Skip to main content

Local AI upgrade intelligence

What does an AI upgrade actually unlock?

A faster GPU is only part of the answer. VRAM can change which models, quantizations, and context sizes are practical. UOS keeps capability, measured speed, system fit, and purchase evidence separate so one benchmark never hides the real trade-off.

Ask the Hardware Upgrade Advisor
Selectable GPU options
9
Requirement profiles
0
Model artifacts
2

Choose a task · inspect exact source records

Find out what the evidence can answer

Task selection is your browsing preference. No reviewed task-specific requirement profile exists, so it does not filter artifacts or predict GPU fit.

Fit status: UNKNOWN

There are 0 verified requirement profiles. A model artifact or a GPU specification is not a runtime requirement or measurement.

Verified model artifact records (2)

Qwen3.5-27B

Exact model revision
fc05daec18b0a78c049392ed2e771dde82bdf654
Repository snapshot
Qwen/Qwen3.5-27B@fc05daec18b0a78c049392ed2e771dde82bdf654
Revision relationship
Model and artifact revisions match exactly
Task / use case
Not documented in reviewed model facts
Format
safetensors
Quantization
Not recorded
Documented parameters
27.000000 B parameters
Architecture metadata
Qwen3_5ForConditionalGeneration
Model-card context limit
262,144 tokens · documented limit, not a tested fit
Repository files
11 files · 55.6 GB (decimal) total stored bytes
Hardware fit / speed
Unknown · no verified requirement profile or exact runtime measurement

File bytes do not establish GPU memory need. Context length from a model card does not prove that this model runs at that context on a particular GPU.

Qwen3.6-27B

Exact model revision
6a9e13bd6fc8f0983b9b99948120bc37f49c13e9
Repository snapshot
Qwen/Qwen3.6-27B@6a9e13bd6fc8f0983b9b99948120bc37f49c13e9
Revision relationship
Model and artifact revisions match exactly
Task / use case
Not documented in reviewed model facts
Format
safetensors
Quantization
Not recorded
Documented parameters
27.000000 B parameters
Architecture metadata
Qwen3_5ForConditionalGeneration
Model-card context limit
262,144 tokens · documented limit, not a tested fit
Repository files
15 files · 55.6 GB (decimal) total stored bytes
Hardware fit / speed
Unknown · no verified requirement profile or exact runtime measurement

File bytes do not establish GPU memory need. Context length from a model card does not prove that this model runs at that context on a particular GPU.

Reviewed source facts and observation dates
  • architecture: Qwen3_5ForConditionalGenerationQwen official Hugging Face repository · OFFICIAL_REPOSITORY_METADATA · pinned repository config.json · observed 2026-09-24T01:23:59.531ZOpen source fact
  • checkpoint artifact bytes: 55563006400 bytesQwen official Hugging Face repository · MODEL_ARTIFACT_REPOSITORY · pinned repository API blob metadata · observed 2026-09-24T01:23:59.531ZOpen source fact
  • max context tokens: 262144.000000 tokensQwen official Hugging Face repository · MODEL_AUTHOR_FIRST_PARTY · official model card · observed 2026-09-24T01:23:59.531ZOpen source fact
  • parameter count b: 27.000000 B parametersQwen official Hugging Face repository · MODEL_AUTHOR_FIRST_PARTY · official model card · observed 2026-09-24T01:23:59.531ZOpen source fact

Exact catalog hardware

GPU candidates with reviewed specifications (9)

These are sourced GPU facts, not a model-fit ranking. A candidate with no linked exact product page remains unlinked rather than guessed.

Measured runtime behavior

Local AI runtime benchmark comparison

An independent test measured multiple model/runtime configurations on one NVIDIA GeForce RTX 5080. These are runtime measurements on a fixed GPU, not GPU-vs-GPU scores, local FPS, or a hardware upgrade recommendation.

Each capture and condition is shown separately. Host CPU and OS were not published and remain unknown. GPU memory is the source-reported runtime allocation in MiB, not the card’s capacity.

