How to Configure Multi-GPU Systems for Resolve and After Effects

If you run Resolve all day, build around one very strong GPU first, then add a second only when your timelines are actually GPU-bound. If you run After Effects all day, spend the second-GPU money on RAM, CPU, fast cache storage, or a better primary GPU before you expect a meaningful multi-GPU win. That's the shortest useful rule. Resolve’s image pipeline is heavily GPU-based, especially for grading, noise reduction, temporal effects, debayering, scaling, and GPU-aware OFX. After Effects uses the GPU for specific effects, viewers, color management, and some rendering paths, but the app is usually gated by CPU, RAM, cache, project structure, plugins, and Multi-Frame Rendering. A dual-GPU workstation can be great for a Resolve finishing room and underwhelming for an AE motion graphics seat.

Resolve finishing can use two GPUs well, while many After Effects seats benefit more from memory, CPU, and cache speed.

You need the right app, motherboard lanes, BIOS behavior, driver stack, thermal layout, and expectations about VRAM. Two GPUs don't magically become one giant GPU. In most post workflows, they're two workers sharing frame processing, not one combined memory pool.

Multiple GPUs share work, but their VRAM stays separate rather than merging into one larger pool.

Match the hardware to the workload first

Multi-GPU makes the most sense when the workstation spends a lot of time doing operations that can be split across GPUs. In Resolve, that often means high-resolution finishing and grading where the GPU is doing real image processing for every frame. Good candidates include:

4K, 6K, and 8K timelines with heavy color nodes
Temporal noise reduction and motion effects
GPU-accelerated OFX chains
RAW debayering at high quality
Deliverables where image processing dominates render time, not codec export

Those workloads are where a second GPU is most likely to have something useful to do. Poor candidates include:

Simple edit timelines with light correction
Long-GOP camera formats where decode is the bottleneck
Exports limited by a single hardware encoder
Network storage that can't feed frames fast enough
After Effects comps dominated by expressions, shape layers, CPU-only effects, or single-threaded plugins

In those situations, another GPU is unlikely to fix the active bottleneck. Blackmagic’s own configuration guidance has long described Resolve as GPU-based image processing software, not a traditional editing app that only uses the GPU occasionally. That's why Blackmagic has historically recommended high-performance GPUs, and in larger systems multiple GPUs, for serious grading and finishing. But the same guidance also ties performance to source format, codec, timeline resolution, and render target. The GPU is central, but it isn't the only bottleneck. After Effects is different. Adobe documents GPU requirements and GPU acceleration, and specific GPU-accelerated effects can be much faster than CPU-only rendering. Adobe has described some GPU-accelerated effects as rendering several times faster than CPU-only paths. But that doesn't mean two GPUs double After Effects. Most AE artists get more predictable gains from enough RAM for Multi-Frame Rendering, a fast NVMe cache, current GPU drivers, and reducing CPU-only bottlenecks inside the comp.

What “multi-GPU” actually means in post apps

There are two ideas people often mix up: load balancing and memory pooling.

Concept	What it means	Resolve impact	After Effects impact	Buying implication
Load balancing	The application splits work across more than one GPU.	Can improve heavy grading, temporal noise reduction, scaling, and GPU-aware effects when the GPU is the bottleneck.	Usually limited. AE may use GPU acceleration for specific features, but a second GPU rarely changes overall comp performance much.	Useful when one GPU is already saturated and VRAM use is still comfortable.
VRAM pooling	Multiple GPUs act like one larger shared memory space.	Usually not available in the way editors expect. Each GPU still needs enough usable VRAM for the workload.	Not a practical planning feature for typical AE workstations.	Do not buy two smaller cards to solve a VRAM shortage. Buy one card with more VRAM instead.
Mixed-card scaling	Different GPUs work together in the same system.	Can work, but the slower or smaller-VRAM card may reduce the practical benefit and complicate support.	Usually not worth planning around unless another application needs it.	Prefer matched GPUs for Resolve. Prefer one stronger GPU for AE-first systems.

