3-10300 4 (8T) 3.7 GHz (4.4) 8MB Dual Yes 65W $143
i3-10100 4 (8T) 3.6 GHz (4.3) 6MB Dual Yes 65W $122

While this article will focus on the overall performance to be derived from our tested CPUs, it’s really difficult to sum-up a single chip and have those thoughts apply to everyone. Our workloads are all different, so you need to analyze yours to some extent to gauge what exactly is important to you.

As more rendering workloads have become so darn fast on our GPUs, many will be best-suited with a smaller core-count CPU that has great clocks over a many core-count CPU with modest clocks. The better or more efficient a CPU’s clock speed, IPC (instructions-per-clock), and general single-threaded performance is, the faster responsiveness will be – from the OS to the software within.

If you’re in need of many cores, either because of heavy rendering, encoding, or perhaps even the tendency to run many virtual machines, it will be worth sacrificing a bit of clock speed just to gain plenty of breathing room. Or, in the case of Core X or Ryzen Threadripper, users would also gain the advantage of a quad-channel memory controller for greatly improved bandwidth.

We have a lot to cover, so we’ll dive straight into things on the next page. First up is a look at our testing methodology, as well as detailed configuration specs. If you don’t care about the methods to our madness, you can skip right to page three to kick things off with a look at rendering performance.

Test Methodology & Systems

Benchmarking a CPU may sound like a simple enough task, but in order to deliver accurate, repeatable results, strict guidelines need to be adhered to. This makes for rigorous, time-consuming testing, but we feel that the effort is worth it.

This page exists so that we can be open about how we test, and give those who care about testing procedures an opportunity to review our methodology before flaming us in the comments. Here, you can see a breakdown of all of our test machines, specifics about the tests themselves, and other general information that might be useful.

Let’s start with a look at the test platforms. For testing AMD’s Ryzen, we use ASRock’s X570 TAICHI motherboard, while for Threadripper, we use ASUS’ Zenith II Extreme Alpha. On the Intel side, Core chips are tested in ASUS’ ROG STRIX Z390-E GAMING, while Core X uses the ASUS ROG STRIX X299-E GAMING.

The same 64GB kit of Corsair DOMINATOR memory is used on each one of the test platforms. After XMP is applied, the DRAM speed is reduced from DDR4-3600 to DDR4-3200. All of the tested CPUs and their respective platforms should be able to run DDR4-3600 memory speeds without issue, but we’ve encountered enough time-wasting behavior in the past to simply stick to 3200.

Neither of the AMD motherboards we use for testing apply an automatic overclock to the CPUs, and for the sake of testing reference clocks, Precision Boost Overdrive is left disabled. On the Intel motherboards, ASUS’ MultiCore Enhancement acts as an automatic overclock, so after enabling XMP, we change the option to enforce reference clocks. Both PBO and MCE will deliver a decent improvement to performance at the expense of additional power draw and higher temperatures, but it’s worth considering if your PC’s cooling is up to the task.

We used the latest version of Windows 10 (20H2) for our testing, ensuring that the latest bits are pulled in by Windows Update. We also ensure that the platform-specific chipset driver is up-to-date, and in all cases, the settings that driver applies are left alone.

All of the tests in this article were run with the help of our automated test scripts which ensure the same settings and actions are applied each and every time. This generally means we don’t need to run a test too many times to generate an accurate and repeatable result, but generally speaking, most tests are run at least 3 times over.

Here’s the full breakdown of the test platforms:

Techgage’s CPU Testing Platforms

AMD AM4 Test Platform
Processors AMD Ryzen 9 5950X (3.4GHz, 16C/32T)
AMD Ryzen 9 5900X (3.7GHz, 12C/24T)
AMD Ryzen 7 5800X (3.8GHz, 8C/16T)
AMD Ryzen 5 5600X (3.7GHz, 6C/12T)
Motherboard ASRock X570 TAICHI
CPUs tested with BIOS P4.00 (Jan 19, 2021)
Memory Corsair VENGEANCE (CMT64GX4M4Z3600C16) 16GB x4
Operates at DDR4-3200 16-18-18 (1.35V)
Graphics NVIDIA RTX 3070 (8GB; GeForce 460.89)
Storage WD Blue 3D NAND 1TB (SATA 6Gbps)
Power Supply Corsair RM850X (850W)
Chassis Fractal Design Define C
Cooling Corsair Hydro H100i PRO RGB (240mm)
Et cetera Windows 10 Pro (20H2, Build 19042)
As tested configuration: AMD Ryzen 9 5950X
As tested configuration: AMD Ryzen 9 5900X
As tested configuration: AMD Ryzen 7 5800X
As tested configuration: AMD Ryzen 5 5600X

