What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
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
keep_alive: -1 tells Ollama to keep a model loaded; it does not, by itself, establish that Ollama will deadlock when switching models. A hang during a model switch could reflect VRAM pressure and eviction, or a scheduler problem—but the available report of a scheduler hang used a 32 GB GPU, not a 6 GB card. The exact cause of the 6 GB symptom remains unconfirmed.
What `keep_alive: -1` does
Ollama’s FAQ says the default idle residency is five minutes. A negative `keep_alive` value, including `-1`, keeps a model in memory; `0` unloads it after the response. You can also provide a duration string or a number of seconds. The API’s `keep_alive` parameter overrides the server-wide `OLLAMA_KEEP_ALIVE` setting.
Persistent residency is a memory policy, not a documented guarantee that multiple models will fit simultaneously. Ollama may unload idle models to make room for another model. You can unload a model manually with ollama stop <model>.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Why a 6 GB GPU may struggle to switch models
Ollama documents that concurrent GPU model loads require the models to fit entirely in VRAM. If a model—or the combination of resident models and other GPU workloads—does not fit, the server may need to unload an idle model before loading another. That is expected memory pressure and eviction; it is not evidence on its own of a deadlock.
#1 Best Overall
- System Compatibility Note: 2‑slot ITX card, 169.9x123.5x39.2mm, single 8‑pin power, recommended 500W PSU. Verify chassis clearance before purchase.
- Dedicated Support: Please contact us directly through Amazon for any product questions or assistance you may require.
- Intel Arc A380 GPU: Powered by Intel Xe architecture with 6GB GDDR6 on 96‑bit bus – ideal for compact gaming, HTPC, and media builds.
- 2250MHz GPU Clock: Factory overclocked core delivers solid performance for esports titles and everyday creative tasks.
- Small Form Factor ITX Design: Compact 2‑slot card fits easily into mini‑ITX and small form factor cases without sacrificing performance.
Nominal GPU capacity alone does not determine whether a workload fits. Ollama says memory requirements for parallel requests scale with the number of parallel requests multiplied by context length. Model size, context, parallelism, other GPU processes, and whether a runner uses CPU or GPU all matter. The FAQ documents OLLAMA_MAX_LOADED_MODELS as a limit conditional on available memory, OLLAMA_NUM_PARALLEL as the maximum parallel requests per model (default 1), and OLLAMA_MAX_QUEUE as the queue cap (default 512). These controls describe loading and queueing behavior; raising a limit is not an established fix for a scheduler hang.
What the reported scheduler hang does—and does not—show
Open Ollama issue #17408, filed July 26, 2026, describes a silent hang during a particular concurrent eviction path. The reporter says new, cold /api/generate requests stopped completing when a concurrent request reached the runner selected for eviction. Requests to already-loaded models and endpoints including /api/ps, /api/tags, and /api/embed reportedly continued to work; restarting the server recovered service.
Rank #2
- NVIDIA Ampere Streaming Multiprocessors: The all-new Ampere SM brings 2X the FP32 throughput and improved power efficiency.
- 2nd Generation RT Cores: Experience 2X the throughput of 1st gen RT Cores, plus concurrent RT and shading for a whole new level of ray-tracing performance.
- 3rd Generation Tensor Cores: Get up to 2X the throughput with structural sparsity and advanced AI algorithms such as DLSS. These cores deliver a massive boost in game performance and all-new AI capabilities.
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure.
- OC Mode : 1500 MHz (Boost Clock)/Default Mode : 1470 MHz (Boost Clock)
The reported environment was Ollama 0.31.1 on Ubuntu 24.04.4, with an RTX 5090 32 GB, OLLAMA_NUM_PARALLEL=2, OLLAMA_KEEP_ALIVE=-1, a 32768-token context, Flash Attention enabled, and an f16 KV cache. The completion model was gemma4:26b Q4_K_M; an embedding runner was CPU-only (num_gpu: 0) and pinned with keep_alive: -1. The proposed explanation—that concurrent activity can overwrite an eviction mark and leave a scheduler operation waiting indefinitely—is the reporter’s analysis, not an upstream-confirmed root cause.
This is a substantial user report, not confirmation that every installation has the same bug. Its 32 GB GPU setup does not verify a 6 GB scenario, and the available evidence establishes neither how common the failure is nor whether a fix covers the exact symptom.
