Jump to content
Search In
  • More options...
Find results that contain...
Find results in...


  • Content Count

  • Joined

  • Last visited

Everything posted by Falkentyne

  1. The PSU was still a good purchase. I wish I had known you were using an AVX offset however. Crashing in non AVX apps with an offset is NOT an Asus bios bug but a known issue that happens regardless of ODM. In order NOT to crash, you have to basically be fully AVX stable (like prime95 29.8 build 3 small FFT stable). This is a guardband tolerance issue. Sometimes reducing cache ratio can help. https://rog.asus.com/forum/showthread.php?106375-MCE-explanations-and-others&highlight=explanations
  2. This is not true. The Ultra's VRM's are better on paper than the Master's VRM's (SiC634 vs IR3553), and the VRM controller I'm not sure how to rate (not sure): ISL69138 vs IR35201. I don't think this translates to overclockability but they are more efficient.
  3. The Ultra isn't the 'better' board than the Master. You need to read exactly what the review (I assume, by buildzoid?) or other reviews are clearly saying. What was said was that ON PAPER, the Ultra's VRM's are better than the Master because they are more efficient (Intersil tends to have better components than International Rectifier, but as far as the VRM's themselves, Vishay, not sure, but the power stages themselves are more efficient (Intersil), but both have the same # of phases with doublers. The Ultra/Pro use the intersil 69138 while the Master uses the IR 35201. However the Master has a socketed primary bios (thus easy to remove and replace with a spare from Gigabyte if it goes bad, or even to hardware flash it yourself if you have a SPI programmer, without needing a Pomona 5250 clip and jumper cables), a dual bios jumper select switch (makes it very easy to switch bioses, provided you are not on the early release bioses which would often swap positions whenever they wanted), and extremely good VRM cooling. That makes the Master a decent choice.
  4. MSI and Asrock's LLC are inverted from Asus and Gigabyte, and none of them tell you the mOhms values either. Try Loadline Calibration Mode 2 and 1.35v. Set Uncore at 4300 mhz and DISABLE AVX Offset. Then test. And do not use prime95 small FFT with AVX to test your stability, either. You can use 29.8 build 3 with AVX disabled in the new stress test options (useful) as well as the usual Cinebench R20 (custom 3600 seconds), and Realbench 2.56.
  5. Hello, First of all, each connector that has an LED is polarity keyed because LED powered connectors need a + and - to transmit voltage signals to the front I/O for the lights. The HDD LED and Power LED are both polarity keyed. So what it means is, Pin "1" for HDD LED has positive polarity and pin "3" for HDD LED has negative polarity. So that's where the pins go. Pin numbers on circuit boards are *usually* counted CLOCKWISE for chip IC's (like SOIC8 or DIP8 chips). However there seems to be no clear standard for how they are labeled on jumper pins. In your example, you have them labeled wrong. In this case, they are staggered in groups, so pin 1 is bottom left and pin 2 is top left, then pin 3 is to the RIGHT of bottom left, and pin 4 is to the RIGHT of top left! So your label is wrong. Please don't make fun of my paint skills. I have zero artistic ability. But this is how they are supposed to be labeled. They mark "1" and "8" so you know where to start and stop. So as you can see, pins 1 and 3 are HORIZONTAL orientation on the rectangle sideways (long way). So HDD LED pin "+" goes in the bottom left, and the "-" pin goes right next to it
  6. 4 cores is so 2008. And we've had 4 cores with hyperthreading ever since Nehalem. I would not buy ANY new system (laptop or desktop) with only 4 cores at this point in time. Multithreading is becoming more important and with core counts increasing quickly (compared to the.....ahem....last 8 years, where we were stuck on 4 cores forever), I would say 6 core minimum. I would also NOT buy a 6 core without hyperthreading (regardless of what the "exploit" fearmongers keep saying about being scared of exploits. Normal end users will NOT be affected by exploits, and mitigations can be disabled).
  7. Working as expected. In windows you are never going to see the x50 multiplier with turbo boost and rarely the x49 multiplier when at load. Windows task scheduler constantly shifts tasks between different threads and there are background processes also running on more than one thread. This interferes with turbo boost. If you use Throttlestop 8.70+ to monitor clock speeds (instead of HWinfo64) and set it to "More data", you may see your x49 or x50 more often. I don't think HWinfo64 will update faster than 250ms per poll. If you really want to see the higher ratios active at lower loads, you need to run something like Linux.
  8. Is multicore enhancement enabled or disabled? Did you at any time set a manual multiplier? I believe setting a manual multiplier in one of the options forces the multiplier for all cores to be that multiplier if a second option is also changed. I also 'think' one of the power options if disabled stops the cores from boosting higher than the multiplier. I don't know what, because I always, instantly disable MCE, EIST, Speedstep, and all C-states and Ring Down Bin, then I set a fixed multiplier and cache ratio (like 50/47 or 47/44), and my chip runs at that multiplier and cache-300 mhz at all times without downclocking. With the above, when I set x47 for the core and x44 for the cache, the CPU runs at x47 and never clocks higher with a lower core thread load. Keep in mind forcing all cores to x50 requires higher voltage and may require CPU Vcore Loadline Calibration set to turbo. Also the VR VOUT section of your HWinfo64 shows true CPU voltages and VRM and amps monitoring (if there are two sections, one is for the iGPU and you can hide that section if it shows 0.004v for VR VOUT).
  9. That PSU is a "Great Wall" PSU and was mediocre quality in *2011*. OCZ used to make decent products and PC Power and Cooling used to be a Seasonic OEM, but when OCZ bought PCP&C, all of their products went downhill in quality *FAST*. Even their RAM went downhill and their QA started slipping. They were even making SSDs of questionable quality. Needless to say that company went bankrupt not long after and good riddance. God only knows how much power that Great Wall PSU can deliver to the 12v CPU rail (plugging in both 8 pin power connectors helps minimize resistance and voltage drop on the 12v rail also). Also you didn't tell any of us what Loadline Calibration level you tried to use. Try level 6 on the Asus.
  10. Falkentyne