AMD TRX40 Test Platform
Processor AMD Ryzen Threadripper 3990X (2.9GHz, 64C/128T)
AMD Ryzen Threadripper 3970X (3.7GHz, 24C/48T)
AMD Ryzen Threadripper 3960X (3.8GHz, 32C/64T)
Motherboard ASUS Zenith II Extreme Alpha
CPUs tested with BIOS 1303 (Nov 11, 2020)
Memory Corsair VENGEANCE (CMT64GX4M4Z3600C16) 16GB x4
Operates at DDR4-3200 16-18-18 (1.35V)
Graphics NVIDIA RTX 3070 (8GB; GeForce 460.89)
Storage WD Blue 3D NAND 1TB (SATA 6Gbps)
Power Supply Cooler Master Silent Pro Hybrid (1300W)
Chassis NZXT H710i
Cooling NZXT Kraken X63 (280mm)
Et cetera Windows 10 Pro (20H2, Build 19042)
As tested configuration: AMD Ryzen Threadripper 3990X
As tested configuration: AMD Ryzen Threadripper 3970X
As tested configuration: AMD Ryzen Threadripper 3960X

Intel LGA1151 Test Platform
Processors Intel Core i9-10900K (3.7GHz, 10C/20T)
Intel Core i5-10600K (4.1GHz, 6C/12T)
Motherboard ASUS ROG Maximus XII HERO Wi-Fi
CPUs tested with BIOS 2004 (Jan 13, 2021)
Memory Corsair VENGEANCE (CMT64GX4M4Z3600C16) 16GB x4
Operates at DDR4-3200 16-18-18 (1.35V)
Graphics NVIDIA RTX 3070 (8GB; GeForce 460.89)
Storage WD Blue 3D NAND 1TB (SATA 6Gbps)
Power Supply EVGA Bronze 600B1 (600W)
Chassis Corsair Crystal X570 RGB
Cooling Corsair Hydro H115i PRO RGB (280mm)
Et cetera Windows 10 Pro (20H2, Build 19042)
As tested configuration: Intel Core i9-10900K
As tested configuration: Intel Core i5-10600K

Intel LGA2011-3 Test Platform
Processors Intel Core i9-10980XE (3.0GHz, 18C/36T)
Motherboard ASUS ROG STRIX X299-E GAMING
CPU tested with BIOS 3301 (Nov 5, 2020)
Memory Corsair VENGEANCE (CMT64GX4M4Z3600C16) 16GB x4
Operates at DDR4-3200 16-18-18 (1.35V)
Graphics NVIDIA RTX 3070 (8GB; GeForce 460.89)
Storage WD Blue 3D NAND 1TB (SATA 6Gbps)
Power Supply Corsair Gold AX1200 (1200W)
Chassis Corsair Carbide 600C
Cooling NZXT Kraken X62 (280mm)
Et cetera Windows 10 Pro (20H2, Build 19042)
As tested configuration: Intel Core i9-10980XE

Testing Considerations

As covered earlier, we use an up-to-date Windows 10 (20H2) for our testing, as well as the latest chipset driver for each respective platform. In the pursuit of accurate, repeatable benchmarks, here are some basic guidelines we follow:


Encoding Tests

Adobe Lightroom Classic
Adobe Premiere Pro
Agisoft Metashape
Blackmagic RAW Speed Test
LameXP
MAGIX Vegas Pro

Go straight to test:


Rendering Tests

Autodesk Maya with Arnold Autodesk Maya with Arnold
Autodesk Maya with Arnold
Autodesk Maya with Arnold
Blender Blender
Blender
Blender
Maxon Cinebench Maxon Cinebench
Maxon Cinebench
Maxon Cinebench
Maxon Cinema 4D Maxon Cinema 4D
Maxon Cinema 4D
Maxon Cinema 4D
Autodesk 3ds Max with Corona Renderer Autodesk 3ds Max with Corona Renderer
Autodesk 3ds Max with Corona Renderer
Autodesk 3ds Max with Corona Renderer
Luxion KeyShot Luxion KeyShot
Luxion KeyShot
Luxion KeyShot
LuxMark LuxMark
LuxMark
LuxMark
POV-Ray POV-Ray
POV-Ray
POV-Ray
Autodesk 3ds Max with V-Ray Autodesk 3ds Max with V-Ray
Autodesk 3ds Max with V-Ray
Autodesk 3ds Max with V-Ray
Chaos Group V-Ray Benchmark Chaos Group V-Ray Benchmark
Chaos Group V-Ray Benchmark
Chaos Group V-Ray Benchmark

