Announcement

Collapse
No announcement yet.

AMD FX-8350 "Vishera" Linux Benchmarks

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

  • #76
    A more modern way to do the testing

    Since with specific workloads there's a huge difference between the two in some tests as shown in an older article, I'd like to see all these tests both on intel and amd cpus also done with the code compiled using -Ofast instead of -O3 (also wouldn't mind -O2 since for workloads that don't involve raytracing, computational fluid dynamics or other applications using very large data sets which use a memory addressing pattern which results in a lot of cache misses it seems to be faster over -O3, even postgresql test in mentioned article shows -O2 as better than -O3 and it would be something significant for servers, also some graphicsmagick operations which use adjacent memory locations thus resulting in fewer cache misses).

    Comment


    • #77
      I was wondering why Anand's results were not brought up yet. Michael's tests are heavily mutithreaded, but Anand's seem to be more balanced/representative, imho. Plus, the power consumption graphs are pretty clear-cut.

      Comment


      • #78
        Cores

        I develop CAE post processing software for a living. There are days when I spent a LOT of time waiting for code to recompile after catching up my local source tree with our source code repository. At work I have an i7-x990 CPU, which is insanely expensive. However, it's SO worth it to run "make -j10" to do parallel builds, and still be able to get other background tasks done while it grinds away.

        We're also working very hard to leverage those multiple cores in our products. It's not easy!

        Anyway, I'd love to be able to do something similar when I work from home. There's NO WAY I can afford a high end Intel CPU on my personal budget. I'd be VERY interested to see how the 8250 or 8350 performs running parallel builds. I have a Phenom 2 1090 now, and am wondering if I'd get a significant performance boost from 8 cores. I don't overclock much, because I don't want to risk an unstable overclock causing erratic behavior in our code.

        Comment


        • #79
          Originally posted by jjmcwill2003 View Post
          I develop CAE post processing software for a living. There are days when I spent a LOT of time waiting for code to recompile after catching up my local source tree with our source code repository. At work I have an i7-x990 CPU, which is insanely expensive. However, it's SO worth it to run "make -j10" to do parallel builds, and still be able to get other background tasks done while it grinds away.

          We're also working very hard to leverage those multiple cores in our products. It's not easy!

          Anyway, I'd love to be able to do something similar when I work from home. There's NO WAY I can afford a high end Intel CPU on my personal budget. I'd be VERY interested to see how the 8250 or 8350 performs running parallel builds. I have a Phenom 2 1090 now, and am wondering if I'd get a significant performance boost from 8 cores. I don't overclock much, because I don't want to risk an unstable overclock causing erratic behavior in our code.
          You're exaggerating. You can have the latest from intel (i7-3770) for less than $300. Of course, if you're happy with an 8350, that will be cheaper.

          Comment


          • #80
            So how would a i7-3770 compare to the AMD 8350 on something like "make -j8" ?

            I was referring to what I paid for the i7-990X when I had my work PC built. Is the 3770 considered a high end Intel CPU? I thought only the 3930K and 3960X fit that category?

            Comment


            • #81
              Originally posted by jjmcwill2003 View Post
              So how would a i7-3770 compare to the AMD 8350 on something like "make -j8" ?

              I was referring to what I paid for the i7-990X when I had my work PC built. Is the 3770 considered a high end Intel CPU? I thought only the 3930K and 3960X fit that category?
              3960X is the highest end, but it's also in a different league than 8350. The 3770 gives you 8 threads, same as 8350. And while I don't know the results of make -j8, here's a comparison using Visual Studio: http://www.anandtech.com/show/6396/t...x4300-tested/3

              Comment


              • #82
                Thanks for the Visual Studio compile time benchmark link!

                Comment


                • #83
                  Originally posted by sturmflut View Post
                  These are measurements from Anandtech:
                  Dafuq Anandtech still manipulates graphs by picking base POWER value of 50 instead of 0 ??!

                  Anyone doing his is *ALREADY* biased.

                  Originally posted by sturmflut View Post
                  Intel gets the same job done in about the same time, but the whole system consumes about half the power. Idle power consumption is also lower.
                  From the graphs, Intel does the job 1/3 longer. Vishera comes first.
                  Also, from many other tests, Vishera idle is on paar to SB - 60W vs 70W.
                  And it costs less.
                  And has much more features.
                  And it overclocks.
                  And fits old socket.
                  And its better for multithreading.

                  Its very attractive CPU. Eats more, yet costs less and offers more.

                  Originally posted by bug77 View Post
                  3960X is the highest end, but it's also in a different league than 8350. The 3770 gives you 8 threads, same as 8350. And while I don't know the results of make -j8, here's a comparison using Visual Studio: http://www.anandtech.com/show/6396/t...x4300-tested/3
                  Matter of buying CPU, installing PTS and performing timed kernel compile.
                  Last edited by crazycheese; 10-24-2012, 12:22 PM.

                  Comment


                  • #84
                    I have corrected Anandtech fake graph.



                    1) Added missing zero base. Nonzero base is very known statistic fallacy.
                    2) Added arrow to point out where AMD CPUs are already done and idling, while Intel is still working.

                    Granted the AMD is still NOT as efficient as Intel when it comes to LOAD, it is still efficient at idling and the unefficiency is *in acceptable range* now, unlike Bulldozer.

