As creatives in the VFX industry, we know how long render times can be, and with our commitment to providing you with high-performance workstations in mind, we decided to put our latest additions to the test.
These rendering tests were designed to test different aspects of rendering performance and ultimately showcase how the CPU chosen for GPU rendering engines, like Redshift, can profoundly affect render times.
The three different setups tested were as follows:
HP z840 Dual Xeon (20 cores total @ 3.1GHz) 64GB of RAM
VFX i9 Workstation 7980XE (18 Cores total @ upto 4.2GHz) 64GB of RAM
VFX i7 Workstation (6 Cores total @ GHz 4.8GHz) 64GB of RAM
TEST 1
The first rendering test consisted of a scene which rendered almost instantly in the viewport, but it was tasked with writing a very heavy multichannel EXR file containing cryptomatte among other render AOV’s.
The i7 wins by about 10 seconds per frame over the i9, which demonstrates how fast the machines can take the rendered data from buffer then write it out to a large multilayer EXR files.
This process is about 98% CPU as the render itself would have been about 2 seconds per frame to compute. Although the i7 might have fewer cores and potentially lower performance cache, its raw clock speed wins. Running at 3.7GHz and over 4GHz with turbo-boost versus the i9's 2.1GHz.
Both i-series processors beat out the z840 by a large margin.
TEST 2
The second rendering test we ran tested how effectively the GPU communicates with both the CPU and RAM.
In Redshift, the CPU is constantly working to shuttle the data back and forth, which is a multi-threaded process. The scene in this test is a high compute scene set to very low samples, so we can filter out raw times shown to CPU centric times.
Again, raw clock speed wins with fast renders with the i7 showing a 7 second advantage over the i9 on an 18 second task. The i9 is not far behind, but it does lose due to raw clock speed. The z840 did much worse in this test. Showing a major bottleneck in the dual CPU's communicating with the GPU's
TEST 3
The next rendering test is heavy on the GPU. It includes a scene running all kinds of heavy elements, like rendering in 4K with bloated AOV's, DOF, glossy refractions, SSS through refractions.
It's built to test how different CPU's affect GPU performance over a much longer period of time, which shows us the kind of percentage impact it will have.
This time the i9 won with more cores to do more shuttling of data when compared to the GPU's along with a higher performance processor all around. Those frames completed in 15m.
The i7 fell behind by about 2 minutes per frame although bolstered by its higher clock speed with each core operating at almost double the i9 clock speed.
The z840 lagged behind the i9 about 6 minutes.
TEST 4
The last rendering test we ran included a Maya scene, which was relatively hard to open due to it containing thousands of objects. The render time on this was moderate as we focused on the GPU translation time rather than rendering.
The i7 came out on top again due to its higher clock speed. Then came the i9 with the z840 lagging behind the most out.
Opening heavy scenes is a single threaded process, which is why the i7 majorly outperformed here. The data translation to the GPU is multi-threaded, so the i9 was able to catch up a bit.
The z840 boards with the dual CPU just aren’t really made for efficient communication with the GPU although they might be better for running more hyper-threaded CPU computations.
Real-Time Raytracing
While deferred rendering is still industry standard, there is a new method of rendering that is making its mark on the industry. Even with the new line of GPU’s making a big leap in performance from the previous release, it’s still extremely time consuming. Renders could take hours, days, even weeks to complete for a single scene. But with real-time raytracing, render times have significantly decreased while maintaining very similar quality. Along with decreasing render times, real-time raytracing enhances the user experience allowing for an improved workflow which results in increased productivity. Although real-time raytracing is still in its early stages, the advantage is clear and many studios are beginning to make the leap and transition over into Unreal Engine. For now, there is still a place for standard deferred rendering. But as the software continues to develop, real-time raytracing could replace many of the standard methods of rendering we have today.
FINAL THOUGHTS
The i7 and i9 are the clear winners here, so spending more money doesn’t necessarily equate to faster render times.
The Dual xenon z840 was vastly out performed by these much cheaper CPUs when the GPUs are equal. Throughout our rendering tests, the i9 does have a slight advantage over the i7 when doing production scale renders, where data throughput to and from the GPU is very important.
However, we recommend the i9 9900K, which is only a fraction of the cost of the more expensive 7880XE and still carries a higher clock-speed. This setup would have the advantage of the i9 architecture while carrying the higher clock-speeds, which, as shown by the tests, are a very important component even in GPU rendering.
RTX 3080
The next generation RTX cards provide the best performance per dollar, and even when money between the last generation and next generation cards isn’t considered, the performance gains are simply staggering. Here's the Octane Bench Test for easy comparison.
When rendering heavy scenes that include CAD data, motion blur, depth of field, and refractions, in Redshift all at once, then the 3080 consistently cut render times in about half without a major change in software ( note: Redshift requires a new point release for compatibility with the Ampere architecture). The new OptiX enabled feature in Redshift allows the card to use the RTX cores for calculation which increases the speed significantly.
The 2.5 slot form factor does allow for a dual card setup, but I was recording much higher temperatures in the card which had its fans just millimeters from the other card. This resulted in some thermal throttling of the upper primary PCI slot. When rendering for extended periods of time or running Unreal engine in full RTX mode, we didn't push the cards too hard while compromising the thermal solution.
Our solution was to put either the 2080TI or 2070Super in the upper slot and the 3080 in the lower slot. This ensured plenty of airflow to both cards. The render times were even faster challenging production renders to the realm of 1 - 2 minutes a frame. When enabling AOV’s and cryptomatte, this went up to about 2.5 - 3 minutes a frame. Still blazing performance!
While not avid gamers, we test and research games to gawk at their amazing graphics, but we rarely set all the sliders and settings to their maximum because there's always a drop in frame rate or some performance hit. However, the 3080 allowed us to set every setting to maximum and set supersampling to the biggest it would go; it seemed as though the machine wasn’t really trying very hard to keep up.
For VR, the card allows for buttery smooth frame-rates, and we never encountered a hitch, which is something I would encounter often on the 2070Super.
When coupled with the right CPU’s, software settings, and render optimizations, this card and the new line of RTX cards take the anxiety of rendering away from freelancers and independent contractors. Now we can focus more and more on the look and push the settings higher from an earlier point of the production process; clients and creatives alike no longer have to see past grainy preview renders or playblasts.
The next step will be to incorporate these cards and setups into VFX technologies’ new line of modular and highly rugged machines. This will allow for operators to work on the field, on set, in the elements, and bring high performance wherever they go.