    I7 9700K Not Turbo Boosting?

    Also current version of prime95 is 29.8 build 3. https://www.mersenne.org/download/
  11. Falkentyne

    9900k on a z370 board ?

    VRM's will only get that hot when running power viruses like Prime95. You don't need to run prime95 to enjoy a PC.
  12. Falkentyne

    New Microsoft Flight Simulator Announced

    I'm sorry, but the flat painted textures of cars in parking lots when the plane was landing just did it for me. I'm emotionally scarred for life. I mean...we had that stuff during the Geforce 2 Era. Yikes....
  13. Just ask all the hackers in free to play games, who get banned for cheating, then keep coming back with new spawned accounts and keep doing the same thing over and over. Sure, eventually the hacker gets tired and gives up. Companies will just treat him as inmate#65535. Then you take an arrow to the knee when you realize the thousands of other hackers are doing the exact same thing. Oof.
  14. Falkentyne

    i7-7700k and RTX 2080 Bottleneck

    Battlefield 5 can use more than 8 threads. I believe it can use up to 12 threads, making an 8700K / 8086K do some nice work. Pairing a 2080 RTX with a 4 core (8 thread HT) CPU is going to make the CPU choke as the extremely fast video card is going to be held back by that processor. That's just how it goes. If you tested it with an 8700K with that card, you would get much more consistent performance. Remember we've had 4 cores with hyperthreading for at least 10 years now.
  15. 5.2 ghz is a completely unrealistic scenario on a non R0 stepping or non KC (whatever that iGPU disabled version is called) 9900K. Silicon Lottery themselves can only get 8% of their samples at 5.1 ghz stable. 5.2 ghz with HT enabled is a complete pipe dream unless you have a golden CPU or you go direct die plus water cooling or sub-ambient cooling. Disabling hyperthreading will usually get you an extra speed bin. LLC on Ultra will not only actually require MORE load voltage to remain stable due to transient response being worst case horrible at more than light load (based on VR VOUT voltage at load, which is very accurate CPU on-die sense voltage; HWinfo64 will show you this very useful voltage), but should *NEVER* be used. Using more than 1.28v with Ultra Extreme LLC puts your CPU at high risk of slow degradation if you apply a heavy load to it (like more than 120 amps measured in Current iOUT in HWinfo64). LLC Ultra Extreme violates Intel's loadline specifications for voltage droop (removing all droop) which then prevents your CPU from operating at a safe VR VOUT voltage when current goes up. Safe voltages are based on current, starting at 1.52v with a ZERO AMP current (CPU must NOT be subject to any clock signal at 1.52v at VR VOUT), and going down to 1.218v at 193 amps of current. These are absolute MAXIMUM specifications for long term reliability and are based on a 1.6 mOhms loadline (Loadline Calibration=Standard / Normal is 1.6 mOhms, Intel defaults). Increasing loadline calibration reduces this loadline (reducing vdroop by lowering the mOhms) which obviously throws the entire specification on the floor, since load voltage is no longer reduced as much. You are lucky that your board has access to On-die sense voltage and amps and load power monitoring, so it's easy to keep voltages within specification. Some estimated maximum safe voltages (VR VOUT, aka VCC_Sense) at some amps limits I've calculated for 9900K/9700K: 193 amps (virus TDP mode): 1.218v 150 amps: 1.280v (real world heavy load) 100 amps: 1.360v. (real world medium load) 50 amps: (non real world extremely light load): about 1.440v. LLC should be left at Turbo in most cases for normal users. Reducing LLC will help stability at very heavy amps, at a certain target voltage VR VOUT, when a certain load voltage is required to maintain stability. This may require raising bios voltage to compensate for increased vdroop, or changing the AC loadline on auto voltages (or combined with auto voltage then applying SVID offset to it, when reducing AC loadline, which is an advanced tweak I can't explain here without writing a gigantic essay).
