Short answer: Early Ashes of the Singularity testing showed a real advantage for AMD’s GCN architecture when a game mixed graphics and compute work asynchronously. Nvidia’s Maxwell GPUs could expose related functionality and remained compatible with DirectX 12, but Oxide found that enabling async compute on Maxwell was much slower than disabling it for that workload. The result was an architectural and engine-specific finding—not proof that AMD won every DX12 game or that Maxwell was “not DX12.”
This was a 2015-era controversy, when DX12 games, drivers and hardware paths were still developing. It is useful history, not a performance forecast for current Radeon or GeForce products.
What the original controversy was about
The early Ashes of the Singularity results raised a specific question: did DirectX 12 expose a major AMD advantage because GCN was better at asynchronous compute? Contemporary coverage, including ExtremeTech’s report, often compressed several separate issues into a yes-or-no argument about “DX12 support.”
The evidence was more nuanced. AMD GCN was well suited to the asynchronous workload Oxide tested. Maxwell was not simply incapable of DirectX 12 or of every form of concurrent GPU work; its scheduling and switching behavior made this particular implementation inefficient enough that Oxide used a separate Nvidia path with async compute disabled.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minute#1 Best Overall
- Powered by Radeon RX 9070 XT
- WINDFORCE Cooling System
- Hawk Fan
- Server-grade Thermal Conductive Gel
- RGB Lighting
What asynchronous compute actually means
Async compute lets an engine submit compute jobs—such as particles, lighting, post-processing or other GPU calculations—so they can be scheduled alongside graphics work instead of waiting for the graphics queue to become completely idle. The goal is to fill otherwise unused execution capacity.
That description does not mean that every shader runs literally simultaneously. Four related capabilities must be separated:
- Multiple queues: whether graphics and compute commands can be submitted through separate queues.
- Concurrent execution: whether different work types can occupy GPU resources at the same time.
- Scheduling: how effectively the hardware distributes those jobs among its execution resources.
- Preemption and context switching: how quickly active work can be interrupted, suspended or interleaved.
An engine also has to produce suitable independent jobs. If graphics already saturate the GPU, or if dependencies require compute to wait for graphics, adding another queue may provide no gain and can add synchronization overhead.
Is async compute required for DirectX 12?
No. DirectX 12 is a broad low-level API. It provides explicit resource management, command lists, descriptor management and tools for reducing CPU driver overhead, among other capabilities. Async compute is one optional execution strategy within that model, not a definition of DX12 itself.
Rank #2
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
A game can use DX12 and benefit from lower CPU overhead while making little or no use of asynchronous compute. Conversely, an exposed async-compute capability does not guarantee useful performance; the driver, engine and workload still determine the result. Oxide’s developer explicitly said async compute was not a specific D3D12 requirement in the developer discussion.
Why AMD GCN attracted attention
GCN included Asynchronous Compute Engines, commonly called ACEs, intended to manage compute workloads separately from traditional graphics work. AMD had also emphasized general-purpose GPU computing, and the queueing model fit the low-level, console-influenced engine designs being discussed at the time.
That architecture could be advantageous when a workload had enough independent compute to overlap with graphics. It did not make every game faster automatically. The engine still had to submit useful work, the GPU needed available capacity, and barriers or synchronization could erase the benefit. Contemporary technical commentary about GCN and ACEs should therefore be read as architectural explanation, not as a universal benchmark result.
What Maxwell’s situation was
The careful statement is not “Maxwell could not do async compute.” Maxwell drivers could report or expose related API functionality, and Nvidia GPUs remained DirectX 12 products. The practical issue was efficiency in the workload Oxide tested.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsRank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
Oxide reported that enabling async compute on Maxwell was substantially slower than leaving it off and described the result in extremely negative terms. The team therefore created a separate Nvidia path that disabled the feature. The likely distinction was not a total absence of concurrent execution, but the way Maxwell scheduled and switched between graphics and compute work.
| Term | What it tells you |
|---|---|
| API or feature exposure | The driver reports that an operation is available. |
| Hardware behavior | The architecture’s queues, schedulers and execution resources determine how it runs. |
| Practical performance | A specific engine may gain, break even or lose performance when the feature is enabled. |
| DX12 feature-level support | A broader compliance category; it does not promise equivalent performance for every optional technique. |
| Game support | The developer may enable, disable or tune a path separately for each vendor. |
What Oxide said about Ashes of the Singularity
Oxide’s comments, reproduced in the Linus Tech Tips forum discussion, provide the most direct account of the disputed behavior.
- Ashes used a modest amount of async compute rather than making the entire game an extreme demonstration of the feature.
- Async compute produced a noticeable improvement on AMD hardware in the tested build.