Go straight to test:

Synthetic Tests

SiSoftware Sandra SiSoftware Sandra
SiSoftware Sandra
SiSoftware Sandra

Go straight to test:


If you think there’s some information lacking on this page, or you simply want clarification on anything in particular, don’t hesitate to leave a comment.

Encoding: Premiere Pro, Vegas Pro & Agisoft Metashape

We’re going to kick off this performance look with a handful of encode tests. Encoding is one of those scenarios that can be extremely hit-or-miss when it comes to taking good advantage of big CPUs. Sometimes, applications will give the impression that they’re making proper use of the CPU, but we’ve found more than once that some applications actually just use the entire CPU very poorly.

Fortunately, the situation is getting a lot better over time. As an example, for most of its life, Adobe’s Lightroom didn’t use more than a few cores and threads. Today, the application can use most of whatever CPU you can hand it.

The performance look on this page is going to tackle Adobe’s ever-popular Premiere Pro, as well as MAGIX’s Vegas Pro. That duo takes care of video encoding for this page, while Agisoft’s Metashape will help with a photogrammetry scenario.

Adobe Premiere Pro CC – CPU

Adobe Premiere Pro 2020 - 1080p YouTube CPU Encode (AVC) Performance (February 2021)
Adobe Premiere Pro 2020 - 4K YouTube CPU Encode (AVC) Performance (February 2021)

The projects we use for testing could be a bit better developed, but as you can see, they still take good advantage of whichever CPU they’re given – even the easier 1080p test. At the top-end, the many-core chips flip-flop over each other a little bit. The 5900X 12-core seems to be one of the best bang-for-the-buck CPUs here, but you will thankfully see obvious improvement if you go even higher.

Adobe Premiere Pro CC – Codec Comparisons

Adobe Premiere Pro 2020 - 4K60 AVC to 1080p AVC Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 4K60 AVC to 1080p HEVC Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 8K24 RED to 1080p AVC Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 8K24 RED to 1080p HEVC Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 8K24 ProRes 422 to 1080p AVC Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 8K24 ProRes 422 to 1080p HEVC Encode Performance (February 2021)

While projects are varied in design and give the CPU a lot of different assets to work with, transcoding one single file from one format to another is a lot more straight-forward, and gives every CPU a better opportunity to strut their stuff.

With the project encode tests, we saw the Threadripper chips flip-flop over each other, and for the most part, we see the same with these transcode tests. It was only with the RED to AVC test that the 64-core Threadripper 3990X managed to take the top spot. In most other cases, it was actually the 32-core 3970X that took the top spot.

Ultimately, the scaling overall is quite good here. It becomes clear that you really don’t want to go too low-end on your CPU, or else you’ll be waiting a lot longer on each encode. That said, as with most rendering, CPUs tend to not be quite as fast as GPUs, but combing their efforts could lead to even better overall performance. Let’s add a GeForce RTX 3070 to the mix:

Adobe Premiere Pro CC – CPU + GPU

Adobe Premiere Pro 2020 - 1080p YouTube CPU Encode (CUDA, AVC) Performance (February 2021)
Adobe Premiere Pro 2020 - 4K YouTube CPU Encode (CUDA, AVC) Performance (February 2021)
Adobe Premiere Pro 2020 - 4K60 AVC to 1080p AVC (CUDA) Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 4K60 AVC to 1080p HEVC (CUDA) Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 8K24 RED to 1080p AVC (CUDA) Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 8K24 RED to 1080p HEVC (CUDA) Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 8K24 ProRes 422 to 1080p AVC (CUDA) Encode Performance (February 2021)
Adobe Premiere Pro 2020 - 8K24 ProRes 422 to 1080p HEVC (CUDA) Encode Performance (February 2021)

Adding a GPU to the encoding gives us some really interesting results. Somehow, the six-core Ryzen 5 5600X took the top spot of the 1080p project encode, but with this type of workload, we could see slight differences on another day (encoding is a lot more sporadic than rendering).