  16. Computers aren't a value purchase. Common sense tells you to buy the MOST computer you can safely afford, but to avoid buying anything right before new tier hardware comes out, unless there is a fire sale on the older gen gear. So if you have the money to buy a 3950X and a 3900X, why would you or anyone choose the 3900X when they can afford the 3950X? Buy the most you can afford. Buyer's remorse is always when they go value basement and buy low or mid tier, then wind up unhappy because they don't have something they could have had. No, I'm not telling you guys to buy the 3950. Read my post again. i said buy as much as you can afford safely. That's the right way to buy hardware. (of course do your research, wait for reviews, and do NOT buy buggy, defective or hardware prone to catch fire and burn down your city block (Alienware 51m, I'm looking at you).
  17. I don't think you know how prime95 works at all. What you're asking for is completely impossible. An iteration finishes based on how fast your processor is and the # of threads, NOT by the time you give the FFT to be tested for. What you asked it to do with your settings is to do the 128K iteration for 60 minutes, then to start over. This would be obvious if you used a preset, then changed to a custom (e.g. 24K-85K range) and then did 60 minutes. Then you would see 24K last for 60 minutes then move on to 25K (or 26K, whatever is the next valid FFT). What you're asking is for is for something like m2605473 to be calculated repeatedly. This is basically impossible. You may be able to edit prime.txt or local.txt or something but this is advanced stuff and beyond what most any normal user would ever do and is completely useless for you. And the "m"iteration is completely pointless. It's the FFT size that matters, and the load size. The only time the FFT size stress can change is when the SSE instruction changes, e.g. if it changes from Pentium 4 to Core 2 to type 0 or type 2, etc. You can prevent this by setting a "0" minute test time (which is basically ending the test at ONE Iteration and restarting). What you actually want per your OP is to set the test time for 0 minutes. Also, the temps and fan speed are not based on the test 1 and test 2. It's based purely on the FFT size. So you will have the same temps whether you do a 0 minute or 60 minute test. If your motherboard had AMPS monitoring in HWinfo64 (VR VOUT for VCC_Sense voltage and IOUT for current (amps) and POUT for power), you would be able to test this directly.
  18. Falkentyne

    Bad i9 or bad motherboard?

    This doesn't mean that the processor is working or undamaged. Also, the processor will NOT Turbo boost if you manually set a CPU multiplier. Turbo boost only works if the multiplier is set to auto, and I think multi-core enhancement is set to enabled. It's not easy to explain. Several options have to be enabled for the 1-2, 3-4, 5-6 and 7-8 "core load" Turbo boost ratios to work. But we don't want turbo boost to work like that. We want a multiplier override to force a ratio on ALL cores at once. Set your processor to 4700 mhz (4.7 ghz). Set the cache ratio manually to 4.4 ghz. Then try setting CPU Core voltage to 1.25v, VCCIO to 1.15v, VCCSA to 1.15v, max out the Power Limit 1 and Power limit 2 TDP values, set CPU Vcore Loadline Calibration to medium-high (Gigabyte= Turbo, Asus=Level 6, MSi and Asrock Level 2) and then test the processor. If it still crashes, RMA it. It's damaged.
  19. Yes you can. BUT DO NOT USE LOADLINE CALIBRATION AT ALL when using auto voltages!! The IA AC/DC Loadlines will handle the voltage for you. the loadline isn't going to be a linear drop from what you see in the bios when using auto voltages. The AC Loadline will boost the voltage "higher" when a load is applied, then vdroop will drop it down to a safe level.
  20. Falkentyne

    Bad i9 or bad motherboard?

