Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteiTechGuides is reader-supported.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
Your RTX 5070 can sit at near-zero utilization while an AI app runs on integrated graphics because Windows, the app’s backend, or the installed CUDA software has selected a different device. Integrated graphics are the normal power-saving choice for light workloads, while the RTX 5070 is used only when the application requests a compatible CUDA device. Set the app to High performance in Windows 11, then confirm its own device setting, driver, toolkit, and model backend support Blackwell (compute capability 12.0). If any one of those layers is CPU-only or incompatible, a Windows preference alone cannot move inference to the RTX 5070.
Why is my RTX 5070 idle while integrated graphics run the AI app?
Windows supports systems with both an integrated GPU (iGPU) and a discrete GPU (dGPU). Microsoft describes integrated GPUs as generally less powerful but more power-efficient, and discrete GPUs as the choice for heavier processing. Windows can switch between them by task to balance performance and battery or system power.
That policy does not guarantee that every AI program will use the dGPU. The application must ask Windows for the high-performance adapter, then use a backend such as CUDA that understands the RTX 5070’s Blackwell architecture. A CPU-only runtime, an iGPU-specific backend, or a CUDA build without Blackwell-compatible kernels can leave the RTX 5070 idle even though Device Manager and Windows Graphics settings detect it.
Also distinguish an idle reading from a fault. Many AI workloads launch short GPU kernels, copy data, or wait for the CPU between steps. Utilization can therefore appear at zero when you look outside an active inference, or remain low when the model is too small to fill the card.
#1 Best Overall
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
How does Windows decide which GPU an AI application uses?
On Windows 11, the per-application Graphics preference is the first setting to check. Microsoft exposes three choices:
- Let Windows decide (Power saving): Windows chooses an adapter according to its policy.
- Power saving: normally the integrated GPU.
- High performance: normally the discrete GPU, such as an RTX 5070.
The preference applies to the executable you add, not necessarily every launcher or helper process. An AI tool may start a separate worker executable, service, or runtime process; that process can have a different preference.
How do I force the AI app to use the RTX 5070?
Use Windows’ documented per-app setting, then restart the application so it creates a new graphics and CUDA context.
- Open Settings > System > Display > Graphics.
- Under the app list, select the AI application. If it is missing, choose Browse and add the application’s actual
.exefile. - Select the app and click Options.
- Choose High performance, verify that the displayed adapter is the RTX 5070, and select Save.
- Close every window belonging to the app, including a tray icon or background worker, and launch it again.
If the program has its own device menu, choose the RTX 5070 or a CUDA backend there as well. Some applications expose labels such as “GPU,” “CUDA,” “NVIDIA,” or a device index; others require a startup option or configuration file. Windows cannot override a program that deliberately initializes CPU inference or an incompatible backend.
What hardware and software must support the RTX 5070?
The RTX 5070 is a Blackwell discrete GPU. NVIDIA lists the following current product specifications:
| Specification | RTX 5070 value |
|---|---|
| Architecture | Blackwell |
| CUDA cores | 6,144 |
| Memory | 12 GB GDDR7 |
| AI performance | 988 AI TOPS |
| Memory bandwidth | 672 GB/s |
| Total graphics power | 250 W |
| Recommended system power | 650 W |
| CUDA capability | 12.0 |
Those specifications describe what the hardware can provide; they do not mean every AI package is ready for it. NVIDIA’s CUDA architecture matrix lists Blackwell compute capability 10.0 and 12.0 as first supported by CUDA Toolkit 12.8. TensorRT-RTX has a separate requirement: its current packages require CUDA 12.9 Update 1 or CUDA 13.4, plus a driver that meets the selected toolkit’s minimum. A framework may bundle its own CUDA components, so installing a newer system toolkit does not automatically upgrade the runtime inside an application.
Rank #2
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
Check the requirements for the exact framework, not only the card’s driver version. A package built for older architectures can detect the RTX 5070 and still fail when it tries to launch a kernel.
How can I verify that the CUDA stack sees the card?
Perform these checks while the application is closed, then repeat monitoring during a real inference.
