Processors etc

Comments

• Duff

  October 2011

  When Intel released Core 2, everything changed. Since then, AMD hasn't been able to keep up. When it comes to high-end CPUs, AMD really doesn't offer any competition.
  
  http://www.cpubenchmark.net/high_end_cpus.html
  
  While that's only one benchmark, it does give you an idea of the difference. If you want a cheap (but still decent) system, go with AMD. If you want performance, go Intel.
• nightice

  October 2011

  The basic idea in modern systems is that parallelism wins over clock speed. The P4's NetBurst and Prescott microarchitectures were total losses in just about every category. A missed branch could be devastating as the processor basically had to stop, undo everything, back up to the branch location, and start over. The entire pipeline was very serial and took 31 stages (a stage is, roughly, fetch instruction -> decode instruction -> fetch operands -> schedule instruction -> execute instruction -> store result) for a complete operation. A basic tenet of a pipelined architecture is that if there are several discrete units that are needed to execute an instruction, then it is possible to start the next instruction once one is done with a unit. (illustration) Intel's strategy was to shorten each pipeline operation to a minimal set of transistors to allow for faster and faster clock cycles to push more instructions through faster.
  
  Unfortunately this came back to bite them with the well-known power draw and thermal output issues. So they started over with a simpler architecture (Pentium M, which was an updated Pentium III) and everything since then has been on a clear evolutionary path. Nehalem (Core i7, 2008) is Core 2 quite literally on steroids and with hyperthreading, Clarkdale (Core i5/i3, 2010) is Nehalem with an integrated north bridge and GPU, Sandy Bridge (Core i7/i5/i3, 2011) is Clarkdale with better caching and higher-level instructions, etc.
  
  AMD never got that crazy with their designs and historically had somewhat of an edge in heavily threaded applications with Opteron, but that seems to have gone away with Bulldozer. They can probably turn it around next generation. Bulldozer at its core is a good idea IMO - sharing certain parts of the processor (cache and memory the controller) and parallelizing integer and floating point operations - but it sounds like it was rather rushed and sloppily engineered. AMD claims better performance per watt, which may be true, but it's sounding like it's fairly important to optimize important things like OS schedulers for BD. Windows 7 prefers to schedule threads to physical cores before logical ones (e.g. hyperthreading) but I don't recall how a BD processor is broken down to the scheduler - a single module has two parallel integer and FP units, and a chip branded as "dual-core" has two modules.
  
  In the case of both Intel and AMD, low-power lines (Atom, Bobcat) are generally unrelated to the other product lines, and server products are generally similar to the desktop ones but with a lot more cache and supporting multi-processor setups.
• Q

  October 2011

  OK, I appreciate the new information and backstory(s) you're giving me.
  
  Next question: Should I wait for the next thing from Intel/AMD or not? IE. Is what's next looking like it'll be a complete blockbuster or just better version of the current stuff?
  
  -Q
• Lupinthe3rd

  November 2011

  A bit late but still, if you want to get something which is gonna stand most people recommend a Core i52500K Sandy Bridge. Because it's a rock solid performer, and overclocks very well to 4+ GHz. The problem is Intel have released two or three new pin allignments and are planning anotehr with their next main release.