    Don't be so sure. There have been a number of 9900K's that have spontaneously died like this, crashing and BSOD'ing and only working when downclocked to like 1600 mhz with the exact same symptoms. I strongly suggest the OP gets that 9900K fully tested and confirmed to be working and not damaged. And yes, the TUF board is complete garbage also, but it's a very high possibility that CPU's IMC is damaged.
  21. Falkentyne

    Which is my correct VCORE voltage?

    Sorry but this is wrong. Explained in the other reply.
  22. Falkentyne

    Which is my correct VCORE voltage?

    Sorry but you guys are wrong. VR VOUT is the correct value. Not vcore. And definitely not the Super I/O vcore either. https://www.overclock.net/forum/5-intel-cpus/1714622-9900k-large-vdroop-load-voltage-main-1-watch-2.html#post27736104 https://www.overclock.net/forum/27686004-post2664.html The 8792E Vcore is the Socket MLCC caps, which wont show live vdroop properly at all. If you set 1.20v in the bios with LLC Turbo, and run a 150 amp prime95 AVX stress test, it will fluctuate between 1.188v to 1.212v, regardless of the load, which doesn't make any sense. That's because vdroop (loadline) is R*I, resistance times amps. In this exact test I mentioned, VR VOUT in this example would show about 1.137v if you calculated it as mentioned below. You can easily calculate the accurate VR VOUT if you know the "mOhms" value for your loadline calibration. For 4 and 6 core CFL, the default (Standard/Normal) loadline if LLC is set to those settings--aka Intel default settings--is 2.1 mOhms. For 8 core CFL it's 1.6 mOhms. LLC High has 50% reduced vdroop (so 1.05 mOhms for 6 core, 0.8 mOhms for 8 core). LLC Turbo is 75% reduced vdroop, so that's 0.4 mOhms for 8 core and 0.52 mOhms for 6 and 4 core. LLC Extreme is 80% reduced vdroop and Ultra Extreme is no vdroop (99%) or 0.01 mOhms. Then to calculate your VR VOUT, take your bios voltage you set and save it, then take the amps (Current IOUT) and multiply it by the resistance (Amps * mOhms or I * R). Record that value (it will be in millivolts, since the resistance is in milliohms). Then subtract that value from your bios voltage (change the bios voltage to millivolts, e.g. 1.200v=1200mv), and that should be equal to your VR VOUT. The Asus Maximus XI boards have been recalibrated to read the same value on their vcore as Gigabyte (and some Asrock and MSI boards) do on VR VOUT.
  23. Falkentyne

    safe Vcores?

    This isn't so simple. Current Intel chips are based a baseline up to 1.52v, where up to 100C, 1.52v to VCC_Sense must have a 0 amp load going to the CPU, and must be subject to Intel default loadline slope. Yes, higher than 1.52v is NOT safe (even at idle) unless you go sub ambient. This loadline is 1.6 mOhms for 8 core CFL and 2.1 mOhms for 4 and 6 core. Using loadline calibration throws a huge wrench into this. I explained in more detail above.
  24. Falkentyne

    safe Vcores?

