X58 Uber RAM Speed?

Discussion in 'Die-hard Overclocking & Case Modifications' started by BLEH!, Dec 4, 2012.

  1. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    I've been tinkering again. Figured since I don't use 24 GB of RAM very often I might as well ditch half of it and run the rest faster. I've got the mad 2000 MHz Corsair Dominator GT, currently at 1866, standard timings for 2K (9-10-9-27-2T), but I'm wandering if it's possible to get 'em to 2133 or even 2400 MHz via looser timings or some other weird/magical trick. My CPU (think 980X but a Xeon) has the multi's for it, even at stock bclock, so it should be possible, in theory. Just a question of whether the uncore can handle the required frequencies (up to 3600 MHz).

    Any ideas gurus?
     
  2. Veteran

    Veteran Ancient Guru

    Messages:
    12,089
    Likes Received:
    6
    GPU:
    2xTitan XM@1590Mhz-CH20
    The only way your gonna know is try it.
    Your Cpu OC will go up as well ( I take it your not using an XMP profile).

    Question is.....Your Ram may be capable but is your cpu? Does your cpu have the right cooling to go higher?

    Is the NB/SB/Mosfets cooled enough to allow stability once you OC all of it?

    Even if you have all of the above you need to find the right settings.

    Only takes one of these to fail and the whole lot fails.

    Personally 4.4 with your cpu is very good on Air, I think if you want to higher than 4.5 your gonna need to get better cooling.

    Try it and see, Best of luck:)
     
  3. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    I can hit 4.53 GHz at 1.4V, but that's as high as I like to go with this chip, starts hitting high 70s under stringent benches there, 80 under IBT/Linx. The NB/SB/2xNF200/mosfets all have this fancy heatpipe/radiator array sat on top of em which I reckon keeps em cool, never gets more than warm to the touch even when under max stress. I'm fortunate with this CPU in having an unlocked multi, but it might be worth dropping that a bit and upping the bclock to see how high I can get at the "stock" ram timings, beyond 2000 MHz. Given that X58's standard bclock is 133, to get to 2K it has to up it to 142 anyway, so that should be reachable, increases QPI bandwidth a little as well then. I'll give it a shot tomorrow.

    **EDIT**

    Currently running P95 blend at 2000 MHz ish, seems stable, will be stressing more at higer speeds later :D
     
    Last edited: Dec 5, 2012
  4. Veteran

    Veteran Ancient Guru

    Messages:
    12,089
    Likes Received:
    6
    GPU:
    2xTitan XM@1590Mhz-CH20
    Yeah thats pretty impressive oc for 4.5, I havent even seen many 980's do that on air infact i have not seen any do that on air.

    You have a good chip there well worth hanging onto, if you had water setup you could probably go 4.7 minimum...?Saying that not worth the amount of volts you would have to push through it.

    Im at 1.40.5v for 4.4 24/7 use no problems whatsoever.
     
    Last edited: Dec 5, 2012

  5. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    It's stable at 2K RAM, now to push that a bit further. I'm not sure whether this is an uber cherry picked Westmere chip but it really is killer for what I paid for it (£400 ish), half the price of the 980X AND with the unlocked turbo multi, too. I have contemplated water but it is a LOT of effort and probably cost at least £500 with all the gear I have in this case, not to mention I'd need probably 480 mm of rads just to cool everything, and the custom blocks for my board are impossible to come by nowadays. I might try some benching at higher clocks by ripping the side off my case and whacking some extra fans on there to see how far I can push it voltage wise under reasonable temps, but hell 4.5 isn't bad for 6 cores on an air cooler really.

    **EDIT** OK so it won't even boot at 2050 MHz, I'll have to try something else. timings maybe?

    **FURTHER EDIT** OK, so even anything a little over 2000 MHz is totally unstable, won't boot or crashes on loading, similarly trying to get any more out of the uncore over 3000 at such speeds is seemingly impossible. Maybe needs a bit more juice (currently at 1.3375V IIRC). Max i'm willing to go is maybe 1.35V, no more. Either that or knock back to 1866 and try tightening timings.
     