Ultimately, what is proven here is that a GPU is really important to video encoding, and if for some reason you still wish to stick strictly to CPU-only, you’ll want a beefy chip. Adding a GPU to the mix almost breathes new life into a CPU, allowing both to work in tandem to get the encode done much quicker. In some cases, the faster clock speeds of smaller core-count chips can help propel them high in a chart.

MAGIX Vegas

MAGIX Vegas Pro 18 - Median FX CPU Encode Performance - (February 2021)
MAGIX Vegas Pro 18 - Median FX NVENC Encode Performance - (February 2021)

With MAGIX’s Vegas Pro, we’re seeing similar scaling as we did with Premiere Pro, with the 32-core Threadripper once again utilizing its cores better than the big sibling 3990X. Adding a GPU changes the scaling entirely, although we should stress that the CPU+GPU test uses just one filter, and across the entire file. We’re currently investigating if there are better ways we can test Vegas, but in all of our testing, it’s the Median FX that has proven to be the most grueling, so it acts well as a worst-case scenario.

MAGIX Vegas Pro 18 - Colorize FX CPU Encode Performance - (February 2021)
MAGIX Vegas Pro 18 - Style Transfer Encode Performance - (February 2021)

With the release of Vegas Pro 18, MAGIX added the software’s first FX features to utilize AI. That primarily involves the Colorize and Style Transfer options, which are pretty self-explanatory, and fun to use. They also require a lot more time than most other FX types to encode; each of the AI encodes above encompass a source that was just five seconds long.

As with some of the previous results, the top-end of the chart shakes things up, and scales the models in an order we wouldn’t expect. Intel’s 18-core i9-10980XE seems particularly strong here, while the 64-core Threadripper sits behind even the 24-core. If only encoder scaling was more predictable!

Agisoft Metashape

Agisoft Metashape Photogrammetry Performance - Build Depth Maps (February 2021)
Agisoft Metashape Photogrammetry Performance - Build Dense Cloud (February 2021)
Agisoft Metashape Photogrammetry Performance - Build Mesh (February 2021)

More than most workloads, it’s difficult for there to be such thing as a “perfect” CPU for photogrammetry. From start to finish, a photogrammetry task will utilize the CPU and GPU differently throughout, so it might only be with one particular step that one CPU may excel over others.

The Depth Maps process is the first one we record, and it’s one that uses both the CPU and GPU together. The follow-up Dense Cloud and Build Mesh steps use only the CPU. As usual, the Mesh graph puts an interesting chip at the top. It’s as though the 8-core Ryzen 7 5800X has the perfect blend of cores and clocks – but only for that test.

Overall, AMD’s 12-core Ryzen 9 5900X looks to be the most effective for its given cores and clocks, but for the most part, it’s really hard to pick a single SKU as being the best here, since no one chip excels in every case. It seems a lot safer to eye a mainstream CPU over the enthusiast Core X and Threadrippers, however.

Encoding: Adobe Lightroom, BRAW Speed Test, HandBrake & LameXP

On this page, we’re going to be tackling a few additional encoding-type projects. Since the beginning of its life, we’ve benchmarked with Adobe’s Lightroom, but dropped it for about a year or two a few years back because it wouldn’t reliably scale. Over time, things changed, and now the application seems pretty efficient on multi-core CPUs.

In addition to Lightroom, we’ve also tested Blackmagic RAW Speed Test, which acts as a simple way to see how a CPU can handle playback of BRAW footage at different compression levels. Finally, we’re also testing with LameXP, an open-source music encoder that can take advantage of many-core CPUs, as well as the super-popular HandBrake transcoder.

Adobe Lightroom Classic

Adobe Lightroom Classic - RAW to JPEG Export Performance (February 2021)
Adobe Lightroom Classic - RAW to DNG Export Performance (February 2021)

As hard as it is to believe sometimes, we’ve been benchmarking Adobe’s Lightroom for nearly 14 years. For most of that time, we used the same photo set that came out of our Nikon D80. Recognizing the aging set, a friend of the website provided us with a new set of higher-resolution RAW files from a Canon DSLR. To our surprise, scaling hasn’t changed much, but the bigger files make for a more strenuous test.

To date, we’ve only tested Lightroom’s JPG export, which involves a resize and also an application of a matte finish. This time around, we added DNG export, and are glad we did, because as you can see from the results above, the scaling changes up quite a bit.

