Ziskind
Ziskind is a technology blogger or developer focused on Apple hardware and local AI performance testing.
Timeline (last 90 days)
In Ziskind's hands-on tests, the M5 Ultra achieved three- to four-fold prefill gains on long prompts, with only 1.6x on short prompts; MacStories measured about 2.5x on a 16,000-token prompt, with time to first token dropping from 13.9 seconds to 5.6 seconds.
Ziskind tested DeepSeek V4 Flash with a 32,000-token codebase: the 256GB M5 Ultra took about 50 seconds to first token, while two DGX Spark units totaling about $10,000 took only 17 seconds.
Ziskind tested DeepSeek V4 Flash concurrency: the Mac's total throughput peaked at four users and declined at eight users; with eight users and chat history, each person waited over a minute to first token, after which generation was about 5 tokens per second.
All Facts (3)
In Ziskind's hands-on tests, the M5 Ultra achieved three- to four-fold prefill gains on long prompts, with only 1.6x on short prompts; MacStories measured about 2.5x on a 16,000-token prompt, with time to first token dropping from 13.9 seconds to 5.6 seconds.
50%UnverifiedZiskind tested DeepSeek V4 Flash with a 32,000-token codebase: the 256GB M5 Ultra took about 50 seconds to first token, while two DGX Spark units totaling about $10,000 took only 17 seconds.
50%UnverifiedZiskind tested DeepSeek V4 Flash concurrency: the Mac's total throughput peaked at four users and declined at eight users; with eight users and chat history, each person waited over a minute to first token, after which generation was about 5 tokens per second.
50%