- Turning it on for Nvidia hardware was slower, so Oxide disabled it on that path.
- Nvidia had actively collaborated with Oxide during development; the result was not explained by a simple claim that Nvidia had been excluded.
- The game was not intended to be a definitive showcase of every advanced GCN capability.
Oxide also relayed early reports from console developers of gains as high as roughly 30% in some GPU workloads. That was an uncertain workload-specific estimate, not a general PC-game multiplier. Results depend on queue occupancy, shader balance, synchronization, driver behavior and whether the GPU has spare resources to fill.
How to read the benchmark correctly
CPU overhead and GPU throughput were different stories
Oxide said Nvidia’s DX12 CPU overhead was better than its DX11 overhead in the tested engine, while also believing async compute was helping AMD’s GPU performance. Those observations can both be true. DX12 can improve command submission and CPU scaling independently of how a GPU schedules overlapping compute.
The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Rank #4
- AI Performance: 767 AI TOPS
- OC mode: 2632 MHz (OC mode)/ 2602 MHz (Default mode)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- A 2.5-slot design maximizes compatibility and cooling efficiency for superior performance in small chassis
The AMD result was workload-specific
Ashes mixed graphics and compute in a way that favored GCN’s approach. A graphics-heavy game with little independent compute, or an engine that synchronized frequently, could produce a different ranking. Average frame rate alone also would not reveal every scheduling or frame-time trade-off.
Drivers and engine maturity mattered
The game, drivers and vendor paths were still being tuned. Comparing different builds or driver versions could change the outcome, and a feature toggle only shows the net result after scheduling and synchronization costs—not the theoretical capability in isolation.
Async compute is not preemption
These terms were frequently blurred in 2015. Async compute concerns submitting and scheduling compute work alongside graphics. Preemption concerns interrupting or suspending active work so another task can run. A GPU can support multiple queues without offering low-latency interruption at arbitrary points.
Preemption and context-switch costs matter especially when a workload needs rapid handoffs, such as latency-sensitive rendering. Efficient asynchronous execution does not require every shader to be interrupted at any instruction; engines can instead divide work into schedulable batches. The period’s technical discussion of this distinction appears in the Oxide discussion.
Free tools Windows power users keep installed
One-click scans. No signup required.
Best Value
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Tier 2 versus Tier 3 binding was a separate issue
Early DX12 coverage also mixed up resource-binding tiers with async compute. Oxide said Nvidia’s Tier 2 and AMD’s Tier 3 binding hardware were separate from asynchronous execution and were not expected to be a significant performance issue in Ashes.
This is a useful warning: DX12 contains many capabilities, and a difference in one feature does not automatically explain a result attributed to another.
What the benchmark established—and what it did not
| It established or strongly suggested | It did not establish |
|---|---|
| AMD GCN could gain from async compute in the tested Ashes workload. | That AMD would beat Nvidia in every DX12 game. |
| Maxwell’s async path was inefficient enough for Oxide to disable it. | That Maxwell lacked DX12 support or all asynchronous execution. |
| DX12 could alter the balance between CPU overhead and GPU throughput. | That DX12 inherently favored AMD. |
| Vendor-specific rendering paths could be necessary. | That a single benchmark predicted Pascal, later Nvidia designs or modern GPUs. |
| Architecture and workload mix mattered. | That the game was proven biased solely by Stardock’s AMD marketing relationship. |
Oxide said Nvidia had been involved in development, in some respects more actively than AMD during the relevant period. That is developer testimony, not independent proof of neutrality. The responsible conclusion is that a marketing relationship deserved disclosure, while the measured scheduling behavior still required technical explanation.
What a 2015 GPU buyer should have concluded
- Do not choose a GPU solely from one early Ashes result.
- Treat AMD’s async-compute advantage as a legitimate architectural consideration if the games you play use mixed graphics and compute workloads.
- Compare the games, resolution, CPU, power requirements and driver history that actually matter to your system.
- Look for results across multiple engines and mature driver releases before making a universal DX12 prediction.
- Do not interpret “full DX12 support” as equivalent performance for every workload-dependent feature.
Why the historical lesson still matters
The durable lesson is about evaluating low-level APIs. A feature label describes an interface contract; it does not describe how efficiently every architecture executes every workload behind that contract. Queueing, overlap, scheduling, preemption, synchronization, driver quality and engine design all contribute to the final frame time.
That is why the early AMD-versus-Nvidia argument could contain several true statements at once: AMD GCN could be faster in the tested async-compute pattern, Nvidia could show strong CPU behavior in the same engine, and neither observation could predict the entire DX12 market. For current hardware, consult the Microsoft DirectX 12 programming guide and modern independent benchmarks rather than carrying Maxwell-era conclusions forward unchanged.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
