Announcement

Collapse
No announcement yet.

Amazon EC2 Cloud Benchmarks vs. AMD Ryzen, Various AMD/Intel Systems

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • Amazon EC2 Cloud Benchmarks vs. AMD Ryzen, Various AMD/Intel Systems

    Phoronix: Amazon EC2 Cloud Benchmarks vs. AMD Ryzen, Various AMD/Intel Systems

    For putting the AMD Ryzen 7 Linux performance in additional perspective and showing how various Amazon EC2 cloud instances compare to bare metal hardware, here are fresh benchmarks of many different Amazon EC2 instance types compared to various Linux systems in our lab. This comes down to a 29-way comparison of different cloud instances and bare metal systems.

    http://www.phoronix.com/vr.php?view=24559

  • #2
    IOW, Cloud program sucks balls compared to Ryzen for most worloads, when one plans to use mroe than 150+ish CPU hours.

    Comment


    • #3
      Very interesting to see CPUs like the 5960X and 6900K fall behind the R7 1700 in more than one occasion. The price to performance ratio continues to surprise in a positive way.

      Comment


      • #4
        Imho, the most important thing about cloud computing is the break even point. Sure, on demand resources are nice for infrequent scenarios, but if you're going to use a machine in the cloud continuously, after some time it inevitably becomes more cost-effective to buy the actual hardware.

        Comment


        • #5
          Clouds may be scalable and good with load balancing. And "on demand" (TM by IBM?). But most important is: You are no longer in control. Your or other people's data is running through numbercrunching god-knows-where and nobody can guarantee you that the data is not copied, compromised or just being seen by someone else.
          But still a lot of enterprises outsource their (or better: customer's) data to cloud services.

          But happy to see Ryzen 7 so often on top.
          Stop TCPA, stupid software patents and corrupt politicians!

          Comment


          • #6
            Originally posted by bug77 View Post
            Imho, the most important thing about cloud computing is the break even point. Sure, on demand resources are nice for infrequent scenarios, but if you're going to use a machine in the cloud continuously, after some time it inevitably becomes more cost-effective to buy the actual hardware.

            Intersting thing is, this runs contrary to 99,999% of cloud marketing babble that I was forced to hear last quite a few years.

            Cloud is coming...clouds everyhwere... everything in a cloud... etc etc crap.

            Maketing morons were constatnly pushing HW price where clouds would supposedly be able to crush everthing since they buy en mass.

            And now, we came to breakeven ponts, which aren't even that high, even before considering other factors and costs.


            Comment


            • #7
              Originally posted by Brane215 View Post


              Intersting thing is, this runs contrary to 99,999% of cloud marketing babble that I was forced to hear last quite a few years.

              Cloud is coming...clouds everyhwere... everything in a cloud... etc etc crap.

              Maketing morons were constatnly pushing HW price where clouds would supposedly be able to crush everthing since they buy en mass.

              And now, we came to breakeven ponts, which aren't even that high, even before considering other factors and costs.

              Yeah, we found this out the hard way (despite my repeated warnings) when Amazon presented us the bill at the end of the year.
              The cloud is ok, you don't need to buy a 64 core 64GB RAM server if you only need it for a couple of days each year. But the cloud is NOT the default answer to everything.

              Comment


              • #8
                Benchmarking cloud based VM's is troublesome because it is simply the performance measurement in one point of time on that particular day.

                When you bench a Ryzen or Core in your lab, you have a reasonable expectation that it will function the same each time.

                Not so much in a cloud setting. Clouds have to be benchmarked like internet providers, with tests over several time frames and different points in time to collect a performance aggregate.

                Unless you order (an expensive) Amazon EC Bare Metal service, you have no control on how over the subscription rates EC may be tasking your particular VM. If you push your VM hard, it may be days before EC will motion your VM to a less tasked host. When that happens, your performance profile will change. If you run a basic VM, they may not move you at all. Its best available service and if your neighbor VM is "noisy" your performance profile will change as well.

                This exercise was absolutely valuable as a comparison with your local bare metal options and the cost benefit varies depending on how much compute over time you really need.

                But one must put cloud based compute in the proper context relative to what your needs are. This is a great way to start the discussion.

                Comment


                • #9
                  The title is a typo, it should have been "Ryzen is quite competitive with Amazon cloud and even Intel's high end stuff".
                  Last edited by starshipeleven; 04-27-2017, 02:02 PM.

                  Comment


                  • #10
                    Originally posted by edwaleni View Post
                    Benchmarking cloud based VM's is troublesome because it is simply the performance measurement in one point of time on that particular day.

                    When you bench a Ryzen or Core in your lab, you have a reasonable expectation that it will function the same each time.

                    Not so much in a cloud setting. Clouds have to be benchmarked like internet providers, with tests over several time frames and different points in time to collect a performance aggregate.
                    I don't think it's a fair point to make.
                    You don't pay a monthly subscription with advertised peak values - you pay hourly rates for different performance classes.
                    If an hour costs you 40 cents, and the performance drops by 40% in the next hour, why would you pay 40 cents for this one?
                    The whole point is to pay hourly rates for hourly compute performance. Why would you accept a one hour job to take 2 hours, especially when physical hardware is payed off in a limited time period and will guarantee you time to complete the job?

                    Comment

                    Working...
                    X