Announcement

Collapse
No announcement yet.

DDR4 VCCSA (System Agent) Voltages

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

  • DDR4 VCCSA (System Agent) Voltages

    I'm using ripjaws (F4-2400c15q-16GRK) and I'm wondering what typical voltages would be like for System Agent to stabilize this kit. It would appear that an offset of 0.250 yielding voltages of 1.150 ~ 1.160 seems to work. Anything lower seems to trigger a bluescreen (usually code 124).

    Is this higher than usual? What has been the experience of others using such memory?

    Bonus question: I recently got a bluescreen of code 0x9f when attempting to overclock my memory. To test, I left the memory at stock and overclocked the CPU yet didn't trigger this error. The system ran fine a for a number of days and passed many hours of stress testing. This particular code seems to indicate a power-state issue. I overclocked the memory again and attempted to sleep the machine. No errors. I'd likely suspect a offset voltage setting to be the issue but can't seem to pin point the genesis of this error.

  • #2
    That is normal, but generally the motherboard should be able to automatically run the RAM without a manual boost.

    If you raise voltage or timings, you should be able to reach a higher DRAM Frequency. Let us know how it goes.

    Comment


    • #3
      Originally posted by GSKILL TECH View Post
      That is normal, but generally the motherboard should be able to automatically run the RAM without a manual boost.

      If you raise voltage or timings, you should be able to reach a higher DRAM Frequency. Let us know how it goes.
      Thanks for the update! Good to know that those voltages are not far outside the reasonable margin most would expect.

      To thoroughly test the system, I've decided to go back to basics and strip the overclock down to just the CPU. Currently stress testing now actually.

      CPU: 5930k @ 4.2 Ghz
      motherboard: Asus x99 Deluxe
      vcore: 1.27 v
      input voltage: 1.9 v
      Load-line calibration: level 1

      After 1 hour and 15 minutes I'm still stable with those settings. I'll run for another hour to ensure these settings. Maybe even another two hours. At that point, I'll open up the browser, a media player, and start monkey-smashing the system to instigate a crash if I can. If not, on to the memory!

      I intend to start with auto settings on my motherboard with the CPU settings still at manual. This will remove any ambiguity with offset/adaptive voltage and stability. I recall the motherboard initially suggesting a system agent voltage of 1.05 volts and having failed at that setting. But since then I've found many online posts and articles suggesting an increase to vcore when encountering a BSOD of 0x124. 0x101 seems to definitely suggest upping the vcore but 0x124 seems a touch trickier. Part of me wonders if the system agent voltage would need a boost due to that error akin to the days of Nehelam based chips. I recall needing to boost the uncore when I got 0x124 as an error during testing on a i7 930. I'm not sure that is a transferable method to a Haswell-e based chip though. 0x9C and 0x0A are referenced as system agent problems at hardwareluxx. Soon to find out I suppose.

      I will release my findings as they come! Hopefully my tinkering will help others.

      Comment


      • #4
        After 2 hours and 24 minutes of stress testing on prime95 I feel confident that the CPU is stable. With that in mind, I decided to proceed onto the memory and get that overclocked.

        Heading into the bios, I set the defaults for voltage to begin with while enabling the XMP profile for the memory. The board has decided on the following values by default:

        Input Voltage: 1.9 volts
        System Agent Voltage: 1.05 volts
        DRAM Voltage: 1.20 volts

        I ran MemTest86+ to screen for an initial errors. After one hour, everything passed.

        Next I moved back into the operating system. Here I ran Prime95 again, only this time I specified for it to chew up 12 gb of system memory.

        It has been running for 2 hours stable. So I moved on to opening up a browser (which I'm using now to make this post) with 5 or 6 tabs, Windows Media Player playing some tunes, and a few other simple applications. All seems well. Time will tell now. Seems like it is time to fire up some games and test that out. Before I do, I'll drop my vcore to adaptive and test things out. I'll also sleep the system and attempt to wake it up several times to test power settings.

        Comment


        • #5
          Final results are in until I otherwise find the system to be unstable.

          Current settings:
          CPU: 5930k @ 4.2 Ghz
          Memory: F4-2400C15Q-16GRK @ 2400
          Vcore: 1.275 volts
          Input Voltage: 1.9 volts
          Load-line Calibration: Level 1 (0%)
          DRAM Voltage: Auto (1.2 volts)
          System Agent: Auto (1.05 volts)
          Cache Voltage: Auto (1 volt I believe, possibly less)

          Stable in all tests so far. Anyone thinking about picking up memory who happens upon this post, the kit I picked up is highly recommended.

          Notes:
          I will note here that a great deal of guides and posts online state people getting up to 4.5 Ghz and beyond stable at 1.2 ~ 1.3 volts on the vcore. I don't buy this. I think a great deal of these overclocks are only using Aida64 for stability testing. Please don't take that as a knock to Aida64 as a utility. It is a perfectly fine application offering much more than just stability and stress testing. But I'm left with the impression that if I were to stroll up to one of these systems and run Prime95 for even 10 minutes they'd likely encounter a bluescreen. The take away here is to test YOUR processor to assert what it can do and to do so to YOUR satisfaction. Perhaps some of these people won the silicon lottery. Maybe I'm just to strict on stability testing. In either case, you need to run your own testing.

          There is a great deal of talk about Prime95 over-drawing voltage on Haswell-e. While this might be the case (though I personally didn't witness such a thing in my testing) you should test with static voltages. Adding adaptive or offset voltages should be the crowning moment; it shouldn't be a starting point. Run stress tests at your own risk but know that I tested with Prime95 right up to 1.3 volts on my vcore and my processor didn't go up in smoke. Again, it's your call.

          With cashmere components like the Asus x99 Deluxe, it is advisable (as was pointed out by the representative in this thread) to use 'Auto' unless you fail stability testing or encounter stability issues with the system. I've found that on the latest bios (version 1206) I'm able to run most settings at Auto just fine. Hell, even the vcore on Auto at 4 Ghz was stable. Top end companies put a great deal of time into their testing to provide a smoother experience for you. If you are willing to fork over the cash then be at least willing to test their defaults. You might be surprised.

          Overall, I'm very happy!

          Comment

          Working...
          X