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
Use NVIDIA’s model lookup, not the GeForce, Quadro, or Tesla family name alone. First identify the exact GPU model, search it in NVIDIA’s CUDA GPU Compute Capability table (or the legacy table), and confirm that a compute-capability number is listed. A listed model is evidence that the hardware is CUDA-capable.
That check answers only the hardware question. A CUDA program can still fail because of its required compute-capability minimum, NVIDIA driver, CUDA Toolkit, operating system, compiler or other toolchain. Check those software requirements separately before installing or upgrading anything.
What it means for an NVIDIA GPU to support CUDA
CUDA is NVIDIA’s platform for running general-purpose computations on NVIDIA GPUs. Support is a property of a specific GPU architecture and model. NVIDIA describes CUDA as a standard feature in GeForce, Quadro and Tesla GPUs, as well as NVIDIA GRID solutions; the practical way to verify an individual card is still the model-by-model lookup in NVIDIA’s GPU tables. See the CUDA FAQ for NVIDIA’s product-family explanation.
Free tools Windows power users keep installed
One-click scans. No signup required.
The key value in the lookup is compute capability (CC). NVIDIA defines it as the hardware features and supported instructions for a GPU architecture. It is written as a major.minor number, such as 6.1, 8.6 or 12.0. A CUDA-capable model can execute CUDA code, but each application may support only a range of compute capabilities. Therefore, “supports CUDA” means the hardware has CUDA capability; it does not promise that every CUDA application or every Toolkit release will run on it.
#1 Best Overall
- 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
Do not confuse compute capability with the CUDA Toolkit version. Compute capability belongs to the GPU. The Toolkit is the software development package installed on the computer, and its version changes independently of the card’s CC number.
The reliable four-step check
- Identify the exact model. Record the complete name reported by the operating system or NVIDIA utility, including series and model number. “GeForce RTX” or “GTX” by itself is not precise enough.
- Open NVIDIA’s current table. Go to CUDA GPU Compute Capability and search the page for the model name. If the card is older and absent from the current list, check CUDA GPUs — Legacy.
- Read the CC column. A row containing your exact model and a major.minor compute-capability value confirms that NVIDIA lists the GPU as CUDA-capable. Save the value; software documentation may express it in a different notation.
- Compare the target program’s requirements. Check the application’s minimum compute capability, supported driver range, Toolkit or runtime version, operating-system support and compiler/toolchain requirements. A model-table match is necessary for many programs, but it is not a complete application-compatibility test.
How to identify the exact GPU model
Use nvidia-smi when the NVIDIA driver is installed
On a system with an installed NVIDIA driver, open a command prompt or terminal and run:
nvidia-smiThe command reports the NVIDIA devices visible to the driver, including the GPU name. Copy the model name exactly as shown. If more than one GPU appears, check each one; a CUDA application may select a different device from the one you expected.
Recommended Free Tools
If the shell reports that nvidia-smi cannot be found, the driver may not be installed, the utility may not be on the command path, or the machine may not expose an NVIDIA device to the current environment. That result is an identification problem, not proof that the hardware lacks CUDA. Use the computer’s built-in device-information view or the machine’s hardware documentation to obtain the model, then perform the table lookup.
Use the operating system’s device information
Every major desktop operating system provides a device-information or hardware-details view. Open the section that lists display adapters or graphics devices and copy the full NVIDIA product name. Menu labels vary by operating-system edition and release, so do not rely on a path written for a different version. The important output is the exact model, not the generic display-adapter category.
On a workstation, laptop, server or virtual machine, note whether the name identifies an NVIDIA GPU or only a virtual display adapter. A virtualized environment can expose a device name that differs from the physical card, so use the model visible to the CUDA software in that environment when you perform the lookup.
Record the model before changing drivers
Write down the model, operating system and driver version before troubleshooting. Driver changes can alter what utilities report, while the physical GPU model does not change. This small record prevents a broad family name or a previous card’s name from being used in the compatibility check.
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 glitchesHow to search NVIDIA’s compute-capability tables
Current GPU table
Open https://developer.nvidia.com/cuda/gpus and use your browser’s find function for the complete model name. Match the manufacturer’s wording as closely as possible. If several rows share a family name, compare the model number, memory variant or mobile designation rather than selecting the first similar result.
