1 Reply Latest reply on Nov 15, 2017 8:12 AM by black_zion

    processor performance, how to do the assessment?

    gigiux

      Dear all,

      I upgraded my PC from an AMD A4 6300 dual core 3.7 GHz (16 Gb RAM) to an AMD FX8350 octa core 4.0 GHz (32 GB RAM). I was expecting a (8-2 cores) / 2 cores = 3-fold increase in performances and I tested it by extracting a tar archive that produces a 52 Gb file. With the A4 processor it has taken 561 seconds to extract the file, with the FX processor it has taken 454 s. Thus the fold increase of performance for this task has been: (561 s – 454 s) / 561 s = 0.19, which is a bit small. So maybe what I have to look at is (4.0-3.7 GHz) / 3.7 GHz = 0.08 fold increase.

      My questions are:

      1. how do I assess the true computational power of a machine? If on the paper I have a given number of processors and a certain speed level, is there a standard measure that allows me to compare two machines for real (what I obtained with the reported process suggest me that I could have kept my previous machine and save the money since the increased performances are minimal)

      2. what is the impact of the motherboard on the processor performances? I understand that also the motherboard is involved in the final performance of the machine, so that the final speed of the processor (let's say 4 GHz) is just nominal: the motherboard might hamper the speed and reduce the performances. So what shall I look in the motherboard specification in roder to achieve higher computational power?

      Thank you

        • Re: processor performance, how to do the assessment?
          black_zion

          Was there any particular reason you chose the dead and obsolete Socket AM3 platform instead of the far superior Socket AM4? Also, 114.5MB/s is butting up against the typical sustained write speed of a hard drive, especially if you are reading and extracting the archive on the same drive. Also, the FX-8350 is a 4 module 8 thread processor so it still only has 4 floating point processors.

           

          As for performance benchmarks, there's no shortage of them, but there's also no universal benchmark as each one is situational, even Linpack has its detractors, which is why reputable review sites focus more on a suite of real world applications, not premade benchmarks. So the answer to your first question is there is no way to obtain the "true" computational power of a machine because there is no "true" computational value, especially when it comes to multithreading because not every program will utilize it in the same way, some do not at all, some achieve nearly 100% scaling, and most fall between those two values.

           

          As for the motherboard, it has little to do with the computational speed outside the supported RAM speeds, overclocking ability, and drive interfaces, but if you take two motherboards, one an entry level model, say a AMD 960 chipset board, and one a high end model, say the ASUS Crosshair V, stick the same CPU, RAM, and drives in there, raw computational performance will be equal.