Load balancing means the application can divide work across multiple GPUs. In Resolve, this can help when image processing is the bottleneck. One GPU may process one frame while another processes another, or the app may divide certain operations depending on the processing path. The result can be faster playback for heavy grades or shorter renders, assuming the rest of the system keeps up. Memory pooling means multiple cards combine VRAM into a single larger memory space. That isn't what most Resolve and After Effects workstations get. For most dual-GPU post systems:

Two 24 GB GPUs don't behave like one 48 GB GPU.
Mixed-VRAM cards usually behave like the smaller card is the real limit for shared workloads.
NVLink, where available, isn't a universal “combine VRAM” switch.
SLI isn't required for Resolve and is generally not relevant to modern DCC performance.
Don't plan After Effects around VRAM pooling.

This matters when deciding between one large GPU and two smaller ones. If your Resolve timeline needs more than 12 GB of VRAM because of 8K media, high-resolution stills, temporal processing, or large Fusion comps, two 12 GB cards are usually the wrong answer. A single 24 GB or 48 GB GPU is safer. If your timeline fits comfortably in VRAM but needs more compute, dual GPUs can make sense. For After Effects, choose the larger, stronger single GPU almost every time unless you also use a GPU renderer, 3D application, Resolve, or another tool that explicitly scales across multiple devices.

Choose GPUs that behave well together

The cleanest multi-GPU Resolve system uses identical GPUs. Same model, same VRAM, same driver family, same cooling behavior. Mixed cards can work, but they make troubleshooting harder and can reduce the benefit. Your team will usually get a more stable post workstation by choosing matched professional or high-end consumer GPUs rather than a grab bag of available cards. The main variables are VRAM, compute performance, cooling, slot width, power draw, and driver support. Use these guidelines when choosing cards:

Prefer matching GPUs for Resolve multi-GPU systems.
Prioritize VRAM per card for high-resolution finishing.
Avoid blower/open-air combinations that fight each other thermally.
Leave physical space between cards when possible.
Confirm the power supply can handle sustained render load, not just peak desktop use.
Use current Studio, Pro, or enterprise drivers rather than gaming-first drivers on production systems.
Don't mix NVIDIA and AMD in the same Resolve workstation unless you have a very specific reason and have tested it.

For Resolve on Windows and Linux, NVIDIA CUDA configurations are common because of driver maturity and broad plugin support. AMD can be viable, especially on macOS or specific workstation builds, but the important part is consistency. On macOS, Apple Silicon changes the conversation because the GPU and unified memory are part of the SoC, not a traditional PCIe multi-GPU workstation.

Motherboard and BIOS settings that matter

A lot of multi-GPU problems start before the operating system loads. The motherboard must expose both GPUs with enough PCIe lanes, enough address space, and a sane slot configuration.

Before software settings matter, the platform needs the right slot layout, PCIe lanes, and physical spacing.

On workstation and high-end desktop platforms, two GPUs often run at x16/x16 or x8/x8. PCIe x8 is usually fine for many GPU compute workloads, especially on newer PCIe generations, but you don't want the second card hanging off a chipset slot with limited bandwidth if you expect serious performance. Look for these BIOS and platform settings:

Set PCIe slot configuration to x8/x8 or x16/x16 where applicable
Configure PCIe bifurcation correctly if the board requires manual setup
Enable Above 4G Decoding for multiple large PCIe devices
Enable Resizable BAR only if it's stable with your GPU and driver stack
Set Primary display to the GPU that drives the UI monitor
Disable integrated graphics unless you intentionally use it for display or Quick Sync workflows
Disable CSM on modern UEFI Windows systems
Set PCIe generation to Auto, or manually lower it only to troubleshoot instability
Set adequate fan curves for sustained GPU load

After changing BIOS settings, confirm both GPUs appear correctly in the OS before opening Resolve or AE. In Windows, Device Manager should show both cards without warning icons. Task Manager can show individual GPU load if you switch the graph to CUDA, Compute, 3D, Copy, Encode, or Decode depending on what you're testing. NVIDIA users can also use command-line tools to see utilization, memory use, power, and thermals during playback or render. The takeaway is simple: if the OS doesn't see two healthy GPUs, Resolve can't fix that for you.

