Increasingly the demands of virtual production and high-end visual effects demand powerful GPUs, both for the graphics capabilities and the enormous requirements of Machine Learning.  There is somewhat of a worldwide shortage of graphics cards, and yet growing demand. Into that world, NVIDIA now offers the A5000 card. This is effectively the enterprise version of the RTX 3090. We have been test-driving the A5000 alongside our other RTX 3080 and other hardcore graphics cards such as the RTX 8000 cards. This has not been a benchmark test so much as a practical exploration of pipelines for both virtual production and digital humans.

In terms of fxguide testing our pipeline, Epic Games presented a perfect vehicle with the new UE5 early access last week. While UE5 runs on a range of hardware including the AMD RX Vega 64 or higher, like many others, we have been focused on the NVIDIA hardware. We use the Radeon Pro Vega cards in our Mac machines, but for our high-end PC projects, we use NVIDIA.

Epic recommends the minimum requirement for general UE5 use is an NVIDIA 2080 card, but most of the demos at the launch of UE5 were performed on an RTX 3080 level card. The UE5 install states that the NVIDIA GTX 1080 is the minimum spec-ed graphics card but speaking personally, we would not use anything below an NVIDIA RTX 2080. The new sample project that was released with UE5, Valley of the Ancient, is a 100Gig download and it really shows the power of the NVIDIA cards to full advantage. Our test machine had both an RTX3080 and a newA5000 installed, and the results were impressive.

Test driving UE5

The Valley of the Ancient, and UE5 were designed to run at 30fps on the PS5 and, naturally, on the newer graphics cards such as the RTX 3090 and A5000. These powerful cards really allow UE5 to shine. At 4K resolution with the large Epic demo scene, UE5 runs at close to 30fps full screen, on our system, which is a Gigabyte Z590 Aorus Master (basically a gaming motherboard) with Intel i9 CPU, with SSD fast drives ([email protected]/sec) and 64Gig of CPU RAM. On our Asus XG27UQ 27″ UHD 4K monitor, Valley of the Ancient is glorious – no really – just wonderful. A key point to note is with two cards in our machine, each only gets half the PCIE bandwidth (so a 16x slot becomes effectively two 8x slots).

The gold master full version of UE5 is expected to be released early next year and there are no current plans for any major additional releases between this early preview and the shipping full version, although there may be some bug fix releases if issues come up, as they sometimes can. From our testing, these NVIDIA cards have enough capacity to allow you to develop and push UE5 for quite some time.


The NVIDIA RTX A5000 delivers the power, performance, and capabilities similar to the RTX3090 which is also based on the latest NVIDIA Ampere 7th generation GA104 architecture. The RTX 3090 was released in September last year, while the A5000 has only just started shipping. The RTX A5000 provides 64 second-generation RT Cores, 256 third-generation Tensor Cores, and 8,192 CUDA cores with 24 GB of graphics memory. The A5000 provides AI, graphics, and advanced computational power, all in a slim and compact form factor that only requires a single 8-pin power connector, drawing a maximum of 230W power, much less than the 320W power draw of an RTX 3080. There are differences between the cards, but mainly the A5000 offers a more robust, reliable workhorse version of the RTX 3090, not a vastly different graphics performance difference.

Post-render vector motion blur artifacts including banding.
Full Render Motion Blur


The A5000 also offers 50% more memory than the previous generation RTX 5000. The new card’s 24GB of GDDR6 allows the A5000 to work with even larger 3D datasets. It also provides valuable rendering benefits. Second generation RTX GPUs like the RTX A5000 support hardware-accelerated motion blur, allowing these GPUs to significantly speed up motion blur during the rendering process, and avoid artifacts that can be produced with post-processed ‘vector-motion blur’.

When rendering 3D, motion blur can either be computed by the render engine during the rendering process, or it can be added as a post-processing effect on a non-motion blurred series of frames. Computing motion blur during the render process produces exceptionally high-quality results. However, it has traditionally significantly increased render times due to the additional computation overhead associated with computing the motion blur. With the A5000 artists can take advantage of the higher quality full rendered motion blur, with less impact to render times. In Blender, adding full motion blur could double the render time compared to the previous RTX 5000 card. With the new A5000, full motion blur may only add 20-25% to the render time, but that render will already be almost twice as fast as it was on the old 5000 card. Part of this is due to the 50% increase in graphics memory on the new A5000.

