iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
A game’s ordinary rendering runs on the GPU’s graphics engines; an AI accelerator is involved only when the game dispatches machine-learning work, such as a supported DLSS, XeSS, ML-based FSR, or DirectML feature. To check, identify the feature in the game’s settings, compare the same scene with it on and off, and watch GPU-engine activity and frame counters. Task Manager’s 3D percentage alone cannot prove that AI-specific cores are working.
What does it mean for a game to use an AI accelerator?
A graphics card is not a single undifferentiated processor. It contains independently scheduled GPU engines, such as 3D, copy, and video, and may also contain specialized units designed to accelerate particular kinds of work. When a game renders a scene, its graphics workload runs on GPU graphics hardware. When the game invokes machine-learning work, compatible hardware may accelerate that work on AI-focused units.
That distinction matters because the words “GPU usage” in an overlay usually describe a broad workload or engine, not the activity of a specific AI unit. Microsoft’s DirectX engineering explanation describes a GPU engine as an independent unit of silicon that can be scheduled and can operate in parallel with other engines. An active 3D engine therefore shows that graphics work is happening; it does not identify which internal units performed every part of that work.
AI acceleration is also not an alternative to using the GPU. A game can render on its GPU while also dispatching machine-learning operations to specialized units on that same GPU. The feature must be supported by the game or its middleware, and the result depends on the relevant game, hardware, driver, API, and feature version.
#1 Best Overall
- 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
Which game features can use AI hardware?
Look for a named feature in the game’s graphics or video settings. A menu option establishes that the game offers an integration; telemetry and a controlled comparison help establish whether it is active in the scene you are testing.
| Technology | What its vendor documents | What to check in a game |
|---|---|---|
| NVIDIA DLSS | NVIDIA describes DLSS as neural-rendering technology powered by RTX Tensor Cores. Its suite includes Super Resolution, Frame Generation, Ray Reconstruction, and DLAA. | Check for the specific DLSS option and whether it is enabled. If Frame Generation is on, record generated-frame FPS separately from base game FPS. |
| Intel XeSS | Intel describes XeSS as AI-based upscaling. Intel’s implementation uses DirectX 12 compute-shader passes and supports Arc and other GPUs. | Check the game setting and Intel’s supported-games list. Intel says the list changes and advises checking with the game developer if a title is absent. |
| AMD FSR | AMD documents game-integrated upscaling and frame-generation features, including ML-based paths for supported titles. | Check which FSR version and feature the game offers, whether it is enabled, and whether AMD Software lists support for the title. |
| Microsoft DirectML | DirectML is a low-level, vendor-agnostic machine-learning API. It can execute ML operators on a GPU or on AI-acceleration cores when available. | Consumer overlays may not label a DirectML dispatch as AI. Application documentation, API logs, or a developer profiler can provide more specific attribution. |
These technologies should not be treated as interchangeable labels. A game may offer one feature without offering the others, and a feature’s implementation and hardware requirements can vary by title and version. For example, the presence of a setting called “Frame Generation” does not, by itself, tell you which other upscaling or reconstruction features are active.
How can I check whether a game is using an AI feature?
Use a repeatable on/off comparison rather than relying on one utilization reading. Keep the scene, resolution, quality preset, and camera path as consistent as possible so the change you observe is more likely to come from the feature you toggled.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
- Record the hardware and driver. Note the GPU model and its driver version. These help you interpret whether the game’s feature is supported by your setup.
- Check the game’s graphics or video settings. Find the exact option, such as DLSS, XeSS, FSR, Frame Generation, or Ray Reconstruction. Record whether it is available and enabled; do not assume that a similarly named option represents the same feature across games.
- Choose a repeatable scene. Use the same save, benchmark sequence, or in-game location. Keep resolution, quality preset, and camera movement constant for the comparison.
- Record overlay readings. Steam’s performance monitor can show GPU utilization, video-memory use, temperature, and frame rates. If the overlay provides separate base-game and generated-frame FPS, record both.