    Last edited: Dec 5, 2012
  6. Veteran

    Veteran Ancient Guru

    Messages:
    12,089
    Likes Received:
    6
    GPU:
    2xTitan XM@1590Mhz-CH20
    Its either voltages, Cooling, Timings or could be other settings that are in the bios, Could be anything, Once you start hitting th maximum threshold its hard to know without better cooling.
     
  7. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    I've stuck it back at 1866, see where I can get there, although it doesn't even seem to be stable at 9-9-9-27, dropping the 2nd one from 10, although this does seem to allow the uncore to run a good deal faster, which might make up for lack of timing tightness. IIRC, though the big one with memory is the first number right?
     
  8. Veteran

    Veteran Ancient Guru

    Messages:
    12,089
    Likes Received:
    6
    GPU:
    2xTitan XM@1590Mhz-CH20
    Im not sure, I just ran my timings at recommended timings and took everything else off auto.

    They was unstable until i put an aircooler( Dominator Airfan) on them.
     
  9. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    I do have one of those, two in fact, might be worth sticking it on, see what difference it makes.
     
  10. ---TK---

    ---TK--- Ancient Guru

    Messages:
    22,135
    Likes Received:
    0
    GPU:
    2x 980Ti Gaming 1430/7296
    you got a real good chip there I would keep it at 4.4 and call it a day. if you can get the ram to 2000 I would call it a day also. test your memory bandwidth at 1866 and at 2000 with aida64 and see if its worth it with looser timings
     

  11. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    It runs basically at the same timings at 1866 and 2000, just I think there's a bit more in the cache/uncore to be had at 1866. I'll do a bit more testing tomorrow and find out. Cheers.
     
  12. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    Right, I've done some more testing at 1866 and 2000. Seems The QPI voltage was holding me back a little bit, I've been able to get much more out of 1866 MHz at 1.35 V vs my previous 1.325. Tested bandwidth over various uncore settings at each speed, at least the ones that were stable.

    At 133 MHz blcock (RAM 1866 MHz):
    Uncore (MHz)/Bandwidth (GB/s)
    2933/23.87
    3066/34.49
    3200/24.92
    3333/25.38
    3466/25.80
    3600/26.24
    3733/26.58 <-- seems like this is where it starts to tail off.
    3866 - Not stable at 1.35 V, might be at 1.375, but that's pushing it with my chip.

    At 143 MHz bclock (RAM 2000 MHz)
    Uncore (MHz)/Bandwidth (GB/s)
    3003/24.67
    3146/25.50
    3289 - Wouldn't even boot here, bootmgr unstable, even at 1.375 V, and I'm not risking 1.4 V.

    Gunna do a bit more later with tighter timings at 1866 MHz (now that I've found the QPI voltage again), and maybe run some tests at 1600 MHz, see where that gets me.

    Any more suggestions/comments welcome. I do have the excel graph if anyones interested. Bandwidth measured using SANDRA.

    Also seems to like to forget a stick at higher uncore speeds...
     
    Last edited: Dec 6, 2012
  13. Veteran

    Veteran Ancient Guru

    Messages:
    12,089
    Likes Received:
    6
    GPU:
    2xTitan XM@1590Mhz-CH20
    Sorry for late reply been out all weekend.

    I think that from what you have shown the only way your going to find stability at the higher OC is to up the voltage which as youve said you dont want to do.

    Might be wise leaving as it is if you not going to go higher....decisions decisions:)
     
  14. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    Indeed. I'll be doing a lot more testing in the coming days, just in the middle of moving into my own flat atm, so haven't been able to run all the possible iterations of speed/timings yet. Needless to say it's coming out with some interesting results. I'll post 'em when I've got some reasonable conclusions to draw. There are a LOT of variables in this to test.
     
  15. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    So to reopen an old thread a little. Here's the effect of RAM speed/timings at 1600/1866/2000 MHz with uncore ratio.

    [​IMG]

    Some of the 1866 ones aren't wholly stable at the higher settings. I was able to drop the timings a bit with higher QPI/VTT voltage to 8-8-8-2X, can't remember the exact settings, but the gains aren't anywhere near as good as with running the RAM at 2000 MHz. Might give that a go, OC it the old fashioned way, see if I can get 2200 by upping the BClock. The extra QPI/VTT seems to give it a good deal more overhead anyway, with both uncore multi and timing tightness, though 2K may be pushing the upper limits of what the memory controller can handle, people running 2400 on X58 isn't unheard of.