                    Comment


                    • #85
                      Originally posted by crazycheese View Post
                      I have corrected Anandtech fake graph.



                      1) Added missing zero base. Nonzero base is very known statistic fallacy.
                      2) Added arrow to point out where AMD CPUs are already done and idling, while Intel is still working.

                      Granted the AMD is still NOT as efficient as Intel when it comes to LOAD, it is still efficient at idling and the unefficiency is *in acceptable range* now, unlike Bulldozer.
                      Great. Now that you fixed the graph, let me help you read it:
                      Vishera: 200W x 1,000s = 180,000Ws = 50Wh
                      Ivy Bridge: 105W x 1,100s = 115,500Ws = 32.08Wh
                      Enjoy.

                      Comment


                      • #86
                        Originally posted by jjmcwill2003 View Post
                        Thanks for the Visual Studio compile time benchmark link!
                        PTS should automatically use a good -j option, I think it uses the common 2x + 1 (so 17 for this 8-core). Note I'm not sure what the exact formula is, but it's certainly better than -j1.

                        In the linux kernel compile on page 8, 8350 beat the 3770.

                        Comment


                        • #87
                          Originally posted by pingufunkybeat View Post
                          The average user will be playing Mp3s from a collection (probably being organised by a database backend) while browsing the web, which means 10 to 20 tabs open, which means 10-20 threads, many of them running flash or something and buffering from youtube while the user is reading something else. And they probably have an anti-virus running on the background and a chat program and a p2p client downloading 3-4 movies minimised somewhere in the taskbar. They might even sync their smartphone or MP3 player with their collection at the same time, which takes a while, so they minimise that. You don't need fluid dynamics to need good multi-threaded performance.

                          The user who runs one process which then needs 300 GFLOPS of sustained single-thread performance is rather the exception. Browser, Word and Excel and most games run fine on 5-year old stuff.
                          What you wrote goes without saying. My first comment was in address to the dullard who thinks professional work only happens at the office.

                          Your scenario which is common amongst hundreds of millions of consumers will alone turn a system to a pile of goo real fast, especially those poorly developed Zynga games that are averaging over 2 GB each to run CityVille, CastleVille, Farmville, etc. Browsers eat system resources and the more threads to process the less the system has to wait for its turn.

                          Most commonly used applications you note are all multi-threaded. To see graphs of single-thread performance is insulting. Intel recognizes the direction industry needs to go: Heterogeneous Systems Architectures.

                          The Industry is jumping on board. When Intel does it will be just another vindication of AMD being ahead of the curve.

                          Comment


                          • #88
                            Originally posted by bug77 View Post
                            Great. Now that you fixed the graph, let me help you read it:
                            Vishera: 200W x 1,000s = 180,000Ws = 50Wh
                            Ivy Bridge: 105W x 1,100s = 115,500Ws = 32.08Wh
                            Enjoy.
                            8350 Vishera ended at 900 seconds and used average of 200W, where a rival 3570K ended at 1100 and consumed 120W.

                            200W x 900s / (60*60) = 50 Wh
                            120W x 1100s / (60*60) = 36,7 Wh

                            Thats 73,4% of Intel electro efficiency, intel is just 26% more efficient. Not 100% or 50%, but mere 26%.

                            Minimum price in europe:
                            3570k = 198,48€
                            8350 = 179,16€

                            8350 supports ecc ram, vtd and is unlocked. Also scales very well at load when downclocked.

                            I think its VERY competitive CPU compared to Intel. My plan for AMD - cut management income and get more *good* engineers. They need only to improve power management at LOAD and probably add 16 minicores version for enthusiasts on this technology (with several cores to run at full speed and offload unimportant lowpriority tasks to lowerspeed cores) - they need engineers at linux and offtopicos kernel to implement this and they are back competitive.

                            Comment


                            • #89
                              Originally posted by crazycheese View Post
                              Also scales very well at load when downclocked.
                              downvolted. Fixed.

                              Comment


                              • #90
                                Originally posted by crazycheese View Post
                                8350 supports ecc ram, vtd
                                amd-vi

                                Originally posted by crazycheese View Post
                                I think its VERY competitive CPU compared to Intel.
                                I still don't get why people get so worked up about the amd cpus being "nowhere near" intel cpus and that the performance is supposedly so immensely lacking and that it's "trash", etc.

                                Imagine for a moment intel didn't exist. The AMD cpu would be considered quite good. It still is, even if there are intel cpus that are better by a slight margin.
                                Why the idea that one product must have at least exactly the same performance as its competitors to be "competitive"? We get it, amd is a bit behind in terms of power consumption. But for me it's well in the region of being competitive.. Let's see what prices do in the next time, maybe they go down enough to make the difference?

                                When bulldozer was new there was much talk about how since that was a completely new design it would be far more optimize-able than the intel cpus that have been there for quite a while.
                                I think relatively speaking bdver2 is - relatively speaking - nearer at the recent intel cpus than bdver1 was at the intel cpus back then. And isn't power consumption better too?
                                Of course intel's cpus get better too. But I don't see as much people talking about the trends as I thought I would see.

                                Comment

                                Working...
                                X