Announcement

Collapse
No announcement yet.

The Sandy Bridge Core i7 3960X Benchmarked Against Today's Six-Core / 12 Thread AMD/Intel CPUs

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

  • #31
    Originally posted by perpetually high View Post
    Btw: that nice command is inverted. That's giving the process a priority of +12 (lower priority) instead of -12 (higher priority) which I'm sure is what you wanted.

    I find "nice -n -12" to be easier to remember and more intuitive than the double hyphen (nice --12) and prevents accidentically using the positive number instead of negative.
    With normal user privileges, you can't usually use a negative niceness.

    Comment


    • #32
      Originally posted by coder View Post
      With normal user privileges, you can't usually use a negative niceness.
      Yeah I have:
      Code:
      myusername  -  nice  -20
      in /etc/security/limits.conf which allows it to be changed. Also required for Feral's GameMode for it to change niceness.

      Comment


      • #33
        Originally posted by perpetually high View Post
        Yeah I have:
        Code:
        myusername - nice -20
        in /etc/security/limits.conf which allows it to be changed. Also required for Feral's GameMode for it to change niceness.
        I did say "usually". If someone doesn't know what they're doing, they're not going to accidentally swamp the CPU. But, if you give yourself permission, of course you can shoot yourself in the foot.

        Edit: And, speaking of shooting yourself in the foot, I wouldn't go all the way to -20. IMO, -8 should be fine. Even if you have background work going on, just increase its niceness.
        Last edited by coder; 09-14-2019, 05:34 PM.

        Comment


        • #34
          Originally posted by coder View Post
          I did say "usually". If someone doesn't know what they're doing, they're not going to accidentally swamp the CPU. But, if you give yourself permission, of course you can shoot yourself in the foot.
          Right on, makes sense for it to be the default. I only mentioned it in case someone else didn't know, as your post was clear since you used normal and usually.

          Comment


          • #35
            That $999 Sandy Bridge would cost $1140 in today's dollars.

            But also, the depreciation left of that same Sandy Bridge in 2019 is down to zero.

            The average used price now is $99.

            Since AMD didn't have anything to counter at the time, it's value was somewhat maintained.

            Now that Zen is out, it clearly shows how badly Intel was overpricing even then.

            Comment


            • #36
              Originally posted by edwaleni View Post
              Now that Zen is out, it clearly shows how badly Intel was overpricing even then.
              Huh? How does Zen show that?

              Comment


              • #37
                Michael
                My 3960X is still running strong in 2019: results are about 1.5-2X of yours with this CPU. I tested against your 1909137-AS-3960XRYZE03 results and almost in every test I get 1.5-2X better score. My 3960X is slightly OC to 4 [email protected] core, and Ubuntu is 20.04(newer compiler, etc), but this could not so massively boost perf. on old hardware. Smth is wrong with your 3960X: it is seriously underpowered. Maybe you're using slow 2xchannel memory(you said it's 4 channel, but please check)? Mine is 16 GB = 4 channel x 4 GB DDR3 2133 CL11 XMP profile (which should be used with these CPUs to achieve proper performance). I see my 3960X only a bit lags behind R5 2600X, and geomean of all results is only about 1.5-2X slower than R5 3600X. What I get in some tests:
                test ph 3960X stock ubuntu 18.04 My 3960X @ 4.0 ubuntu 20.04 diff, X
                build-linux-kernel, s 247 132 1.87
                build-php, s 140 82 1.7
                build-llvm, s 1484 745 1.99
                svt-av1, FPS 1.9 3.4 1.78
                nginx, req. per s 10500 18600 1.77
                apache, req. per s 12386 15595 1.26
                mcperf Get, ops per s 31019 53260 1.71
                redis GET 254685 1906000 7.48
                perl interpreter, s 0.00176 0.00119 1.48
                hackbench c8 process, s 75.7 41.6 1.81
                indigobench super car, Msamples/s 0.94 1.7 1.8
                blender BMW27, s 706 389 1.81
                node-express-loadtest, req. per s 2831 4944 1.74
                rodinia v 2.4 CFD, s 78 46 1.69


















                And this goes on for almost every test in your result named 1909137-AS-3960XRYZE03. Please re-run your tests, you have smth seriously wrong with 3960X.