    *Edit* The axes on that graph should be labelled Uncore speed along the bottom (MHz) and memory bandwidth up the side (GB/s).

    Any thoughts?
     
    Last edited: Dec 26, 2012

  16. Veteran

    Veteran Ancient Guru

    Messages:
    12,089
    Likes Received:
    6
    GPU:
    2xTitan XM@1590Mhz-CH20
    Yeah i found the QPI/VTT has to be pretty much spot on for me to hit my OC on my CPU.

    If i went to high or too low my cpu would become unstable in either Linx, Occt,IBT and Prime 95.

    If my CPU fails in just one of these tests then it fails in my eyes as i like to be stable in all tests as there all designed to find a weakness in a different way to each other otherwise what would be the point in having multiple stability tests, Thats just me though

    So i found through lots of testing that there is a *sweet spot* regarding the QPI/VTT.

    2400mhz is not unheard of but it is rare on x58 when you run in synchronisation with the cpu OC.

    The thing is its easy just buying some 2400mhz ram and using a 2400mhz profile and bang it works as long as your motherboard supports it.

    However once you start throwing CPU OC sync in as well its gets much harder because you need a cherry picked cpu to allow you to get the oc on the cpu up to near the ram maximum which also requires great cooling and even if you succeed in doing this, It is still going to be nigh on impossible as the CPU wont OC as much as the ram will unless its capable of hitting an OC way way up there with the maximum ram potential OC.

    I dont think this is possible even with a 990x on X58.

    High end clocking ram is supposed to be used as an XMP profile but to use it in sync with cpu OC, Well i dont think its possible unless you put the CPU under Phase change or LN as the 2400mhz is just too high for any x58 cpu and for this reason i think that its a waste of time buying high end ram for x58 as it can only really be used as an xmp profile, But by doing this you lose out in the CPU OC.

    I never use XMP profiles, Waste of time for me:)

    Im sure you will push it to the absolute max and i wish you the very best of luck.

    You have a great Cpu there!
     
    Last edited: Dec 26, 2012
  17. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    Thanks Vet. I'm doing a lot more testing now I'm off for xmas/getting that housing crap sorted out. Limiting myself to 1.35V on the QPI/VTT as much more than this CAN degrade the memory controller, and I really don't want to fry my CPU. Vcore is fine as is, I've got enough overhead to hit 4.5 changing other settings should that be the case. I'll post stuff as and when I find out what happens.

    UPDATE 27/12/2012

    Well, at 1.375 V on the Vcore, 1.35 V on the QPI/VTT, 1.9 V on the CPU PLL, 1.64V on Vdimm and a slight bump on the NB Core and PCIe voltages to account for 3 GFX cards (+0.08 V from stock on both), I can safely say that the RAM isn't going to give ANY more than 2015 MHz, regardless of Core multiplier, Uncore multiplier, or X.M.P timings. It seems to be stable at 143 and 144 MHz bclock at XMP settings (for 2000 MHz and 2015 respectively), but bumping it up to 145 renders it totally unstable, even the BIOS has trouble loading, so getting any faster than that is really going to be a struggle. I'm tempted to knock the CPU back to an x15 multi and try fiddling a few voltages, see if I can get it any higher, but the QPI/VTT are REALLY pushing what I want to be at on air cooling, unless the QPI is holding me back at approx 7 GHz, which it shouldn't be, given it is capable of 8. Is NB core voltage likely to have any great effect or not really?

    Thoughts welcome.
     
    Last edited: Dec 27, 2012
  18. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    Giving it 1.66V on the Vdimm has REALLY opened things up.
     
  19. Veteran

    Veteran Ancient Guru

    Messages:
    12,089
    Likes Received:
    6
    GPU:
    2xTitan XM@1590Mhz-CH20
    Yes that will show up in the red, I have mine at 1.65v.
    What settings was yours at before 1.66v?
     
  20. BLEH!

    BLEH! Ancient Guru

    Messages:
    5,654
    Likes Received:
    2
    GPU:
    Sapphire Fury
    1.64V, there's no intermittent setting, unfortunately.

    I am wary about frying the IMC, although at such a low setting it should be alright, given QPI voltage has been raised, although there are a few issues with other stability now.
     

Share This Page