#tradeoffs — Public Fediverse posts

Lenovo's ThinkCentre X Ultra, the first Zenith-class machine, launches November 2026 at $3,699. A key detail: a 30B mixture-of-experts model runs at 70-100 tokens per second on this hardware, but a dense 70B model drops to about 5 tokens per second. Architecture matters more than memory size. https://www.implicator.ai/microsoft-project-zenith-developer-pcs-64gb/ #AI #performance #tradeoffs

Lenovo's ThinkCentre X Ultra, the first Zenith-class machine, launches November 2026 at $3,699. A key detail: a 30B mixture-of-experts model runs at 70-100 tokens per second on this hardware, but a dense 70B model drops to about 5 tokens per second. Architecture matters more than memory size. https://www.implicator.ai/microsoft-project-zenith-developer-pcs-64gb/ #AI #performance #tradeoffs

Lenovo's ThinkCentre X Ultra, the first Zenith-class machine, launches November 2026 at $3,699. A key detail: a 30B mixture-of-experts model runs at 70-100 tokens per second on this hardware, but a dense 70B model drops to about 5 tokens per second. Architecture matters more than memory size. https://www.implicator.ai/microsoft-project-zenith-developer-pcs-64gb/ #AI #performance #tradeoffs

Lenovo's ThinkCentre X Ultra, the first Zenith-class machine, launches November 2026 at $3,699. A key detail: a 30B mixture-of-experts model runs at 70-100 tokens per second on this hardware, but a dense 70B model drops to about 5 tokens per second. Architecture matters more than memory size. https://www.implicator.ai/microsoft-project-zenith-developer-pcs-64gb/ #AI #performance #tradeoffs

Lenovo's ThinkCentre X Ultra, the first Zenith-class machine, launches November 2026 at $3,699. A key detail: a 30B mixture-of-experts model runs at 70-100 tokens per second on this hardware, but a dense 70B model drops to about 5 tokens per second. Architecture matters more than memory size. https://www.implicator.ai/microsoft-project-zenith-developer-pcs-64gb/ #AI #performance #tradeoffs