With the JPG export, the Threadripper chips take the top spots, but fail to do the same with the DNG export. With DNG, it seems as though there is such thing as a perfect blend of cores and clocks, something that helps propel the 16-core 5950X to the top spot. Amusingly, the Threadrippers that dominated the JPG export fall to the absolute bottom in the DNG export. If you need a many-core chip that performs great in Lightroom, you either want the Ryzen 9 5950X or Core i9-10980XE.

Blackmagic RAW Speed Test

Blackmagic RAW Speed Test (February 2021)

BRAW is a format that can take great advantage of CPUs and GPUs alike, something proven in the results here. Once again, the 64-core 3990X doesn’t dominate as much as it should, with the 32-core 3970X placing in front of it. Aside from those top results, the rest scales pretty much as expected all the way down. On a budget, a CPU like the 8-core 5800X or 10-core 10900K looks to offer a decent value overall, but you will definitely see improvements if you opt for a bigger model.

HandBrake

HandBrake AVC Encode Performance - (February 2021)
HandBrake HEVC Encode Performance - (February 2021)

With our HandBrake transcode tests, the 32-core 3970X yet again hits the top spot. It should be clear by now that while the 64-core 3990X is all sorts of impressive in its own way, most encoding software will scale better on the smaller models. We’re really eager to see how the next-gen Threadripper, based around the Zen 3 architecture, changes things up.

In this lineup, the 12-core Ryzen 9 5900X stands out as a good value for the money. It effectively goes toe-to-toe with Intel’s much bigger 18-core i9-10980XE.

LameXP

LameXP - FLAC to MP3 Encode Performance - (February 2021)

As someone who’s encoded tens of thousands of music tracks over the years, a test like this LameXP one hits close to home (even if I don’t encode too much anymore thanks to streaming services). LameXP won’t use every single thread the top-end Threadrippers give it, but they still manage to scale better than the mainstream counterparts.

The 5950X continues to look strong here, but everything aside from Threadripper essentially scales as we’d expect. In time, it’d be great to see how this test would scale if all cores/threads were utilized. With enough threads, a test like this could make for a good storage test, as well.

Rendering: Arnold, Blender, KeyShot, V-Ray

There are few things we find quite as satisfying as rendering: seeing a bunch of assets thrown into a viewport that turn into a beautiful scene. Rendering also happens to be one of the best possible examples of what can take advantage of as much PC hardware as you can throw at it. This is true both for CPUs and GPUs.

On this page and next, we’re tackling many different renderers, because not all renderers behave the same way. That will be proven in a few cases. If you don’t see a renderer that applies to you, it could to some degree in the future, should you decide to make a move to a different design suite or renderer. An example: V-Ray supports more than just 3ds Max; it also supports Cinema 4D, Maya, Rhino, SketchUp, and Houdini.

Autodesk Arnold

Autodesk Arnold 6 CPU Render Performance - Jaguar E-Type Scene (February 2021)
Autodesk Arnold 6 CPU Render Performance - Sophie Scene (February 2021)

We saw AMD’s top-end Ryzen Threadripper 3990X perform sporadically across much of our encode tests on the previous pages, but rendering is one scenario that a many-core chip can prove that it’s better than the rest. That 3990X effortlessly takes the top spot, with the rest of the models scaling downward as we’d largely expect. We’re not quite sure why the six-core Intel i5-10600K performed so much worse than the Ryzen 5 5600X here, but we’ll see if that continues in the next tests.

For the most part, the more you spend on your CPU, the faster your renders are going to be. Naturally, we’d be remiss to ignore the fact that we’d encourage most people to consider GPUs for rendering nowadays, especially if you’ll be able to combine both of the processors together for the same render. Arnold currently does not offer heterogeneous rendering, but we’d expect that to change in time.

Blender – CPU

Blender 2.92 - Linux and Windows Rendering Performance (Cycles CPU, BMW) (March 2021)
Blender 2.92 - Linux and Windows Rendering Performance (Cycles CPU, Classroom) (March 2021)
Blender 2.92 - Linux and Windows Rendering Performance (Cycles CPU, Controller) (March 2021)

For a full-blown look at Blender performance, we’d encourage you to look through our dedicated 2.92 performance deep-dive. That article includes not just rendering, but also viewport tests – and yes, sometimes the CPU can impact performance there (but mostly with Solid and Wireframe modes).

