Sweclockers confirms rumour of limited availability for Coffee Lake until 2018:
"Those who wait for Intel Coffee Lake may have to wait longer than October 5th. According to information obtained by SweClockers, there is a shortage of new processors until after the turn of the year."
PS. The article confirms that motherboard availability is good, though. However, processors may not become widely available until mid-December, with volume in 2018. The source speculates that the effect of moving the launch earlier in time (it was originally scheduled for 1Q 2018), the launch of Coffee Lake now may have an Osborne effect on the Kaby Lake range, and due to poor availability, drive customers to the AMD platform, contrary to the intent of accelerating the launch.
There were games being played between OpEx and R&D over the last several qtrs. Devinder got grilled on that issue in the CC's, but I didn't pay so much attention to it. Intel has been doing something along that line as well, what they did is shift the leading edge product from the low end mobile to servers, and as a result, they charge more of the process development against the datacenter rather than client compute segments. Again, I didn't follow it all, but they are both playing accounting games.
I have no idea. I wouldn't think embedded GPU announcements would move this much, I don't recall them doing so in the past. Its a very small market for AMD, the shift in market cap today would likely equal 5 years of embedded GPU sales, not to mention profits.
There were filings, they just didn't show what was claimed in that article. Those share counts were all stock options they exercised and sold immediately. Along with additional shares. Its the opposite of what that article claimed. There have been zero purchases on the open market of shares by execs at AMD. They have instead been unloading them.
Interesting optimisation results for Ryzen (lower is better, left bars "Modified Vrad" show the improvement achieved by fixing a "false sharing" issue in Valve's Source engine; this issue leading to unnecessary inter-core communication, I presume):
The poster explains the poor scaling. It is probably due to Amdahl's Law, a good portion of sequential code that is not speeded up by parallelisation.
"The speedup in the fixed section of the code is likely a fair bit higher than what's shown here."
But, yes, the case shows how easy it is to introduce a scaling problem if the programmer does not understand multi-threading and shared memory well. Luckily, the shared counter, in this case, was useless and could be removed.