Slightly frustrated with some of his wording; RAM, HHDs, and SSD all DO provide "random access", just not "uniform random access." But other than that a great presentation on the internal workings of the components.
One thing is clear, there's no such thing as "non-uniform random access". That's just "not random access" plain and simple. Otherwise, you could call a tape cassette non-uniform random access because it takes a variable amount of time to access every byte, just like an HDD has to pass through all cylinders to reach the innermost part of the platters.
Myth 1: Single thread performance of CPU are *not* going up. Amdal's law explains while multiprocessing and architectural performance is basically asymptotic, and thus cannot increase forever. What you failed to show in showing the Sandybridge architecture is the comparison with Alpha 21264 architecture from the mind 1990s. It was basically the same. The presenter also doesn't understand ... branching costs almost *0* in modern code (just like it did on the Alpha 21264). You don't need many branch or divide units, because these are relatively rare in real world software. The reason that the text splitting got faster is because the I/O in CPUs have been getting better over time.
![I.N "HALLUCINATION" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/n5B5q1Hwt_U/mqdefault.jpg)
Google for "Mythbusting modern hardware to gain "Mechanical Sympathy" slides" and they show up. It's a pdf.
Great presentation that lays the foundation upon which some decent practices can be built. Hopefully they will be.
Yeah could we please have a version of this video where the slides are visible throughout the talk? This can get really hard to follow at times. :/
Slightly frustrated with some of his wording; RAM, HHDs, and SSD all DO provide "random access", just not "uniform random access." But other than that a great presentation on the internal workings of the components.
One thing is clear, there's no such thing as "non-uniform random access". That's just "not random access" plain and simple.
Otherwise, you could call a tape cassette non-uniform random access because it takes a variable amount of time to access every byte, just like an HDD has to pass through all cylinders to reach the innermost part of the platters.
Myth 1: Single thread performance of CPU are *not* going up. Amdal's law explains while multiprocessing and architectural performance is basically asymptotic, and thus cannot increase forever. What you failed to show in showing the Sandybridge architecture is the comparison with Alpha 21264 architecture from the mind 1990s. It was basically the same. The presenter also doesn't understand ... branching costs almost *0* in modern code (just like it did on the Alpha 21264). You don't need many branch or divide units, because these are relatively rare in real world software. The reason that the text splitting got faster is because the I/O in CPUs have been getting better over time.
Bad slide showing and synchronization
Love it, made me laugh too