As part of that in-depth 2.92 testing, we also tested using a current version of Ubuntu, updated to the latest kernel. As you can see, there are definitely performance improvements to be had, and noticeable ones, at that. Whether it’s enough to make most serious Blender users leap from Windows to Linux, we’re not entirely sure, but a performance boost awaits those who do.

That said, as mentioned before, the GPU is becoming ever-more important for rendering, so if you end up opting for that focus, you can skimp on the CPU a bit, as long as it’s competent for the rest of your tasks. Speaking of GPU, let’s toss that into the mix:

Blender – CPU + GPU

Blender 2.92 - Linux and Windows Rendering Performance (Cycles CPU+GPU, BMW) (March 2021)
Blender 2.92 - Linux and Windows Rendering Performance (Cycles CPU+GPU, Classroom) (March 2021)
Blender 2.92 - Linux and Windows Rendering Performance (Cycles CPU+GPU, Controller) (March 2021)

These results help prove that the GPU is really important for fast render speeds in Blender. Take a look at the Core i5-10600K six-core, for example. In the BMW test, the CPU alone rendered the scene in 209 seconds (in Linux), a value dropped to 27 seconds with the RTX 3070 added in. If you pair a many-core CPU with your GPU, you will see even larger gains – the 64-core dropped from 34 to 17 seconds in the same BMW test.

Note that it’s only the Cycles render engine in Blender that can take advantage of the CPU. If you’re using Eevee, you’ll be strictly looking at GPUs, as the CPU is used for things other than rendering (but you still don’t want to go too low-end; higher clocks do help.)

Luxion KeyShot

Luxion KeyShot 10 - Character Render Performance (February 2021)
Luxion KeyShot 10 - Room Render Performance (February 2021)

We believe that this is the first time we’ve taken a look at CPU rendering with KeyShot since version 10 came out, and as always, it scales really well on big CPUs. Like Arnold, there is no heterogeneous option in KeyShot to take advantage of both the CPU and GPU at the same time for rendering, so you can’t expect speed-ups there. However, the software does have rich multi-client support, so you could have two instances open, using the CPU for one, and GPU for the other. You can even use two instances of KeyShot for the same project at the same time, using a dedicated processor for each.

Chaos Group V-Ray – CPU

Chaos Group V-Ray - Flowers CPU Render Performance (February 2021)

Every single renderer tested in this article is either CPU-specific, or was CPU-specific at one time. As soon as GPUs became better at accelerated ray tracing, they quickly became adopted for the purpose. V-Ray was CPU-only for most of its life, but CUDA support was added in a few years ago, followed by OptiX (required for RT core use) more recently.

Interestingly, the scaling at the top-end here isn’t quite as pronounced as it is in some of the other renderers, but nothing changes about the fact that the bigger the CPU, the faster your renders will be. AMD’s 16-core 5950X even managed to step ahead of the 18-core 10980XE in this test.

As before, let’s add our RTX 3070 to the rendering process and see what changes:

Chaos Group V-Ray – CPU + GPU

Chaos Group V-Ray - Flowers CPU+GPU Render Performance (February 2021)

There are some results here we didn’t quite expect. The 64-core 3990X didn’t exactly blow the 32-core 3970X out of the water in the previous CPU-only test, yet it pulls far ahead in the heterogeneous test. This becomes the best example of CPU+GPU for AMD’s biggest offering.

Fortunately for everyone else not looking for a $3,990 CPU, the GPU can add so much performance to the rendering process, that the importance of the CPU is reduced a bit. As always, you still don’t want to go too low-end, because the faster your clocks, the better the responsiveness. But, if you have needs for a lot of memory, and perhaps a quad-channel memory controller, you’ll pretty much be tied to either the AMD Ryzen Threadripper or Intel Core X platforms.

Chaos Group V-Ray Benchmark

Chaos Group V-Ray 5 Benchmark - CPU Render Score (February 2021)
Chaos Group V-Ray 5 Benchmark - CPU+GPU Render Score (February 2021)

We’re sometimes asked why we like to test with real renderers inside of real design suites, and the above charts can help explain it. In our Flowers test, using only the CPU didn’t really help the 3990X strut its stuff too well, but in this synthetic test, it pulls ahead quite a bit more. Likewise, it didn’t gain quite as much of an advantage in the CPU + GPU test, but it’s still clearly dominant.