NVIDIA’s table, accessed on October 13, 2026, lists GeForce RTX 5060 with compute capability 12.0. That row is a concrete example of the format: the model is the lookup key and 12.0 is the architecture capability. It is not a statement that the card is appropriate for every workload or that it will fit every computer.
Legacy GPU table
Older products may not appear in the main table. Search https://developer.nvidia.com/cuda/gpus/legacy for the same exact model. NVIDIA’s legacy table lists GeForce GTX 1080 at compute capability 6.1. A legacy listing still establishes CUDA capability, but the age of the architecture makes the application’s supported driver, Toolkit and minimum-CC requirements especially important.
What “not found” means
Before concluding that a card is unsupported, check spelling, mobile or workstation suffixes, and the legacy table. Also verify that you copied the GPU model rather than the driver package name or an integrated-graphics label. If the exact model is absent from both NVIDIA tables, the available evidence does not establish CUDA support; consult the card or computer manufacturer’s specification and the software vendor’s requirements instead of guessing from the series name.
Understanding compute capability numbers
Compute capability is an architectural identifier, not a performance score. A higher number generally represents a newer instruction and feature set, but it should not be used as a substitute for benchmarks, memory capacity or application-specific performance data.
CUDA documentation and build systems often remove the decimal point when naming a target. A compute capability of 8.6 is written as sm_86; the digits still represent major version 8 and minor version 6. Keep both forms in your notes so you can translate an application’s requirement correctly. For example, a program that asks for sm_86 is referring to CC 8.6, not to CUDA Toolkit 8.6.
Rank #2
- 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
When a program publishes a minimum CC, compare numbers as architecture versions, not as decimal software versions. If the requirement is CC 7.5, a card listed at 6.1 does not meet that minimum even though both are CUDA-capable. Conversely, a card listed at 12.0 may still be unusable if the program or its binaries do not support that newer architecture.
Hardware capability versus a working CUDA installation
NVIDIA’s getting-started guidance treats a CUDA-capable GPU as one part of a usable setup. You also need a supported operating system, a compatible compiler or toolchain and the CUDA Toolkit when the application requires development components. A precompiled application may bundle or download a runtime, while a project that compiles CUDA code normally needs the Toolkit and a supported compiler.
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 →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Check these layers independently:
| Layer | What to verify | Why it matters |
|---|---|---|
| GPU hardware | Exact model appears in NVIDIA’s current or legacy GPU table and has a CC value. | Establishes that the device architecture is CUDA-capable. |
| Compute capability target | The application or its binaries support your model’s CC, including any stated minimum or maximum. | Code can fail to launch when required instructions or architecture targets are unavailable. |
| Driver | The installed NVIDIA driver falls within the application’s supported range. | The driver mediates access to the GPU and must support the requested runtime and architecture. |
| Operating system | Your OS edition and version are listed as supported. | CUDA packages and drivers are released for specific operating-system environments. |
| Toolkit and compiler | The required CUDA Toolkit release and compiler/toolchain are installed and compatible. | Source builds depend on compiler, headers, libraries and architecture-targeting support. |
| Application build | The application’s own release notes state support for your GPU generation and software stack. | Applications can impose stricter limits than CUDA hardware itself. |
Release requirements change over time. NVIDIA’s CUDA Toolkit Archive listed Toolkit 13.4.2 as the latest release in September 2026. That is a dated archive listing, not a hardware-support cutoff and not a guarantee that every GPU or application supports that Toolkit. Verify the current release and its compatibility matrix when you install.
Check the CUDA requirements of the program you want to run
Find the application’s minimum architecture
Look in the program’s installation guide, release notes or system-requirements page for terms such as “minimum compute capability,” “supported architectures,” “CUDA capability,” or architecture targets like sm_86. Record the exact requirement for the version you plan to install; an older release may support a different set of GPUs than a current release.