- Watch GPU engines in Task Manager. Open Task Manager > Performance > GPU and observe the 3D, compute, copy, and video engines while repeating the scene. Engine graphs show activity by engine, not an exact AI-core cycle count.
- Turn the feature off and repeat. Reload or restart the scene if needed, then compare frame time, image reconstruction, frame counters, and engine activity. Change only the feature under test; otherwise, the comparison becomes harder to interpret.
- Use developer-level tools for stronger attribution. Unreal Insights, RenderDoc, vendor profilers, or API logs can expose more detail than consumer overlays. Profiling adds some overhead, so its measurements may differ from normal play.
Make the A/B test reproducible
Begin with a short test plan. Write down the GPU model, driver version, game build, display resolution, quality preset, selected upscaler mode, and whether frame generation is enabled. Then choose a camera path that can be repeated: a built-in benchmark, a fixed save point, or a short route through the same area. Record several passes in each state rather than relying on a single instant, because a scene can briefly load assets or change effects while you are watching it.
Run the feature-enabled pass first or second consistently, and reload the scene when the game requires it. Do not change the resolution, quality preset, ray-tracing option, or other upscaling mode at the same time. If you test Frame Generation, keep the base upscaler setting documented, since generated-frame output and reconstruction are separate controls in some games.
For each pass, note average or displayed frame rate, frame-time behavior, GPU utilization, video-memory use, temperature, and the 3D, compute, copy, and video graphs. Also note visible image changes such as reconstruction quality or the appearance of generated motion. The goal is not to produce a universal percentage; it is to see whether the named setting changes the same scene in a repeatable way.
What do Task Manager and FPS overlays actually tell me?
Task Manager’s GPU graphs are useful for seeing which broad engines are active, but they are not a direct meter for Tensor Cores, matrix units, or other AI-specific hardware. A rise in 3D or compute activity while a feature is enabled is compatible with the feature doing work, but it does not prove that a particular AI unit handled it. Likewise, a 3D graph that remains busy does not mean that the game ignored its AI feature: rendering and machine-learning work can occur together.
Steam says its performance monitor is intended to help show whether a game is bottlenecked on the GPU. Treat utilization as a simplified indicator, not as a complete account of all GPU work. A percentage from one engine cannot establish which internal hardware unit ran a particular operation.
Rank #2
- 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
Frame Generation needs separate interpretation. Steam’s overlay can distinguish actual game FPS from generated-frame FPS. Generated frames can make motion appear smoother, but they do not represent an equivalent amount of game simulation or input-processing work. A higher generated-frame figure is therefore not the same measurement as a higher base game frame rate.
Reading the common graph patterns
- High 3D activity in both tests: expected when the game is rendering normally. It does not rule AI work in or prove it.
- Compute activity rises with DLSS, XeSS, FSR, or another named option: the toggle changed the workload, but the graph still does not identify whether dedicated AI units or general compute resources performed it.
- No obvious graph change: the feature can still be active. The operation may be brief, overlapped with rendering, or hidden by the graph’s sampling and aggregation.
- GPU utilization falls while frame rate rises: the feature may have reduced the amount of native-resolution rendering, but this is not a direct measurement of AI-core use.
- Generated FPS is much higher than base FPS: the display is receiving interpolated or generated images; compare the base value when judging simulation and input workload.
Task Manager also shows copy and video engines. Activity there can reflect transfers, decoding, or other work and should not be read as an AI indicator. The engine labels describe scheduling domains, not every internal execution unit inside the GPU.
How do the vendors’ features differ?
NVIDIA DLSS
NVIDIA describes DLSS as “a suite of neural rendering technologies powered by NVIDIA RTX Tensor Cores.” The suite includes Super Resolution, Frame Generation, Ray Reconstruction, and DLAA, so identify the exact component selected in the game. Super Resolution changes reconstruction from a lower internal resolution; Frame Generation changes the number of displayed images; Ray Reconstruction addresses ray-traced image reconstruction; and DLAA is an anti-aliasing mode. A menu that says DLSS may expose several of these separately.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →An RTX GPU and a compatible game integration are required for the documented Tensor Core path. In a test, record whether each DLSS component is enabled, then compare the base-game frame rate and any generated-frame value separately. A high 3D graph is expected because the game still renders the scene.