Driver setup for Resolve

For Resolve, install a current production driver and keep it consistent across the room if multiple systems need to behave the same way. Avoid updating drivers in the middle of a show just because a new version exists. Update when you need a fix, a new GPU, a new Resolve version, or a security requirement, then test with real projects. Inside Resolve, the relevant settings live in Preferences, under the system memory and GPU area. The exact labels change by version and platform, but the idea is consistent: Resolve can choose automatically, or you can manually select the GPU processing mode and which GPUs Resolve uses. Typical settings to review include:

GPU processing mode: Auto, CUDA, OpenCL, or Metal depending on platform
GPU selection: Auto or manual
Which GPUs Resolve uses for compute
Whether Resolve also uses the display GPU for image processing
Memory allocation and cache behavior
Video I/O settings if the machine also has Blackmagic playback or capture hardware

On a dedicated grading workstation with two matched GPUs, Auto often works. For troubleshooting, manual selection is useful because it lets you test each card alone, then both together. If one GPU is driving several GUI displays and the other is free, you may get better responsiveness by managing which card handles compute, depending on the system and Resolve version. Don't enable SLI expecting Resolve to become faster. Resolve’s multi-GPU support is application-level, not a gaming SLI profile. In many workstation setups, SLI is irrelevant or actively undesirable. Use a known project that represents the work you actually deliver: same camera formats, same timeline resolution, same node tree style, same noise reduction, same OFX, same render preset. Test one GPU, then both. Watch GPU load, VRAM use, CPU load, disk activity, and render FPS. If GPU load drops while CPU, decode, encode, or disk is pinned, a second GPU won't solve the active bottleneck.

Driver setup for After Effects

After Effects has fewer multi-GPU decisions because there's less useful multi-GPU behavior to configure. The setting most artists actually need is the project renderer under Video Rendering and Effects. Depending on OS and GPU, that might be Mercury GPU Acceleration using CUDA, Metal, or OpenCL. If you set it to Mercury Software Only, GPU-accelerated effects fall back to CPU rendering. Check these AE areas when configuring a workstation:

Project Settings, Video Rendering and Effects
GPU Information in Preferences
Media and Disk Cache location and size
Memory and Performance allocation
Multi-Frame Rendering settings
Current GPU drivers and plugin versions

Adobe’s own performance guidance emphasizes current drivers and plugins, sufficient RAM, and memory allocation. That matches real AE behavior. If you enable Multi-Frame Rendering but the machine doesn't have enough RAM per core, more CPU cores can sit idle or previews can become inconsistent. Adobe has historically recommended enough installed RAM per processor core, and in modern production work you should be generous. Heavy AE seats are rarely upset about having 128 GB or 256 GB of RAM. They're often upset about having two GPUs and not enough cache or memory. If AE doesn't show the expected GPU, fix that at the OS and driver level first. On laptops and eGPU systems, OS-level “prefer external GPU” or high-performance graphics settings may affect which device AE sees, but eGPU behavior is less predictable than an internal workstation card. For production desktops, keep it boring: one primary display GPU, clean driver install, stable OS, fast local cache.

Where the gains show up

In Resolve, a second GPU can produce very real gains, but the curve depends on the task. Heavy temporal noise reduction, high-quality scaling, blur-heavy looks, and GPU-accelerated OFX are the classic places where multiple GPUs help. Render exports may speed up dramatically if the grade is the bottleneck. Where the gains shrink:

CPU-based codec decode, especially some long-GOP media
Hardware encoding limited to one encoder path or export setting
Plugins that don't scale across GPUs
Fusion comps that hit CPU, RAM, or single-GPU limits
Insufficient VRAM on each card
Storage that can't read source media or write renders fast enough
Thermal throttling after a few minutes

For grading, the biggest win is often interactivity. If a colorist can play a heavier node tree in real time without disabling noise reduction or caching every shot, the room feels faster even before final render times improve. That's the right metric for a client-attended suite. For rendering, measure the complete deliverable. A 2x speedup in image processing doesn't mean a 2x faster export if the codec stage, audio muxing, or storage write is the limiter. Render a representative five-minute section instead of a single shot that makes the GPUs look good. In After Effects, the gains from the GPU are real but narrower. GPU-accelerated effects can improve substantially, but a second GPU usually doesn't change the day as much as artists hope. If CPU-only effects, expressions, source decode, RAM starvation, or slow cache bottleneck a comp, another GPU sits there looking expensive.