- Open a terminal and run
nvidia-smi. Confirm that the RTX 5070 appears, note the installed driver version, and check whether the AI process appears when it is running. - If you installed the CUDA Toolkit separately, run
nvcc --version. Compare the reported toolkit with the framework’s documented minimum. An application that bundles CUDA may not use this system-wide compiler, so treat this as a toolkit check rather than proof that the app is using it. - Start a model inference that lasts long enough to observe. Watch the RTX 5070 in
nvidia-smior Task Manager’s GPU graphs while tokens or results are actively being produced. - Record GPU memory usage as well as utilization. A model can reserve VRAM while spending most of its time waiting on CPU preprocessing, disk input, or synchronization.
Do not diagnose from an idle desktop snapshot. Utilization outside an inference says only that no substantial GPU work was occurring at that moment.
What if the app still uses the iGPU after I select High performance?
Work through the layers in this order, because each can independently select or reject the RTX 5070.
The executable is not the one you configured
Launchers often hand work to a worker process. In Task Manager, identify the process that consumes CPU or memory during inference, then add that executable in Graphics settings. Configure each persistent worker only if the application documentation says it is safe to do so.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The app is configured for CPU or an iGPU backend
Look for a device, backend, or accelerator setting inside the app. Select CUDA or the NVIDIA device instead of CPU, DirectML-on-iGPU, or an automatic mode that has cached an earlier choice. Restart after changing it; many runtimes choose a device only during initialization.
Rank #3
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
The CUDA runtime is too old
Blackwell support begins with the toolkit and package versions described by the framework. A current display driver alone cannot add missing Blackwell kernels to an old application bundle. Install a release that explicitly lists compute capability 12.0 or Blackwell support, following that project’s upgrade instructions.
The model does not fit in available VRAM
The RTX 5070 has 12 GB of GDDR7. If a model exceeds available VRAM, software may offload layers to system memory or the CPU. That can produce low GPU utilization and slower generation even though CUDA is enabled. Reduce model size or precision, shorten context, or enable the application’s documented layer-offload controls before concluding that the card is unused.
The workload is too light to occupy the card
Small prompts, embedding batches, or short image operations may complete before monitoring catches them. Increase the batch or run a sustained inference only when the application supports that change; do not inflate a workload merely to force a utilization graph to move.
What do “no kernel image” and similar CUDA errors mean?
NVIDIA’s compatibility guidance explains that a kernel launch fails when neither a compatible compiled binary (cubin) nor forward-compatible PTX is available for the GPU. On a Blackwell card, an application can therefore recognize the device yet fail at execution because its packaged kernels target older architectures without a usable fallback.
Errors may be worded as “no kernel image is available for execution on the device,” “invalid device function,” or a related architecture mismatch. Treat those messages as compatibility failures, not as evidence that the RTX 5070 is physically defective.
- Capture the complete error and the application, framework, driver, and CUDA versions.
- Check whether the framework release explicitly supports Blackwell or compute capability 12.0.
- Upgrade to a release containing compatible cubin or PTX, or rebuild the relevant extension with the architecture settings recommended in NVIDIA’s Blackwell guidance.
- Retest with a small model before restoring your full workload.
Should I disable integrated graphics?
Usually no. The iGPU is useful for display output and light tasks, and Microsoft’s design intentionally uses it when power efficiency matters. Disabling it can remove a display path, increase dGPU idle power, or break a laptop’s display-switching design without fixing an application that lacks CUDA support.
Rank #4
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Broad Compatible: The usb c to usb adapter widely supports all USB C/ USB A devices like laptops, tablets, cellphones, car chargers, and phone chargers. Such as compatible with MacBook Pro/Air 2023/2022, Thunderbolt 4/3 Devices,Apple MagSafe Watch 9/8/7/SE/Ultra, iPad Pro 2022/2021, Samsung Galaxy S23/S20/S10, and iPhone 17/16/15 Pro. Plug and play
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
Use per-app selection instead. Reserve firmware or Device Manager changes for a documented hardware or driver problem, and make a recovery plan for restoring the display adapter before changing it.
Free tools Windows power users keep installed
One-click scans. No signup required.
Which GPU is better for everyday AI work?