Intel XeSS
Intel calls XeSS, or Xe Super Sampling, an AI-based upscaling technology designed to provide a smoother and more consistent gameplay experience. Intel describes its implementation as DirectX 12 compute-shader passes and maintains a supported-games list. That means the presence of XeSS does not make every GPU or every title behave identically.
Check the game’s XeSS setting, the selected quality mode, and whether the title appears on Intel’s current supported-games list. Intel notes that the list changes and advises contacting the developer when a title is absent. During testing, watch both 3D and compute activity, but do not treat either graph as a direct XeSS-core counter.
AMD FSR
AMD documents FSR as game-integrated upscaling and frame-generation technology, with current documentation also covering ML-based paths for supported titles. Therefore, “FSR” alone is not enough information: record the FSR version or feature name shown by the game and whether the title exposes an ML-based path.
Check the in-game toggle and AMD Software’s support information for the title, then repeat the same scene with the feature disabled. Compare frame time, image reconstruction, base FPS, and generated FPS when available. Hardware and driver support can differ by feature and version.
Rank #3
- 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
Microsoft DirectML
Microsoft defines DirectML as a low-level API for machine learning. It is vendor-agnostic and can execute ML operators on a GPU or on available AI-acceleration cores. A game can therefore use DirectML without exposing a consumer-facing setting labeled “AI accelerator.”
Because ordinary overlays may not label a DirectML dispatch, definitive attribution generally requires application documentation, API logs, or a developer profiler. A DirectML workload can coexist with regular 3D rendering, and the same Task Manager limitations apply.
What evidence is strong enough to make a conclusion?
- Menu evidence: a visible, enabled feature shows that the game includes the integration and intends to use it. It does not prove that every scene dispatches the operation.
- Controlled A/B evidence: repeatable changes in frame time, image reconstruction, generated-frame output, or engine activity after toggling one feature support the conclusion that the setting affects the tested scene.
- Hardware evidence: the GPU model and driver establish whether the documented path is supported, but they do not prove that the game invoked it at a particular moment.
- Developer-level evidence: Unreal Insights, RenderDoc, vendor profilers, or API logs can reveal dispatches and timing details that consumer overlays hide. Profiling itself adds overhead, so compare profiler captures with normal-play observations.
The strongest consumer conclusion is usually phrased narrowly: “With this GPU, driver, game version, and scene, enabling this named feature changed the workload and image or frame behavior.” Claiming an exact number of AI-core cycles requires instrumentation that the usual overlays do not provide.
Recommended Free Tools
Why might the result be inconclusive?
Consumer telemetry does not provide a universal counter for exact AI-core cycles. The available evidence is indirect: a game setting, broad engine activity, frame-time or image changes, and, at a deeper level, developer profiling. Each answers a different question, so avoid treating any single percentage as definitive proof.
Support can vary with the game, driver, GPU generation, API, and feature version. A setting may be unavailable on one configuration, implemented differently across titles, or absent from a scene that does not invoke the relevant operation. Intel’s supported-games list can change; if a game is not listed, Intel advises checking with the developer.
Keep the comparison controlled. Changing resolution or a quality preset at the same time as an AI feature can change workload and frame rate for reasons unrelated to that feature. A scene with different camera movement or visual effects can also produce different readings. If the game requires a reload or restart for a setting to take effect, repeat the test after reloading rather than assuming the menu toggle applied immediately.
When the question is simply whether a feature is functioning, compare the image and frame behavior with the feature enabled and disabled. When the question is specifically which hardware units executed the work, broad consumer monitoring may not be enough; use a profiler or API logs, and account for the overhead those tools introduce.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Practical pitfalls to avoid
- Confusing a feature name with a hardware guarantee: DLSS, XeSS, and FSR describe software integrations whose supported paths depend on the title, GPU, driver, and version.
- Using one screenshot as proof: a single utilization value cannot show which engine or internal unit performed a dispatch.
- Mixing frame types: always label base-game FPS and generated-frame FPS separately.
