0

This is not a production question, just something that I've been thinking about.

When speccing out CPUs I know the general rule is a faster clock speed will make a core run faster and the greater number of cores you have the more you can do in parallel. However what makes a "newer" CPU generally faster than an older CPU other than these two things?

For example:

  • There is a noticeable difference between an older i5 and a newer i5 despite having the same number of cores and around the same GHz.

  • My current dual core i5 running at 2.5 (I believe) GHz would surely blow a Pentium 4 out of the water, despite the Pentium running at > 3GHz. I'm sure I could blow a dual core server from 2004 out of the water as well just to make things fair core-wise. I have never done any formal benchmarking, these are just educated guesses.

What are some things to look for when speccing out CPUs?

Kevin Panko
  • 7,346
  • 22
  • 44
  • 53
omgtestadsfasdf
  • 1
  • it seems like you're asking two different questions here. are you asking what makes newer CPUs faster than older ones or what to look for in new CPUs? – nc4pk May 16 '13 at 21:59
  • Don't forget larger and faster cache memory and fast main memory. There is also the motherboard chipset to consider. – hookenz May 16 '13 at 22:51
  • "specing out CPUs" is what benchmarks are for. Unfortunately, it is difficult to find good benchmark results--[SPEC CPU](http://www.spec.org/cpu2006/) is probably the best available for "workstation" workloads--because many benchmarks are broken or have weak reporting rules. Applying benchmark results to estimate performance for one's own workload is a difficult art. –  May 17 '13 at 12:11

1 Answers1

2

Architecture.

A CPU at 3.0Ghz means that the CPU is doing 3 billion somethings per second. As computer architectures grow more efficient, they are able to do more with each clock cycle (the something).

I haven't kept entirely up to date, but back when the Pentiums 4's were hot stuff, the reason comparable AMD processors had a lower clock rate was because they did more per clock cycle than the P4. The trade off is a clock cycle on an AMD processor took longer.

Some processors can execute multiple instructions in a single clock cycle. Some processors take multiple clock cycles to execute a single instruction. Many processors uses a technique called pipelining, which allows them to have instructions that take multiple clock cycles to execute, but still manage to execute roughly 1 instruction per clock cycle.

Darth Android
  • 37,872
  • 5
  • 94
  • 112
  • But how can you actually spec that out, numbers wise? I know an i5 is better than a P4, but I don't see any numbers why. – omgtestadsfasdf May 16 '13 at 22:00
  • Part of the problem is you *can't*, at least not in a general sense. At best you can pick a metric - you'll often see FLOPS, or *floating-point operations per second* as a general idea of raw computing power, but that translates badly to real-world scenarios outside of academic math / numbers crunching, and so isn't really useful. Bitcoining mining is concerned about hashing speed, so you'll find those measured in hashes/second. FLOPS are meaningless for that. You have to design the metric around what you're doing, not the other way around. – Darth Android May 16 '13 at 22:04