What key findings emerged from SemiAnalysis's 43-day performance analysis of DeepSeek V4 Pro inference across AMD MI355X, Nvidia GB300 NVL72DeepSeek V4 Pro's 43-day inference performance timeline across Nvidia, AMD, and Huawei accelerators
AI 提示詞
Create a landscape editorial hero image for this Studio Global article: What key findings emerged from SemiAnalysis's 43-day performance analysis of DeepSeek V4 Pro inference across AMD MI355X, Nvidia GB300 NVL72. Article summary: Nvidia's GB300 NVL72 dominates raw throughput. Huawei's Ascend 950DT matched Nvidia on software readiness — a historic first. AMD's MI355X went from a broken launch to a 110.5x throughput recovery in 26 days, but still t. Topic tags: general, general web, user generated, documentation. Reference image context from search candidates: Reference image 1: visual subject "Sign up for Instagram to stay in the loop. Photo by NVIDIA on February 20, 2026. The latest SemiAnalysis InferenceX data proves that the best performance drives the lowest inferenc" source context "Instagram" Reference image 2: visual subject "DeepSeek V4 Pro 1.6T· GPU comparison. # DeepSeek V4 Pro
openai.com
2026年4月24日,DeepSeek 發表 1.6 兆參數的 V4 Pro 模型時,真正考驗並非模型本身的基準分數──而是市面上有沒有任何硬體能在百萬 Token 的推論情境下,不讓營運商破產。SemiAnalysis 啟動 InferenceX 追蹤平台,連續 43 天記錄所有主流 AI 加速器的表現。最終結果揭露的不只是晶片效能,更是背後軟體生態系的真實面貌。
What key findings emerged from SemiAnalysis's 43-day performance analysis of DeepSeek V4 Pro inference across AMD MI355X, Nvidia GB300 NVL72DeepSeek V4 Pro's 43-day inference performance timeline across Nvidia, AMD, and Huawei accelerators
AI 提示詞
Create a landscape editorial hero image for this Studio Global article: What key findings emerged from SemiAnalysis's 43-day performance analysis of DeepSeek V4 Pro inference across AMD MI355X, Nvidia GB300 NVL72. Article summary: Nvidia's GB300 NVL72 dominates raw throughput. Huawei's Ascend 950DT matched Nvidia on software readiness — a historic first. AMD's MI355X went from a broken launch to a 110.5x throughput recovery in 26 days, but still t. Topic tags: general, general web, user generated, documentation. Reference image context from search candidates: Reference image 1: visual subject "Sign up for Instagram to stay in the loop. Photo by NVIDIA on February 20, 2026. The latest SemiAnalysis InferenceX data proves that the best performance drives the lowest inferenc" source context "Instagram" Reference image 2: visual subject "DeepSeek V4 Pro 1.6T· GPU comparison. # DeepSeek V4 Pro
openai.com
2026年4月24日,DeepSeek 發表 1.6 兆參數的 V4 Pro 模型時,真正考驗並非模型本身的基準分數──而是市面上有沒有任何硬體能在百萬 Token 的推論情境下,不讓營運商破產。SemiAnalysis 啟動 InferenceX 追蹤平台,連續 43 天記錄所有主流 AI 加速器的表現。最終結果揭露的不只是晶片效能,更是背後軟體生態系的真實面貌。