1. This site uses cookies. By continuing to use this site, you are agreeing to our use of cookies. Learn More.

huge difference in CPU score on Time Spy changing mem to C18 from C19

Discussion in 'Overclocking & Cooling' started by Apone70, Feb 10, 2019.

  1. Apone70

    Hitman

    Joined: Aug 15, 2018

    Posts: 571

    Ok so i run my 8pac 3200mhz memory at 4000mhz C19-19-19-39. If i change it to C18-19-19-19-39 my cpu score on 3d mark drops from 8865 down to 6641 (8700k at 5ghz). I was just wondered if anyone knew why as its a huge drop, the system is still stable.

    https://www.3dmark.com/compare/spy/6169803/spy/6169713
     
  2. mattyfez

    Sgarrista

    Joined: Apr 12, 2007

    Posts: 8,941

    I could answer that, but bare in mind that 3dmark is a somewhat an artificial test.

    Real proof of the pudding would be in game average FPS and minimum FPS.
     
  3. Jokester

    Don

    Joined: Aug 7, 2003

    Posts: 37,997

    Location: Aberdeenshire

    I could be wrong here, but a system can cope with an element of errors due to RAM instability that becomes apparent with lower CPU performance as the CPU is having to do extra work to rectify the errors in memory reading.
     
  4. Minstadave

    Capodecina

    Joined: Jan 8, 2004

    Posts: 23,865

    Location: Rutland

    With ECC maybe but I've never seen a system do this.
     
  5. Jokester

    Don

    Joined: Aug 7, 2003

    Posts: 37,997

    Location: Aberdeenshire

    Yes, it's hard to explain but I've seen this in the past, both with system RAM and GPU RAM, where before it falls over completely, performance drops off. The only way I can explain it is if the processor is spending extra time trying to get data to/from RAM, so overall system performance takes a nose dive.
     
  6. andy_mk3

    Sgarrista

    Joined: Oct 5, 2009

    Posts: 8,044

    Location: Lincolnshire

    I've seen it on vram overclocking before.
     
  7. Minstadave

    Capodecina

    Joined: Jan 8, 2004

    Posts: 23,865

    Location: Rutland

    VRAM yes, most VRAM is ECC these days.

    Edit - actually that could be wrong. I thought they had error correction but looks like I'm wrong.
     
  8. EsaT

    Soldato

    Joined: Jun 6, 2008

    Posts: 5,189

    Location: Finland

    That would be like 12x CD-ROM drive I once had:
    It would accelerate but often do fast some read error slowing down for re-read, before accelerating to do another read error.
    When I bought CD-R writer it had only 8x read speed, but was actually faster because of maintaining its read speed continuously.
     
  9. Danny75

    Soldato

    Joined: Aug 26, 2013

    Posts: 6,741

    This thread reminds me of Aesop's Fable 'The Tortoise and The Hare'.
     
  10. Apone70

    Hitman

    Joined: Aug 15, 2018

    Posts: 571

    Lol
     
  11. JCLDJB

    Associate

    Joined: Jan 9, 2019

    Posts: 72

    That is a huge drop for such a small change, but remember that overclocking your RAM to anything above the rated frequency of 3200Mhz, and then tweaking the mem timings, the mem timings will have a greater effect on stability, the higher frequency you go.

    Something isn't right with that C18 from the C19, causing (just repeating what others have stated here) additional memory error in the reading different memory addresses, the CPU will then try to rectify these errors but it takes additional CPU cycles to fix errors, so your overall score goes down.

    I would tweak more than just the C18-19-19-19..try 18-18-18-18 or play around with the '39' at the end, dropping it by 1, or increasing it by 1-2.

    Also, if you want to see if you are producing more memory errors while benchmarking, opel up Task Manager, under Performance and click on Open Resource Monitor then go to the Memory option & on the right you have standard Task Manager graphs/timelines, there is one that shows how many 'faults/sec' you're getting (if I remember this rightly).

    Check the difference in number of memory errors produced from your previous OC to this one.

    Alternatively, if you really want to guarnatee full stability, Memtest for 6-8 hours (or minimum 2 passes, ideally 3 = 10-12hours-ish or more) and you will know for sure, how bad that one little timing change actually was (and how stable your system actually is).
     
  12. Apone70

    Hitman

    Joined: Aug 15, 2018

    Posts: 571

    Cheers i will have a play when i have some time on my hands.