The answer depends on workload size, latency targets, and whether the software supports CUDA on Blackwell.
| Decision factor | Integrated GPU | RTX 5070 |
|---|---|---|
| Light, occasional tasks | Typically lower power draw and less fan noise | Usually unnecessary unless the app requires CUDA |
| Latency or tokens per second on a sustained workload | Generally lower compute headroom | Appropriate when the model and backend are CUDA-compatible |
| Memory capacity | Uses shared system memory; capacity depends on the computer and allocation | 12 GB dedicated GDDR7 |
| Power and cooling | Normally more power-efficient for light work | 250 W total graphics power; system cooling and a 650 W recommended power supply matter |
| Software compatibility | Depends on the app’s iGPU or graphics API backend | Requires a driver and framework with Blackwell compute capability 12.0 support |
| Per-app selection | Windows may label it Power saving | Windows may label it High performance |
There is no benefit to forcing every tiny task onto the RTX 5070. Choose it for sustained, compute-heavy inference when its memory and CUDA acceleration reduce response time enough to justify the additional power and heat.
Does Windows AI Foundry guarantee RTX 5070 acceleration?
No. Windows AI Foundry APIs run locally on supported Windows 11 devices, but each API has its own hardware matrix. Microsoft’s Phi Silica requirements include NVIDIA GeForce RTX 30-series and newer GPUs with at least 6 GB of VRAM, the latest manufacturer driver, and Developer Mode for GPU inference. An RTX 5070 meets the stated GPU generation and memory threshold, yet that support applies to Microsoft’s API path, not automatically to every third-party AI application.
For a third-party tool, follow that tool’s backend and architecture requirements. Its model format, runtime, and packaged kernels determine whether the card is actually used.
The Tool Desk
Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →What is the fastest troubleshooting sequence?
- Set the correct executable to High performance under Settings > System > Display > Graphics.
- Restart the app and select CUDA or the RTX 5070 in its own settings.
- Confirm the driver with
nvidia-smiand, where relevant, the toolkit withnvcc --version. - Run a sustained inference and watch utilization, VRAM, and the process list rather than an idle desktop.
- If an architecture error appears, install a Blackwell-compatible release or rebuild with compatible cubin/PTX.
- If memory is exhausted, reduce the model or context, adjust documented offload settings, and retest.
This sequence separates Windows selection problems from application configuration, resource limits, and binary compatibility. It also avoids the common mistake of reinstalling drivers when the real issue is a CPU-only backend or an outdated framework.
Best Value
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
When is the RTX 5070 working as expected?
During a sustained CUDA-compatible inference, the RTX 5070 process should appear in nvidia-smi, dedicated memory usage should rise when the model is resident, and GPU activity should coincide with generation or other compute stages. Utilization need not remain at 100 percent: preprocessing, synchronization, small batches, and memory transfers can create gaps.
If those signs appear and output speed meets your workload’s needs, intermittent idle periods are normal. If the process never appears, memory never changes, or a kernel-compatibility error is logged, continue troubleshooting the application and CUDA stack rather than disabling the integrated GPU.
Frequently Asked Questions
Will choosing High performance make the RTX 5070 run all the time?
No. The setting affects which adapter an application may use; it does not force continuous computation. The card can return to a low-power state between inference calls, and brief kernels may not produce a sustained utilization reading.
Can an AI model use both the integrated GPU and RTX 5070 at once?
Only if the application explicitly supports multi-device execution or layer offloading. Windows’ per-app preference chooses a graphics adapter; it does not automatically split a model across GPUs.
Why does Task Manager show video-engine activity but little 3D usage?
AI frameworks may use compute, copy, or video engines rather than the 3D engine graph. Select the compute-related graph or confirm the process and memory usage with nvidia-smi during inference.
Does more VRAM always mean faster token generation?
No. VRAM capacity mainly determines whether the model and context fit without offloading. Speed also depends on kernel support, precision, batch size, memory bandwidth, and how much work the model gives the GPU.
What should I do before upgrading CUDA on a working AI installation?
Record the application’s current runtime, driver, and model versions, then read the framework’s upgrade notes. Some applications bundle a tested CUDA runtime, so replacing system components can create a mismatch; keep a rollback installer or restore point.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Clear out junk files and repair common Windows errorsFree Scan →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.