Redshift runs extremely quickly on the A5000

Naturally, these cards are exceptionally well balanced with NVIDIAs increasingly powerful Omniverse toolset. Omniverse is both enabling new applications via SDK tools such as Audio2Face but also encouraging professional collaboration. Omniverse serves a key role in extending the workflow from a powerful user experience into a powerful collaborative experience, which is key for virtual production and advanced M&E pipelines. All of the major renderers work extremely well, especially GPU renderers such as Octane and Redshift.


Why A5000?

The key question is why look at the A5000 compared to the RTX3090? The primary reason is reliability and faultless continuous performance. The 3090 is officially a GeForce RTX3090, and it is a great gaming card. Both it and the 3080 are designed for desktop use, rather than workstation use. The A5000 is a Professional (Quadro) RTX A5000. The difference may seem arbitrary, but in the world of high-performance projects, it is not negligible. The A5000 is designed to run constantly and everything from the airflow over the card to the quality of the memory components reflects that. If you are a gamer interested in running the system hard for only a few hours at a time, the 3080 or 3090 are great. However, in production with long periods of rendering, or (as we and many others now use their GPU systems), for long periods of Machine Learning training, the reliability and robust performance of the A5000 may be a better solution.


For example, the new A5000 board features a custom-built fan design which is specifically designed to allow for multiple RTX A5000 boards to be utilized within a single system. The fan design moves airflow past the GPU and exhausts it out the back of the A5000 board, moving the hot air outside the system. This prevents the A5000 board from exhausting hot air into the case, which allows A5000 boards to be placed directly next to each other without requiring any additional spacing between boards.

If you are on a budget, or only wanting to test small projects, then the new RTX 3080Ti is also a  good option with a cheaper entry price and a good replacement for 2080Ti cards. The 3080Ti is the newest NVIDIA card, but unlike the bigger enterprise-level cards, it still only has 12GB of memory and will no doubt also be in short supply as everyday gamers seek out its strong graphics real-time ray tracing performance enhanced by NVIDIAs DLSS (Deep learning super sampling) ‘turbocharging’. For M&E professionals this might be a good personal option, but it lacks many of the professional high-end NVIDIA key features.

Dual-linked cards.

It is possible to connect two RTX A5000s with NVIDIA NVLink to scale memory and performance with multi-GPU configurations. Two A5000 graphics boards could be connected using NVIDIA NVLink allowing 48GB of combined GPU memory. This allows artists to work with memory-intensive tasks such as large models, ultra-high-resolution rendering, and complex work. In our case, we are running two boards in our system but not linked. One is providing real-time body motion capture, hand capture and face capture, while the other card is doing Machine Learning inferencing and neural rendering.

Virtual Production

Gaming is an enormous industry worth $180 billion annually, but we are focused on the use of these new NVIDIA cards in M&E, and in particular in virtual production (VP). The demands of real-time are incredible in VP. Not only is the A5000 well set up for continuous use but it is also a perfect card for professional VP LED wall volumes. The RTX A5000 can utilize the additional Quadro Sync II card to synchronize the display and image output from multiple compatible cards within a single system, or across a cluster of computers/servers. Each Quadro Sync II card is capable of synchronizing the output of up to four RTX A5000 boards. Two Quadro Sync II boards can be installed in a single system, allowing a workstation or server with 8 GPUs (each with four active independent display outputs) to support up to 32 synchronized video displays. The Quadro Sync II card also offers the ability to generate an external house sync signal output via the BNC connector, eliminating the need for external sync generation hardware.

NVIDIA Mosaic is the technology that provides advanced features required in a professional multi-display solution. Mosaic technology creates a single, unified desktop environment allowing any application to be run natively without having to be multi-display aware. At the most basic level, NVIDIA Mosaic enables applications to “just work”, simplifying the task of configuring and deploying a multi-screen solution. Mosaic handles the individual control of the independent displays and presents the OS with a single “virtual” display that matches the resolution of all the combined displays added together. In addition to managing the virtual combined desktop, Mosaic offers bezel correction (and/or projector overlap) to provide an aligned LED display. Used in conjunction with the Quadro Sync II card, Mosaic properly synchronizes the display across all the LED monitors, offering a smooth, aligned visual solution.

In short, the top-end GPUs, such as the A5000 offer an important trifecta of graphics, machine learning and collaboration, including multi-card/multi-screen use, for high-end virtual production. Are they enough? Heck no!  NVIDIA is providing great power, but virtual production is not a solved problem and no one in VP is suggesting next year they will need less power, or being doing less ML or require less power from their GPUs, – but for now these top-end cards will prove invaluable.