For a high-end best bang-for-the-buck, the Ryzen 9 5950X and Core i9-10980XE deliver both great clocks and core counts for the purpose, but it’s not the end of the world if you need to adopt a more modest CPU. With a GPU added, every CPU suddenly seems so much better.

Rendering: Cinebench, Cinema 4D, Corona, LuxMark, POV-Ray

We covered a handful of major renderers on the previous page, but we’re not done yet. On this page, we’re going to take a look at a few more, including some industry mainstays and newbies. That includes Corona Renderer, which we recently upgraded to version 6.

To give you an opportunity to test your own hardware against ours, we’re also including the popular synthetic tests POV-Ray and LuxMark, as well as Cinebench, which represents current R23 performance. For good measure, we’ll also be testing the real Cinema 4D R23 with multiple scenes.

Cinebench R23

Maxon Cinebench R23 - Multi-threaded Score (February 2021)
Maxon Cinebench R23 - Single-threaded Score (February 2021)

Cinebench is one of the best standalone benchmarks that anyone could test with, as it directly represents real Cinema 4D workloads. While many of the renderers in our lineup for this article can use either the CPU or GPU, C4D’s default render engine still relies entirely on the CPU. If you want GPU compatibility, you will need to opt for another renderer, such as Arnold, Redshift, or V-Ray.

As you’d expect, the more cores you feed C4D, the quicker your render will complete. Let’s see how these results stack up with our real-world tests:

Cinema 4D R23

Maxon Cinema 4D R23 - Staircase Render Performance (February 2021)
Maxon Cinema 4D R23 - Energy Drink Render Performance (February 2021)
Maxon Cinema 4D R23 - Abstract Loop Animation Render Performance (February 2021)

C4D’s render engine is one of the most consistent we’ve ever seen. All three of our real-world projects scale almost the same, with the top 64-core chip not exactly pulling as far ahead as we’d like to see. Still, the bigger your CPU, the less time you’ll have to spend waiting on your renders.

Corona Renderer

Chaos Czech Corona Renderer 5 Performance - Livingroom Scene (February 2021)
Chaos Czech Corona Renderer 5 Performance - Sales Gallery Scene (February 2021)

Corona is another renderer that focuses squarely on the CPU, and over the past few years, the developers seem to have ramped up their efforts to pile on more useful features. As with C4D above, the bigger your CPU, the faster Corona is going to render a scene.

It’s interesting to see, however, that AMD appears to be inherently faster in a number of these renderers, including Corona and Cinema 4D. AMD’s six-core beats out Intel’s 10th-gen six-core by a fairly large margin – one we really wouldn’t expect, given the higher peak clock speeds of the Intel chip.

LuxMark

LuxMark Food (C++) Render Performance (February 2021)
LuxMark Hall Bench (C++) Render Performance (February 2021)

As Maxon’s Cinebench benchmark reflects performance of its Cinema 4D application, LuxMark represents performance seen in LuxCoreRender. As this renderer is built around Intel Embree, we expected Intel’s chips to perform stronger than what we’re seeing here. The six-core AMD still manages to beat out Intel’s own six-core, despite Intel’s clock advantage.

Speaking of clocks, the various configurations of these CPUs has led to interesting scaling here. Overall, the Ryzen Threadripper chips don’t fare well at all, falling behind more modest Zen parts most often. Meanwhile, AMD’s 16-core 5950X, which has much faster clocks than the Threadripper chips, soars to the top of the chart.

In yet another example of how different scenes can scale differently, the Hall Bench scene proved to be much more favorable to Intel than the Food one. Overall, though, AMD carves itself a commanding lead.

POV-Ray

POV-Ray 3.8 Multi-threaded Score (February 2021)
POV-Ray 3.8 Single-threaded Score (February 2021)

To wrap things up, we’re using a crowd favorite: POV-Ray. There are few renderers we’ve tested quite as long as we have POV-Ray, so we have a bit of a soft spot for it. As with LuxMark, the scaling is a bit interesting here, with the 64-core Threadripper chip once again falling behind the 32-core – a bit unusual for a rendering test. The single-thread test can really highlight the differences between the many-core modest-clocked and the lower-core high-clocked chips.

System: SiSoftware Sandra

While this article has no lack of synthetic benchmarks, SiSoftware’s Sandra makes it very easy to get reliable performance information on key metrics, such as arithmetic, multimedia, cryptography, and memory. Sandra is designed in such a way that it takes the best advantage of any architecture it’s given, so each CPU always has its best chance to shine.

