Phoronix Test Suite v10.8.4
Installed: pts/ollama-1.0.0
ollama 0.3.15:
pts/ollama-1.0.0
Running only DeepSeek-r1 8B test.
If ollama test can start own ollama server instead of being blocked from starting by other ollama service already using 11434 port, test leaves detached processes gobbling GPU memory. Up to the point where your fourth test run can't fit into GPU memory and starts running on CPU.
PID USER DEV TYPE GPU GPU MEM CPU HOST MEM Command
4143 llm_benchmark 0 Compute N/A 5508MiB 34% 0% 1479MiB ..../bin/ollama runner --ollama-engine --model /h
4384 llm_benchmark 0 Compute N/A 5508MiB 34% 0% 994MiB .../bin/ollama runner --ollama-engine --model /h
4607 llm_benchmark 0 Compute N/A 4490MiB 28% 1% 1990MiB .../bin/ollama runner --ollama-engine --model /h
464506 llm_benchmark 0 Compute N/A 152MiB 1% 596% 6079MiB ...snip....
Phoronix Test Suite v10.8.4
Installed: pts/ollama-1.0.0
ollama 0.3.15:
pts/ollama-1.0.0
Running only DeepSeek-r1 8B test.
If ollama test can start own ollama server instead of being blocked from starting by other ollama service already using 11434 port, test leaves detached processes gobbling GPU memory. Up to the point where your fourth test run can't fit into GPU memory and starts running on CPU.
4143 llm_benchmark 0 Compute N/A 5508MiB 34% 0% 1479MiB ..../bin/ollama runner --ollama-engine --model /h
4384 llm_benchmark 0 Compute N/A 5508MiB 34% 0% 994MiB .../bin/ollama runner --ollama-engine --model /h
4607 llm_benchmark 0 Compute N/A 4490MiB 28% 1% 1990MiB .../bin/ollama runner --ollama-engine --model /h
464506 llm_benchmark 0 Compute N/A 152MiB 1% 596% 6079MiB ...snip....