Common failure modes in real rooms

Most multi-GPU issues fall into a few buckets. Thermals are the first. Two high-end GPUs can dump hundreds of watts into a chassis for hours during renders. A system that passes a short benchmark can still throttle or crash 20 minutes into noise-reduced 8K playback. Watch sustained clock speeds, not just temperature snapshots.

A quick benchmark can pass even when sustained heat and power draw later cause throttling or crashes.

Power is next. Use a PSU with enough capacity and the right cabling. Avoid questionable splitter setups. GPU transient loads can expose weak power delivery even if average draw looks safe. Lane layout is another common trap. Many consumer motherboards have two long PCIe slots, but the second slot may not have the bandwidth or CPU lane connection you expect. Read the board manual before buying the second card. VRAM mismatch causes confusion. If one card has less memory, the smaller card may limit the whole multi-GPU setup for shared work. This is why identical GPUs are easier to support. Drivers can also bite. A driver update that improves one application can break a plugin, change GPU detection, or affect export stability. Keep a known-good installer available so you can roll back quickly. Finally, don't ignore the display path. Resolve can use a GPU for both GUI and processing, but a serious grading setup usually also has dedicated video I/O for the client monitor. Don't confuse GPU display output with color-managed video monitoring. Multi-GPU performance and accurate monitoring are related workstation topics, but they aren't the same problem.

A sensible buying decision

For a Resolve-first room, start with the strongest single GPU you can justify with enough VRAM for your highest-resolution work. If your test timelines show the GPU is saturated and VRAM is comfortable, a second matching GPU can be a smart upgrade. If VRAM is the problem, buy a larger GPU instead of adding a smaller second one. For an After Effects-first artist, buy one strong GPU, lots of RAM, a fast CPU with good per-core performance, and fast NVMe storage for cache. Add a second GPU only if another tool in the workflow needs it, such as Resolve, a GPU renderer, 3D rendering, or machine-learning processing. For a shared finishing workstation that runs both Resolve and AE, design for Resolve’s GPU appetite but don't starve AE’s CPU, RAM, and cache needs. The balanced version often looks like this: one or two matched high-VRAM GPUs, plenty of system RAM, fast local NVMe cache, a CPU with strong single-core and multicore performance, clean airflow, and conservative driver management. The best multi-GPU system is the one where the app, media, motherboard, drivers, and cooling all agree on what the machine is supposed to do.

FAQ

Yes, Resolve can benefit from multiple GPUs when the workload is GPU-bound. The clearest gains usually appear in heavy grading, temporal noise reduction, high-resolution debayering, scaling, GPU-accelerated OFX, HDR finishing, and render jobs where image processing is the bottleneck. If the timeline is limited by codec decode, storage, CPU, export encoding, or insufficient VRAM, a second GPU may deliver little improvement.

Usually not in a way that justifies buying a second GPU for After Effects alone. After Effects uses the GPU for specific effects, viewers, color management, and some rendering paths, but many comps are limited by CPU performance, RAM, cache speed, expressions, plugins, or project structure. Most AE users see better returns from more RAM, a stronger CPU, faster NVMe cache storage, and one better primary GPU.

In most Resolve and After Effects workstations, no. Two 24 GB GPUs generally don't behave like one 48 GB GPU. Shared workloads often require the same frame data on each card, so the practical VRAM limit can remain the capacity of one card. If a project needs more VRAM, one larger GPU is usually safer than two smaller GPUs.

SLI isn't required for Resolve and is generally not relevant to modern post-production GPU scaling. Resolve handles supported multi-GPU processing at the application level. NVLink, where available, isn't a universal switch that combines VRAM for all workflows. It shouldn't be assumed to solve memory limits unless the specific application and workflow explicitly support the behavior you need.

Important BIOS settings include the correct PCIe slot configuration, such as x8/x8 or x16/x16 where supported, Above 4G Decoding enabled for multiple large PCIe devices, stable UEFI settings, and an appropriate primary display GPU. Resizable BAR can be enabled if it's stable with the GPU and driver stack. The motherboard manual should be checked carefully because some long PCIe slots don't provide the bandwidth or CPU lane connection expected for heavy GPU compute.

When you're comparing one GPU versus two GPUs, different driver versions, or revised grades, the output files can multiply quickly. Aspect stacks versions on the same asset so the current render, previous passes, comments, and revision context stay together with version stacking.