That means a couple of things. This is definitely the “best” possible performance outlook for any chip, and doesn’t necessary correlate with real-world performance in other tests. It’s best used as a gauge of what’s possible, and to see where one architecture obviously differs from another.

Multimedia

SiSoftware Sandra 2020 - Multi-media Performance (February 2021)

As mentioned above, Sandra is really good at giving every CPU the fairest shake, and the result is the type of scaling we’d expect to see if we were to guess. The exceptions would be the Intel fights with the 6- and 10-core chips, which both fail to match the nearest AMD competition. The 10-core Intel actually places behind the 8-core AMD, which is the kind of thing we’re not used to seeing, since Intel has historically been the stronger vendor in this test.

At the top-end, it’s nice to see Threadripper scaling so well, as it’s been a bit spotty throughout the rest of the article (namely with video encoding).

Arithmetic

SiSoftware Sandra 2020 - Arithmetic Performance (February 2021)

It turns out that it’s not just with the Multi-Media test that Intel struggles to catch AMD in; it’s also the case with this Arithmetic test. Since this test can take advantage of AVX-512, we expected the i9-10980XE to place even higher than it does. The fact that AVX-512 couldn’t catch the 16-core Ryzen is surprising.

Cryptography

SiSoftware Sandra 2020 - Cryptography (High) Performance (February 2021)
SiSoftware Sandra 2020 - Cryptography (Higher) Performance (February 2021)

We mentioned that AVX-512 didn’t help the i9-10980XE out too much in the Arithmetic test, but it makes up for that in this cryptography one. With the most complex cryptography load, the 10980XE managed to slot in ahead of AMD’s 32-core Threadripper. Oddly, the 6- and 10-core Intel chips fall to dead last in the AES-256 + SHA-256 test, but the scaling becomes much more expected in the AES-256 + SHA-512 test.

Memory Bandwidth

SiSoftware Sandra 2020 - Memory Bandwidth (February 2021)

This last test is one of the simplest of the entire bunch. This graph paints an easy to understand picture that quad-channel enthusiast platforms are able to deliver far more memory bandwidth than mainstream dual-channel ones. Fortunately, whether you’re using AMD or Intel, you’re going to get similar memory bandwidth.

Final Thoughts

As always, which CPU will suit your needs the best will really depend on your workflow. More than ever, many are moving their rendering tasks to the GPU, which makes us want to suggest getting an 8-12 core chip that offers great clock speeds. Faster clocks means faster responses to interactions in software. You can see from our recent Blender 2.92 article that your CPU choice can even impact viewport performance.

For CPU-dedicated rendering, the huge collection of tests in this article have proven that more cores = quicker renders. Unfortunately for Intel, AMD’s latest generation of Ryzen chips are seriously strong in many regards, but especially with rendering. We saw multiple instances where the new AMD six-core 5600X beat out the Intel six-core Core i5-10600K. In a couple of other instances, the eight-core 5800X even managed to edge out the ten-core i9-10900K.

If you’ll primarily be doing your rendering (or encoding) tasks with the GPU’s help, then you’re not as likely to need a many-core CPU. After looking over all of the performance in this article, we feel as though a chip like the $449 Ryzen 7 5800X offers great performance for its price. For many, eight cores should be sufficient, but it’s important to understand how a larger CPU would impact other aspects of your workflow, as well (such as physics, baking, etc.)

AMD and Intel Processor Boxes

For heavier users, a chip like the $549 Ryzen 9 5900X also feels like a good value, given it has 12 cores under its hood, and $100 premium over the eight-core. A chip like that would offer a lot more breathing room for longer, but again, you need to make sure none of your current bottlenecks would remain bottlenecks on an eight-core (or smaller) chip.

For rendering, the CPUs with the most cores win (ignoring anomalies), but in encoding, performance of one chip vs. another is even harder to sum-up. If you looked through our Premiere Pro data, you would have noticed that not only do full projects behave differently from one CPU to another, but individual codecs do, as well. If you work primarily with one codec more than others, you will want to make sure your chosen CPU is actually better than the competition.

As we like to say, not all benchmarks are built alike, and the myriad results in this article highlight that very well. As mentioned before, if you’re interested in any gaming performance, you can look at our most recent look. If we don’t tackle performance that you’d like us to, please leave a comment, and we’ll explore it.

Copyright © 2005-2021 Techgage Networks Inc. - All Rights Reserved.