Capture 2026-08-20

Ollama 0.32.1 vs llama.cpp b10507 · RTX 5080 · same capture is never joined to another date

Methodology: decode median of five runs; long-prefill uses the reported median. Settings and sample details are shown per runtime row.

gpt-oss-20b · capture

The test identifies different model conversions in this group. These are separate runtime configurations, not an identical-artifact comparison.

gpt-oss:20b

llama.cpp b10507 (upstream GGUF)

Published capture

Engine / backend: llama.cpp b10507 (upstream GGUF)

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
221.96 tokens/sec
Long-prefill median
Not reported
GPU memory during run
13,037 MiB
Decode samples
5 runs
Show raw result rowgpt-oss:20b,llama.cpp b10507 (upstream GGUF),221.96,220.53,1078.7,,,13037,19.8,5,185.8|242.6|222.0|231.9|175.3
gpt-oss:20b

ollama 0.32.1

Published capture

Engine / backend: ollama 0.32.1

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
195.79 tokens/sec
Long-prefill median
Not reported
GPU memory during run
14,152 MiB
Decode samples
5 runs
Show raw result rowgpt-oss:20b,ollama 0.32.1,195.79,194.59,3744.3,,,14152,8.6,5,195.8|195.7|196.3|196.1|195.7

llama-3.2-3b · capture

The source does not provide a verified model artifact match for this capture; no same-artifact performance conclusion is made.

llama3.2:3b

llama.cpp b10507

Published capture

Engine / backend: llama.cpp b10507

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
338.22 tokens/sec
Long-prefill median
16,809 tokens/sec
GPU memory during run
4,339 MiB
Decode samples
5 runs
Show raw result rowllama3.2:3b,llama.cpp b10507,338.22,336.01,5321.0,16809,407,4339,2.5,5,339.2|338.2|335.0|338.2|339.7
llama3.2:3b

ollama 0.32.1

Published capture

Engine / backend: ollama 0.32.1

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
318.51 tokens/sec
Long-prefill median
17,444 tokens/sec
GPU memory during run
4,343 MiB
Decode samples
5 runs
Show raw result rowllama3.2:3b,ollama 0.32.1,318.51,313.25,7038.1,17444,433,4343,3.2,5,319.0|318.7|318.2|318.5|317.5

qwen2.5-14b · capture

The source does not provide a verified model artifact match for this capture; no same-artifact performance conclusion is made.

qwen2.5:14b

llama.cpp b10507

Published capture

Engine / backend: llama.cpp b10507

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
101.38 tokens/sec
Long-prefill median
4,697 tokens/sec
GPU memory during run
10,942 MiB
Decode samples
5 runs
Show raw result rowqwen2.5:14b,llama.cpp b10507,101.38,101.29,2037.8,4697,423,10942,8.1,5,101.3|101.2|101.4|101.4|101.4
qwen2.5:14b

ollama 0.32.1

Published capture

Engine / backend: ollama 0.32.1

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
99.43 tokens/sec
Long-prefill median
5,004 tokens/sec
GPU memory during run
10,941 MiB
Decode samples
5 runs
Show raw result rowqwen2.5:14b,ollama 0.32.1,99.43,98.59,3151.4,5004,451,10941,5.6,5,99.5|99.5|99.4|99.4|99.3

Capture 2026-08-26

Ollama 0.32.15 vs LM Studio 0.4.21 · RTX 5080 · same capture is never joined to another date

Methodology: decode median of five runs; long-prefill uses the reported median. Settings and sample details are shown per runtime row.

gpt-oss-20b · defaults

The test identifies different model conversions in this group. These are separate runtime configurations, not an identical-artifact comparison.

gpt-oss-20b

lms LM Studio 0.4.21

defaults

Engine / backend: [email protected]