- Changing several settings: an FPS increase after changing resolution and an upscaler together cannot be attributed to either one with confidence.
- Ignoring scene state: reload or restart when required, and repeat the same camera path after the setting is applied.
- Overinterpreting a missing option: an unavailable feature can reflect hardware, driver, game-version, or API limitations rather than a general absence of AI acceleration.
- Assuming all AI work is visible: short or overlapped dispatches may not create an obvious separate graph pattern.
Summary checklist
- Write down the GPU model, driver, game version, resolution, and quality preset.
- Identify the exact DLSS, XeSS, FSR, Frame Generation, Ray Reconstruction, or other ML option.
- Confirm the option is enabled and note whether a reload or restart is required.
- Run a repeatable scene and record base FPS, generated FPS if shown, frame time, GPU utilization, memory, temperature, and engine graphs.
- Disable only the feature, repeat the same scene, and compare the results.
- Use a developer profiler or API log when you need attribution beyond broad engine activity.
Frequently Asked Questions
Can a game use AI acceleration for something other than upscaling?
Yes. DirectML is a general machine-learning API, not an upscaling feature by definition, and a game or application can use ML operators for different tasks. Whether a particular game does so requires evidence about that title’s implementation.
Rank #4
- Powered by the NVIDIA Blackwell architecture and DLSS 4 OC mode: 2640MHz/Default mode: 2610MHz (Boost Clock)
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.125-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
How do I know if DLSS is running on Tensor Cores?
Confirm that the game exposes and enables a DLSS component on a compatible RTX setup, then compare the same scene with DLSS on and off. NVIDIA documents DLSS as powered by RTX Tensor Cores, but ordinary Task Manager graphs do not provide a direct Tensor Core cycle counter. Exact attribution requires developer-level profiling or API evidence.
Does XeSS use dedicated AI hardware?
XeSS is an AI-based upscaler, and Intel documents its implementation as DirectX 12 compute-shader passes for Arc and other GPUs. The game, GPU, driver, and feature path determine the actual execution. Compute or 3D activity alone cannot identify a dedicated AI unit.
Does FSR always use AI hardware?
No universal conclusion follows from the FSR label. AMD documents multiple game-integrated upscaling and frame-generation paths, including ML-based paths for supported titles. Check the exact FSR version or feature exposed by the game and test it on your hardware.
Can Task Manager show whether a game is using AI cores?
Task Manager can show broad GPU-engine activity such as 3D, compute, copy, and video. It does not normally expose a universal counter for Tensor, matrix, or other AI-specific cycles, so use controlled toggling and, for definitive attribution, a developer profiler or API logs.
Why can GPU utilization stay high when I turn on an AI feature?
The game still has to render its scene, and graphics and machine-learning work can occur in parallel. An unchanged broad utilization figure does not establish that the feature failed or that no specialized hardware was used.
Does a game setting labeled DLSS or XeSS guarantee the feature is active in every scene?
No. The setting shows that the game includes an integration, but scene behavior and the selected feature can affect whether work is dispatched. An on/off comparison in the same scene is more informative than the menu label alone.
Why are generated FPS and game FPS different?
Frame Generation can add displayed frames without adding an equivalent amount of simulation or input-processing work. Steam’s monitor can report generated-frame FPS separately from actual game FPS; record both when evaluating performance.
Free tools Windows power users keep installed
One-click scans. No signup required.
What should I do if my game is missing from Intel’s XeSS list?
Intel says its supported-games list changes and advises checking with the game developer when a title is absent. Also verify the game version, GPU model, driver, and whether the title offers XeSS in its own settings.
Will profiling give the same numbers as normal play?
Not necessarily. Unreal Insights, RenderDoc, vendor profilers, and API logging can add overhead. Use them for stronger attribution, but account for that overhead and compare their findings with a normal-play A/B test.
The Bottom Line
Use the game’s named feature, a controlled on/off comparison, and separate base-versus-generated frame readings. Task Manager can show which broad GPU engines are busy, but only developer-level profiling or API evidence can make a strong claim about a specific AI accelerator.
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.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitches
