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
If Ollama starts using your CPU after an update, first confirm where the model is running, then separate GPU discovery, CUDA initialization and model-placement problems. An ollama ps result showing CPU use does not by itself prove that an sm_86 CUDA kernel is missing. Ollama lists NVIDIA RTX 30-series GPUs at compute capability 8.6, but hardware compatibility does not guarantee that every packaged backend includes kernels for that architecture.
Check whether Ollama actually placed the model on the CPU
With the model loaded, run ollama ps in a terminal. Read the PROCESSOR column: it reports allocation, not the reason for it.
100% GPUmeans the model is loaded entirely in GPU memory.100% CPUmeans it is loaded entirely in system memory.- A split such as
48%/52% CPU/GPUmeans some of the model is placed in each.
Ollama explains these values in its FAQ. A CPU-only result confirms the symptom, but cannot tell you whether the GPU was not detected, CUDA failed to initialize, or placement was affected by another issue.
Crashes, 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 minuteWindows 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 reinstallWhat sm_86 support does—and does not—mean
Ollama’s GPU compatibility documentation lists GeForce RTX 30-series cards at compute capability 8.6, commonly written sm_86. The same documentation says NVIDIA GPUs with compute capability 5.0 or higher are supported and specifies NVIDIA driver 550 or newer; for compute capabilities 5.0 through 6.2, it specifies driver 570 or newer. Check the current documentation for your card and platform, since requirements can change.
#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
These are separate facts: a GPU can be within Ollama’s documented hardware compatibility range while a particular packaged CUDA backend or preset lacks kernels for its architecture. Conversely, a model running on the CPU does not prove a missing kernel. GPU discovery and CUDA initialization can fail for other reasons.
Identify which stage is failing
| What you observe | What it establishes | What it does not establish |
|---|---|---|
| The GPU is not detected | The runtime’s discovery path has not made the GPU available to Ollama. | It does not establish that the GPU is unsupported or that its kernels are missing. |
| GPU discovery occurs, but CUDA/backend initialization fails | The GPU was seen, but the selected backend did not initialize successfully. | It does not identify a missing sm_86 kernel without more diagnostic evidence. |
ollama ps shows a CPU/GPU split |
Some model placement is on the GPU and some is in system memory. | It does not, by itself, explain why placement is partial. |
ollama ps shows 100% CPU |
The model is placed entirely in system memory. | It does not distinguish discovery, initialization, packaging or other causes. |
Collect platform-specific diagnostics before changing backends
Record the Ollama version, operating system, GPU model, NVIDIA driver version, startup logs and ollama ps output. Those details help distinguish a change in a package or backend from a system-specific discovery or initialization failure.
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
Get more diagnostic output
Ollama’s troubleshooting guide advises setting CUDA_ERROR_LEVEL=50 and trying again to obtain additional diagnostic logs. Use the method appropriate to your operating system and how Ollama is launched; a variable set in one shell may not reach a service started separately.
Check Linux kernel messages
On Linux, the guide suggests checking for NVIDIA errors with:
Rank #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
sudo dmesg | grep -i nvrm
sudo dmesg | grep -i nvidiaCheck containers and installation-specific access
If Ollama runs in a container, verify that the container has access to the GPU; host-level GPU visibility alone does not establish container access. The relevant setup depends on the container and platform. The troubleshooting guide also documents OLLAMA_LLM_LIBRARY for selecting an available backend. Do not set it to a backend name based solely on an unrelated report: confirm the available backend and follow steps that match your installation.
What the sm_86 report on Windows actually shows
A user report opened July 16, 2026, describes an RTX 3050 Laptop GPU on Windows with Ollama 0.32.0. The reporter says default GPU discovery did not find the card and identifies it as compute capability 8.6. The issue references a related change adding sm_86 to the cuda_v13_windows preset, but the reporter says their remaining discovery issue was separate from that preset change. They report that explicitly setting OLLAMA_LLM_LIBRARY=cuda_v13 made the GPU appear in their environment. See the Ollama issue.
Rank #4
- 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
This is one user’s platform-specific report, not a general fix for Windows, Linux, or every Ollama release. It illustrates why kernel coverage and GPU discovery should not be treated as the same failure.
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 glitchesIs Ollama 0.32.14’s CUDA 12 fallback missing sm_86 kernels?
The specific claim that Ollama 0.32.14 omitted sm_86 kernels and thereby broke a CUDA 12 fallback is not independently established by the official documentation or primary issue evidence cited here. Do not infer that causal chain from an ollama ps CPU result or from a slowdown that began after an update. Establishing it requires version-specific evidence, such as a directly relevant release note, merged change or reproducible primary report.
Best Value
- 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
When custom builds are relevant
If you built Ollama from source, its development documentation explains that GPU backends are selected explicitly and that architecture overrides are available. That build guidance is for custom or locally built binaries; it is not a universal command-line fix for an official installation.
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.