Context 8,192 · temperature 0 · seed 42 · GPU offload not reported

Decode median
186.26 tokens/sec
Long-prefill median
5,241.3 tokens/sec
GPU memory during run
12,310 MiB
Decode samples
5 runs
Show raw result rowgpt-oss-20b,lms,LM Studio 0.4.21,[email protected],defaults,8192,186.26,184.34,5241.3,474,78.9,12310,1119,8.43,0.36,12917.9,5,183.42|186.26|186.84|186.98|185.4
gpt-oss:20b

ollama ollama 0.32.15

defaults

Engine / backend: vendored llama.cpp 9d77fa172

Context 4,096 · temperature 0 · seed 42 · GPU offload not reported

Decode median
213.54 tokens/sec
Long-prefill median
9,239.4 tokens/sec
GPU memory during run
13,570 MiB
Decode samples
5 runs
Show raw result rowgpt-oss:20b,ollama,ollama 0.32.15,vendored llama.cpp 9d77fa172,defaults,4096,213.54,212.03,9239.4,475,,13570,1053,18.56,23.67,2138.8,5,213.22|213.64|213.48|213.54|213.76

gpt-oss-20b · normalized

The test identifies different model conversions in this group. These are separate runtime configurations, not an identical-artifact comparison.

gpt-oss-20b

lms LM Studio 0.4.21

normalized

Engine / backend: [email protected]

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
242.83 tokens/sec
Long-prefill median
9,617.7 tokens/sec
GPU memory during run
12,533 MiB
Decode samples
5 runs
Show raw result rowgpt-oss-20b,lms,LM Studio 0.4.21,[email protected],normalized,4096,242.83,240.13,9617.7,474,38.6,12533,1101,8.43,0.27,13176.6,5,242.83|242.82|243.14|241.74|244.35
gpt-oss:20b

ollama ollama 0.32.15

normalized

Engine / backend: vendored llama.cpp 9d77fa172

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
213.65 tokens/sec
Long-prefill median
9,187.1 tokens/sec
GPU memory during run
13,570 MiB
Decode samples
5 runs
Show raw result rowgpt-oss:20b,ollama,ollama 0.32.15,vendored llama.cpp 9d77fa172,normalized,4096,213.65,211.8,9187.1,474,,13570,1053,8.05,8.36,2152.1,5,213.62|213.65|213.68|213.25|213.73

llama-3.2-3b · defaults

The model artifact is source-matched, but test settings are not fully verified as matched. Read each row's context and offload details before comparing these configurations.

llama-3.2-3b-q4km

lms LM Studio 0.4.21

defaults

Engine / backend: [email protected]

Context 8,192 · temperature 0 · seed 42 · GPU offload not reported

Decode median
314.86 tokens/sec
Long-prefill median
20,341.8 tokens/sec
GPU memory during run
4,260 MiB
Decode samples
5 runs
Show raw result rowllama-3.2-3b-q4km,lms,LM Studio 0.4.21,[email protected],defaults,8192,314.86,308.68,20341.8,440,9.4,4260,1122,2.75,0.07,3471.9,5,273.28|312.54|316.75|315.22|314.86
llama3.2:3b

ollama ollama 0.32.15

defaults

Engine / backend: vendored llama.cpp 9d77fa172

Context 4,096 · temperature 0 · seed 42 · GPU offload not reported

Decode median
351.22 tokens/sec
Long-prefill median
17,802.7 tokens/sec
GPU memory during run
3,760 MiB
Decode samples
5 runs
Show raw result rowllama3.2:3b,ollama,ollama 0.32.15,vendored llama.cpp 9d77fa172,defaults,4096,351.22,345.19,17802.7,434,,3760,1053,2.67,2.79,1234.0,5,351.15|350.9|351.86|351.32|351.22

llama-3.2-3b · normalized

llama-3.2-3b-q4km

lms LM Studio 0.4.21

normalized

Engine / backend: [email protected]

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
316 tokens/sec
Long-prefill median
20,336.7 tokens/sec
GPU memory during run
3,808 MiB
Decode samples
5 runs
Show raw result rowllama-3.2-3b-q4km,lms,LM Studio 0.4.21,[email protected],normalized,4096,316.0,309.91,20336.7,442,9.2,3808,1103,2.76,0.07,3469.6,5,310.55|316.0|315.86|316.23|316.6
llama3.2:3b

ollama ollama 0.32.15

normalized

Engine / backend: vendored llama.cpp 9d77fa172

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
350.52 tokens/sec
Long-prefill median
17,862.3 tokens/sec
GPU memory during run
3,760 MiB
Decode samples
5 runs
Show raw result rowllama3.2:3b,ollama,ollama 0.32.15,vendored llama.cpp 9d77fa172,normalized,4096,350.52,338.11,17862.3,432,,3760,1057,2.27,2.38,1239.6,5,349.35|350.38|350.52|350.63|350.96

qwen2.5-14b · defaults

The model artifact is source-matched, but test settings are not fully verified as matched. Read each row's context and offload details before comparing these configurations.