                Here is my result published for reference (normalized view):
                https://openbenchmarking.org/result/...AS81&obr_nor=y

                System tests are using newer versions of apps: gimp, darktable, etc, so their results are separate in the end. Disregard indigobench results - these are GPU not CPU result, phoronix-test-suite has a bug: it is not able to get CPU results from the output if both CPU & GPU scores are present.
                Also note the buggy 5.7 GHz frequency detected: it is not able to get the correct CPU freq, actually it is 4.0 GHz.
                Last edited by dad_ph; 11-12-2019, 12:25 PM.

                Comment


                • #38
                  Originally posted by dad_ph View Post
                  And this goes on for almost every test in your result named 1909137-AS-3960XRYZE03. Please re-run your tests, you have smth seriously wrong with 3960X.
                  What about security mitigations? Here's what Michael used:
                  l1tf: Mitigation of PTE Inversion;
                  VMX: conditional cache flushes SMT vulnerable +
                  mds: Mitigation of Clear buffers; SMT vulnerable +
                  meltdown: Mitigation of PTI +
                  spec_store_bypass: Mitigation of SSB disabled via prctl and seccomp +
                  spectre_v1: Mitigation of __user pointer sanitization +
                  spectre_v2: Mitigation of Full generic retpoline
                  IBPB: conditional IBRS_FW
                  STIBP: conditional RSB filling
                  Source: https://openbenchmarking.org/embed.p...ha=df33cac&p=2

                  and storage config?
                  Disk: 500GB Samsung SSD 860
                  Are you using PCIe-based NVME, by any chance? I am.

                  Also, some Sandybridge-E motherboards only have a couple SATA-3 ports. Mine only had 2, with the rest being SATA-2. Maybe he accidentally plugged the SSD into a SATA-2 port.

                  Also, he doesn't say how full it is, which tends to affect write performance. Especially, if you never run fstrim.

                  Anyway, thanks for posting up your results.
                  Last edited by coder; 11-16-2019, 06:30 AM.

                  Comment


                  • #39
                    Originally posted by coder View Post
                    What about security mitigations?
                    See security details in my result (ref. is in the post above):
                    Core i7 3960X 4 GHz DDR3 2133 4x4GB CL11:

                    HTML Code:
                    l1tf: Mitigation of PTE Inversion;
                    VMX: conditional cache flushes SMT vulnerable +
                    mds: Mitigation of Clear buffers; SMT vulnerable +
                    meltdown: Mitigation of PTI +
                    spec_store_bypass: Mitigation of SSB disabled via prctl and seccomp +
                    spectre_v1: Mitigation of usercopy/swapgs barriers and __user pointer sanitization +
                    spectre_v2: Mitigation of Full generic retpoline
                    IBPB: conditional IBRS_FW
                    STIBP: conditional RSB filling
                    Looks exactly the same as Michael used. ( ithink this is standard for all more or less modern ubuntu distros - I not touched any settings related to security)
                    SSD/storage has nothing to do here: all the benchmarks used are only CPU/memory bound, so storage affects 0 results. My system/boot drive is very fast PCI-E SSD (samsung 950 pro 500 GB), but I install and run all the PTS tests off the hard drive. (6 TB WD black, I have only 10 GB space left on my linux partition of SSD)

                    I'm surprised someone is still interested in LGA2011 system results I posted it to emphasize that actually HEDT 6C/12T from 2011 is still not so bad compared to at least AMD R5 2600X, it is ~ R5 1600 I think, and in some games it even may beat it, so the improvement (upgrading to modern 6C/12T CPU) is actually "not so big". And used 6-8 core sandy or ivy bridge e/ep (OC to 4.5+ GHz especially) are still strong in 2019. Intel CPU has so small progress in 8 years: take 9900K and I doubt you could double my results in every possible load, it's a shame. OC SB-E to 4.5-4.7(which should be possible in case of proper liquid cooling) and the advantage of 9900K is even less than that.

                    Comment


                    • #40
                      Originally posted by dad_ph View Post
                      I'm surprised someone is still interested in LGA2011 system results I posted it to emphasize that actually HEDT 6C/12T from 2011 is still not so bad compared to at least AMD R5 2600X,
                      Yeah, I actually have a Xeon E5 quad-core. Stock speeds (cuz I hate instability), with quad DDR3-1600 (16 GB) and NVMe storage. It's fast enough and (more importantly) I'm too lazy to replace it. I think the motherboard's Toslink output just died, though. I was beginning to feel like that machine might run until the end of time.

                      Originally posted by dad_ph View Post
                      liquid cooling
                      Ugh. I just can't. I worry about air getting into it, making weird noises, and me being just lazy enough to live with it.

                      Comment

                      Working...
                      X