Intel Corp (INTC)

[Elmer Phud]

Some thoughts of recent AMD Zen benchmarks

It occurred to me that AMD's recent demo of Zen running an opensource application against Broadwell may reveal more than AMD wanted. The demo showed Zen beating Broadwell by 1-2% while Broadwell was hobbled down to the same 3GHz and turbo mode turned off. Naturally the Faithful went wild with excitement but I thought about it, why would AMD chose a single opensource application to use to compare with Broadwell? Then it became clear.

#1 Without publishing the system configuration or compiler options, no one can verify their results. Had they used a standard benchmark,anyone could see if they get the same results. I suspect they wouldn't See #2

#2 It would not be at all unusual for Zen to still have bugs at this stage of debug that need compiler switches to turn off features or capabilities. Only with a custom benchmark can AMD accomplish this without telling the world what does or still does not yet work on Zen.