Rank #3
- NVIDIA Ampere Streaming Multiprocessors
- 2nd Generation RT Cores
- 3rd Generation Tensor Cores
- Powered by GeForce RTX 3050
- Integrated with 6GB GDDR6 96-bit memory interface
How to distinguish memory pressure from a cold-load hang
Start by checking whether Ollama is waiting to make room for a model or whether new model loads stop progressing even when the request pattern and available memory appear reasonable. Record the relevant details before changing settings:
- Ollama version, operating system, GPU model, backend, and driver.
- Which models are loaded and whether each uses GPU or CPU.
- Model size, context length, parallel-request setting, and other GPU workloads or consumers.
- Whether only requests that load a model from cold hang, or whether requests to resident models fail too.
- Whether requests overlap during model eviction, and whether the server logs show loading, queueing, or GPU initialization errors.
Ollama’s FAQ recommends ollama ps to inspect models in memory and their CPU/GPU placement. If the issue appears to involve GPU discovery or initialization, consult the official troubleshooting guide. It recommends debug and system diagnostics; for AMD it specifically names OLLAMA_DEBUG=1 for more GPU-discovery detail and checking system logs for driver errors. The guide also covers NVIDIA discovery and container access. These checks can help identify an initialization problem, but they do not independently prove a scheduler deadlock.
Rank #4
- Intel Arc A380 Chipset
- 6GB, 96-bit, GDDR6 memory, 15.5 Gbps graphics memory speed
- 3x DisplayPort 2.0 ready, up to 8K@60Hz, 1x HDMI 2.0
Practical next steps
- Inspect residency: run
ollama psand note which models are loaded and whether they use CPU or GPU. Compare that with available VRAM and other GPU workloads. - Test the residency setting: for a controlled test, use a finite duration or
keep_alive: 0instead of-1. If the symptom changes, that is useful evidence about memory pressure; it does not prove the underlying cause. - Reduce concurrent demand: test with fewer overlapping requests and a shorter context. Treat changes to
OLLAMA_NUM_PARALLELorOLLAMA_MAX_LOADED_MODELSas diagnostic variables, not guaranteed fixes. - Capture the failure conditions: record the request sequence, model and context settings, version, GPU/backend and driver,
ollama psoutput, and server logs. Note whether resident-model requests and/api/psstill respond when a cold load stalls. - Recover service if it is stuck: the issue reporter says a server restart restored service in their environment. That is a reported recovery, not a verified fix for other installations; capture useful logs first if possible.
What can be concluded about the 6 GB symptom
A 6 GB GPU can make model residency and switching more constrained, but the supplied facts do not establish that keep_alive: -1 causes a silent deadlock on that hardware, that every other model will hang, or that a particular workaround fixes it. Confirm the Ollama version, exact GPU and backend, models, context, parallelism, other VRAM use, request timing, and logs before attributing the failure to either memory pressure or a scheduler race.
Quick Recap
Best Value
- 【Ultimate Triple Display Connectivity】: Features a versatile output array including HDMI, DisplayPort (DP), and DVI. Whether you're connecting a high-refresh-rate gaming monitor via DP or a standard office screen via HDMI, this card supports triple-monitor setups for maximum productivity.
- 【Compact Size & Wide Compatibility】: Measuring 240x135x45mm (9.45x5.31x1.77 inches), this dual-fan RX 580 fits perfectly into standard ATX Mid-Towers, Micro-ATX (M-ATX), ideal for compact desktop PC upgrades and space-saving gaming builds.
- 【Optimized Gaming Performance】: With 2048 Stream Processors and a 1206 MHz core clock, this card delivers solid frame rates in popular titles like Fortnite, GTA V, Apex Legends, and Valorant. It’s the ideal budget-friendly GPU for entry-level to mid-range gaming rigs.
- 【Advanced Thermal Management】: Engineered with a dual-fan cooling system and high-efficiency heat pipes to ensure stable performance under heavy loads. The intelligent fan control keeps your system quiet during light office work and provides maximum airflow during intense gaming sessions.
- 【Ready for Content Creation】: Supports DirectX 12, Vulkan, and OpenGL 4.6, making it more than just a gaming card. It provides hardware acceleration for video editing in Premiere Pro, 3D rendering in Blender, and smooth streaming for aspiring creators.
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.
