AI 推理部署案例 (硬件 × 模型 × 引擎) 实测数据流https://evokernel.dev/zh-CNhttps://evokernel.dev/cases/case-dsr1-asc910bx16-mindie-001/https://evokernel.dev/cases/case-dsr1-asc910bx16-mindie-001/ascend-910b ×16 · deepseek-r1 · mindie · bf16 · decode 850 tok/s · TTFT p50 280ms · 瓶颈 memory-bandwidthTue, 28 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-dsv4-flash-h100x8-vllm-fp8-001/https://evokernel.dev/cases/case-dsv4-flash-h100x8-vllm-fp8-001/h100-sxm5 ×8 · deepseek-v4-flash · vllm · fp8-e4m3 · decode 4200 tok/s · TTFT p50 220ms · 瓶颈 memory-bandwidthTue, 28 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-dsv4pro-cm384-mindie-001/https://evokernel.dev/cases/case-dsv4pro-cm384-mindie-001/ascend-910c ×384 · deepseek-v4-pro · mindie · bf16 · decode 2400 tok/s · TTFT p50 380ms · 瓶颈 memory-bandwidthTue, 28 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-llama33-a100x8-vllm-001/https://evokernel.dev/cases/case-llama33-a100x8-vllm-001/a100-sxm4 ×8 · llama-3.3-70b · vllm · bf16 · decode 1480 tok/s · TTFT p50 220ms · 瓶颈 memory-bandwidthTue, 28 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-llama4-scout-h100x8-vllm-001/https://evokernel.dev/cases/case-llama4-scout-h100x8-vllm-001/h100-sxm5 ×8 · llama-4-scout · vllm · bf16 · decode 1850 tok/s · TTFT p50 145ms · 瓶颈 memory-bandwidthTue, 28 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-qwencoder-l40sx4-vllm-001/https://evokernel.dev/cases/case-qwencoder-l40sx4-vllm-001/l40s ×4 · qwen2.5-coder-32b · vllm · fp8-e4m3 · decode 580 tok/s · TTFT p50 480ms · 瓶颈 memory-bandwidthTue, 28 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-dsv4flash-disagg-h100-h200-001/https://evokernel.dev/cases/case-dsv4flash-disagg-h100-h200-001/h200-sxm ×16 · deepseek-v4-flash · sglang · fp8-e4m3 · decode 9600 tok/s · TTFT p50 320ms · 瓶颈 memory-bandwidthMon, 27 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-glm51-h200x8-vllm-001/https://evokernel.dev/cases/case-glm51-h200x8-vllm-001/h200-sxm ×8 · glm-5.1 · vllm · bf16 · decode 2400 tok/s · TTFT p50 280ms · 瓶颈 memory-bandwidthSun, 26 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-qwen36-mi325x8-sglang-001/https://evokernel.dev/cases/case-qwen36-mi325x8-sglang-001/mi325x ×8 · qwen3.6-plus · sglang · fp8-e4m3 · decode 3100 tok/s · TTFT p50 240ms · 瓶颈 memory-bandwidthSun, 26 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-llama4mvk-trillium-256-001/https://evokernel.dev/cases/case-llama4mvk-trillium-256-001/trillium ×256 · llama-4-maverick · vllm · bf16 · decode 5800 tok/s · TTFT p50 180ms · 瓶颈 computeSat, 25 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-llama4scout-dcuk100x8-001/https://evokernel.dev/cases/case-llama4scout-dcuk100x8-001/dcu-k100 ×8 · llama-4-scout · vllm · bf16 · decode 850 tok/s · TTFT p50 320ms · 瓶颈 softwareSat, 25 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-qwen35-397b-mi355x8-001/https://evokernel.dev/cases/case-qwen35-397b-mi355x8-001/mi355x ×8 · qwen3.5-397b · vllm · fp4 · decode 4500 tok/s · TTFT p50 220ms · 瓶颈 memory-bandwidthFri, 24 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-dsv4flash-mtts4000x16-001/https://evokernel.dev/cases/case-dsv4flash-mtts4000x16-001/mtt-s4000 ×16 · deepseek-v4-flash · vllm · fp16 · decode 320 tok/s · TTFT p50 540ms · 瓶颈 softwareThu, 23 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-kimik26-mlu590x16-001/https://evokernel.dev/cases/case-kimik26-mlu590x16-001/mlu590 ×16 · kimi-k2.6 · vllm · bf16 · decode 480 tok/s · TTFT p50 460ms · 瓶颈 softwareWed, 22 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-llama4scout-mi300x8-001/https://evokernel.dev/cases/case-llama4scout-mi300x8-001/mi300x ×8 · llama-4-scout · vllm · bf16 · decode 2200 tok/s · TTFT p50 158ms · 瓶颈 memory-bandwidthWed, 22 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-qwen36plus-mlu590x8-001/https://evokernel.dev/cases/case-qwen36plus-mlu590x8-001/mlu590 ×8 · qwen3.6-plus · lmdeploy · int8 · decode 380 tok/s · TTFT p50 580ms · 瓶颈 softwareWed, 22 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-gemma4-h100x4-fp8-001/https://evokernel.dev/cases/case-gemma4-h100x4-fp8-001/h100-sxm5 ×4 · gemma-4 · tensorrt-llm · fp8-e4m3 · decode 6800 tok/s · TTFT p50 95ms · 瓶颈 computeTue, 21 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-glm51-br104x8-001/https://evokernel.dev/cases/case-glm51-br104x8-001/br104 ×8 · glm-5.1 · vllm · int8 · decode 240 tok/s · TTFT p50 720ms · 瓶颈 softwareMon, 20 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-gptoss-gaudi3x8-001/https://evokernel.dev/cases/case-gptoss-gaudi3x8-001/gaudi-3 ×8 · gpt-oss · vllm · fp8-e4m3 · decode 2900 tok/s · TTFT p50 140ms · 瓶颈 memory-bandwidthMon, 20 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-dsv3-trainium2-x64-001/https://evokernel.dev/cases/case-dsv3-trainium2-x64-001/trainium-2 ×64 · deepseek-r1 · vllm · bf16 · decode 3600 tok/s · TTFT p50 320ms · 瓶颈 memory-bandwidthSun, 19 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-gemma4-c500x4-001/https://evokernel.dev/cases/case-gemma4-c500x4-001/metax-c500 ×4 · gemma-4 · vllm · int8 · decode 580 tok/s · TTFT p50 420ms · 瓶颈 memory-bandwidthSat, 18 Apr 2026 00:00:00 GMThttps://evokernel.dev/cases/case-dsr1-tianhe100x16-001/https://evokernel.dev/cases/case-dsr1-tianhe100x16-001/iluvatar-bi ×16 · deepseek-r1 · lmdeploy · int8 · decode 220 tok/s · TTFT p50 980ms · 瓶颈 softwareWed, 15 Apr 2026 00:00:00 GMT