qwen2.5-14b-q4km

lms LM Studio 0.4.21

defaults

Engine / backend: [email protected]

Context 8,192 · temperature 0 · seed 42 · GPU offload not reported

Decode median
98.15 tokens/sec
Long-prefill median
5,195.8 tokens/sec
GPU memory during run
11,194 MiB
Decode samples
5 runs
Show raw result rowqwen2.5-14b-q4km,lms,LM Studio 0.4.21,[email protected],defaults,8192,98.15,97.31,5195.8,454,24.6,11194,1103,7.38,0.15,10406.8,5,97.36|98.14|98.17|98.15|98.17
qwen2.5:14b

ollama ollama 0.32.15

defaults

Engine / backend: vendored llama.cpp 9d77fa172

Context 4,096 · temperature 0 · seed 42 · GPU offload not reported

Decode median
102.25 tokens/sec
Long-prefill median
4,983.8 tokens/sec
GPU memory during run
10,356 MiB
Decode samples
5 runs
Show raw result rowqwen2.5:14b,ollama,ollama 0.32.15,vendored llama.cpp 9d77fa172,defaults,4096,102.25,101.06,4983.8,453,,10356,1053,5.51,5.67,1604.1,5,102.13|102.34|102.25|102.3|102.25

qwen2.5-14b · normalized

qwen2.5-14b-q4km

lms LM Studio 0.4.21

normalized

Engine / backend: [email protected]

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
97.63 tokens/sec
Long-prefill median
5,566.5 tokens/sec
GPU memory during run
10,405 MiB
Decode samples
5 runs
Show raw result rowqwen2.5-14b-q4km,lms,LM Studio 0.4.21,[email protected],normalized,4096,97.63,96.62,5566.5,453,24.7,10405,1106,7.68,0.16,10411.1,5,96.89|97.75|97.46|97.79|97.63
qwen2.5:14b

ollama ollama 0.32.15

normalized

Engine / backend: vendored llama.cpp 9d77fa172

Context 4,096 · temperature 0 · seed 42 · GPU offload full

Decode median
102.21 tokens/sec
Long-prefill median
5,038.3 tokens/sec
GPU memory during run
10,356 MiB
Decode samples
5 runs
Show raw result rowqwen2.5:14b,ollama,ollama 0.32.15,vendored llama.cpp 9d77fa172,normalized,4096,102.21,101.23,5038.3,452,,10356,1053,5.52,5.68,1617.6,5,102.21|102.21|102.17|102.21|102.21
Dataset v1.1.0. Result rows are preserved as a reviewed content snapshot; canonical database insertions: 0.

Coverage scope: these counts represent Local AI decision inputs available in this view, not the complete benchmark or retail catalog. A verified requirement profile is not a measured GPU execution result.

Start with your hardware

Your PC is the baseline.

Save the GPU, CPU, memory, and motherboard you already own. Then choose Local AI as the goal and evaluate a candidate against the system you actually have.

Model fitDoes the selected model/profile fit the reviewed VRAM requirement?Capability
Measured speedUse exact GPU, model, backend, and context measurements where they exist.Performance
System fitPower, connectors, clearance, and supporting parts stay separate from AI speed.Compatibility
Purchase callPerformance and capability are combined with price only when compliant market evidence exists.Decision

How to read Local AI evidence

Capability gain and performance gain are different.

A 16 GB → 32 GB GPU can expand practical model and context choices even when the measured speed gain is smaller than the capability gain. UOS should report both rather than collapse them into one score.

What remains unknown

If an exact GPU/model/backend/context combination has not been measured, UOS does not substitute a sibling GPU or extrapolate a tokens-per-second number. Missing evidence remains research backlog.