Check driver and Toolkit constraints
Record the minimum driver version and the CUDA Toolkit or runtime version named by the application. A card can pass NVIDIA’s hardware lookup while the installed driver is too old, the Toolkit is not installed, or the selected Toolkit is incompatible with the operating system or compiler. Follow the application’s versioned installation instructions rather than assuming that the newest Toolkit is automatically the right one.
Check how the application was built
Some software ships binaries for a fixed list of architectures; other projects compile kernels on the user’s machine. A prebuilt package may omit support for an older CC even though the GPU is CUDA-capable. A source build may require you to select an architecture target such as sm_86. Read the project’s build instructions before changing hardware.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Troubleshooting common results
“My GPU is not listed”
- Copy the model from
nvidia-smior the operating system’s device details again, including mobile, workstation and generation suffixes. - Search the legacy table as well as the current table.
- Make sure you are searching for the GPU model, not the driver name, laptop series or a virtual display adapter.
- If neither table contains the exact model, do not infer support from a similarly named card. Use the manufacturer’s specifications and the target application’s documented hardware list.
“nvidia-smi fails”
A failed command can indicate a missing or inaccessible driver, a restricted container or virtual machine, or a utility that is not on the system path. Obtain the model through the operating system’s hardware-information view first. If CUDA software also cannot see a device, resolve driver or device-passthrough configuration before evaluating application compatibility.
“The card is listed, but the application will not start”
Return to the application’s requirements and compare its minimum CC, supported driver, operating system and Toolkit versions with the installed environment. Confirm that the program is using the NVIDIA device rather than another graphics adapter. If the application provides a diagnostic command or log, use the reported architecture and driver errors to identify which layer failed.
“The card is old and appears only in the legacy table”
Legacy status does not erase CUDA capability. It does mean you must pay closer attention to the application’s minimum CC and to software versions that still ship binaries for that architecture. If a current release has dropped the card, installing an older, explicitly supported application release may be the only software option; that choice can involve security, feature and maintenance trade-offs.
“A newer Toolkit is installed, but the build still fails”
Toolkit recency alone does not solve architecture or toolchain mismatches. Check that the compiler is supported by that Toolkit, that the project targets the GPU’s CC, and that the driver satisfies the runtime requirement. If the project specifies a particular Toolkit release, use that release or follow its documented compatibility path instead of mixing components from unrelated versions.
When a GPU upgrade is the right next step
Consider an upgrade only after the exact failure is known. A replacement is relevant when the current GPU is absent from both NVIDIA tables, below the application’s minimum CC, or no longer supported by the software version you must run. It is not automatically justified when the problem is a missing driver, unsupported operating system, incompatible compiler or incorrect architecture target.
| Finding | Likely action |
|---|---|
| Exact model appears in a table and meets the application’s minimum CC. | Install the supported driver, Toolkit and toolchain; no hardware change is indicated by the lookup. |
| Exact model appears, but its CC is below the application minimum. | Use an application release that supports the card, if one is maintained, or evaluate a GPU that meets the stated minimum. |
| Exact model appears, but the application has no binary for that architecture. | Check whether a source build can target the CC or whether the vendor supports another release. |
| Exact model is absent from current and legacy tables. | Do not assume CUDA support; obtain authoritative hardware specifications before spending on software or hardware. |
| Hardware passes, but driver, OS or Toolkit requirements fail. | Correct the software stack first and retest before buying a different GPU. |
A repeatable checklist
- Run
nvidia-smi, or open the operating system’s device-information view. - Copy the complete NVIDIA model name and note which device is being used.
- Search the current NVIDIA table, then the legacy table if necessary.
- Write down the listed compute capability in major.minor form.
- Translate architecture notation when needed:
sm_86means CC 8.6. - Read the target application’s requirements for minimum CC, driver, operating system, Toolkit and compiler.
- Install only the versions that the application and operating system support.
- If the program still fails, use its diagnostics to determine whether the remaining issue is device visibility, architecture targeting or software compatibility.
The decisive answer to “Does my NVIDIA GPU support CUDA?” comes from the exact model’s presence and compute-capability value in NVIDIA’s tables. The decisive answer to “Will my CUDA application run?” requires that hardware result plus the application’s current software requirements.
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.