    Based on *AMPS*. 1.520v is the absolute MAXIMUM safe vcore provided NO AMPS are going into the CPU, based on using Intel defaults of a 1.6 mOhm VRM Loadline. Meaning the CPU must not be subject to a clock signal if this is actually the VCC_Sense voltage being sent to the CPU after vdroop. At full idle, this is going to be less. Assuming a 30 amp idle, and 1.6 mOhms loadline, 30 * 1.6=48mv in millivolts, so a 48mv vdroop at 30 amps, subtract that from 1520 converted to millivolts, that's 1.472v. At the maximum 193 amp rating for 8 core parts, you get 193 * 1.6 = 308mv. Subtract 1520-308 and you get 1.212v, which is the maximum safe vcore at 193 amps power draw. This may seem 'low' but remember this is based up to 100C also. And 193 amps is basically virus mode. *NOTE* These measurements are based on VCC_SENSE voltage measurements! Some boards have VCC_Sense capability linked to a value in HWinfo64 called VR VOUT. Asus Maximus XI boards use a recalibrated super i/o vcore sensor linked to VCC_Sense. If your board does NOT, then the vcore reading you see may be wildly inaccurate *AND* will be more inaccurate the greater loadline calibration you use too as this affects the ground and power plane impedances even more. Read: https://www.overclock.net/forum/27686004-post2664.html https://www.overclock.net/forum/6-intel-motherboards/1638955-z370-z390-vrm-discussion-thread-398.html#post27860326 https://www.overclock.net/forum/5-intel-cpus/1711718-9900k-overclock-results-questions-136.html#post27908122 https://www.overclock.net/forum/5-intel-cpus/1714622-9900k-large-vdroop-load-voltage-main-1-watch-2.html#post27736104 https://www.overclock.net/forum/6-intel-motherboards/1638955-z370-z390-vrm-discussion-thread-266.html#post27685780 Six core parts have a 2.1 mOhm loadline (and 4 core CFL also). Max amps rating for 6 core CFL (4 core is lower) is 138 amps. Doing the vdroop math (Vdroop = Amps * Resistance, or I * R) = 289.8mv. 1520 - 289=1.231v, if you are drawing 138 amps on a 8700K, 8086K, etc. These are the maximum safe voltages on ambient cooling, up to 100C. The instant you change loadline calibration away from Intel defaults of Standard/Normal (1.6 mOhms for 8 core and 2.1 mOhms for 4 and 6 core), all bets are now off, because the intel specifications are based on vdroop existing fully. For example, if you tried setting a manual bios voltage of 1.35v and set Loadline Calibration to 0.4 mOhms (approx. equal to "Turbo" on Gigabyte and LLC6 on Asus--note the Gigabyte value was tested by me but the Asus value is a guess as I don't have an Asus board), and unless someone were willing to post load and bios voltages on *LLC7* on Maximus XI boards only (Not older boards), I cant calculate it for you), if you then tried pulling a 193 amp prime95 load or OCCT 5.0.1 Linpack load, you are going to be WAY out of specification and that CPU will degrade slowly on air cooling. (1350mv - (0.4 * 193))= 1272mv=1.272v <----way out of specification here. And transient response will be much worse as well (much larger voltage spikes and drops due to the VRM's having to work much harder and not having the full vdroop to cushion the spike and drops). Also note that when you use "Auto" voltages (STILL NOT using any custom loadline calibration) rather than setting a bios voltage of 1.52v WITHOUT loadline calibration, your idle voltage will be lower, because the AC Loadline value will influence the target VID and VRM voltage that is sent to the VRM (usually around 1.4v), rather than if you used a manual override voltage of 1.52v. This may not make much sense, but when you apply a load to the CPU (Depending on type of load and even on the AC Loadline value if it's set below the max defaults), the AC Loadline will boost the default VID and thus the VRM voltage target for you up to 1.52v (Yes, that's fully safe as long as you are NOT using a custom loadline, and are not using a positive voltage offset either). DC loadline is responsible for dropping the VID lower but this is NOT sent to the VRM at all--it's only used for power measurements (Namely CPU Package Power, which is equal to VID * Amps). DC Loadline's "VID droop" follows the exact same specification as VRM Loadline with respect to Vcore droop (VCC_Sense), namely VID - (Amps * DC Loadline mOhms)=VID reported to system, as opposed to VRM Vcore - (Amps * VRM Loadline mOhms)= VCC_Sense voltage. VCC_Sense voltage is also known as true vcore or on-die sense voltage (not affected by power plane impedance, if you were measuring it with a scope, and grounded to VSS_Sense).
  25. Falkentyne

    Help Moving forward on 9900K overclock

    Clock watchdog timeout is always insufficient vcore, all the time. Sometimes it's simply the cache ratio being too high. Try the following. 1) set your AVX offset to 0 (yes, zero) then re-do realbench 2.56. (AVX offsets trigger guardband penalties as the PLL's go to sleep when the ratio change is triggered, triggering the -worst case- guardband penalty; that's why some people have instability with AVX offsets and none without offsets. 2) (or) Set your cache ratio to x43 and re-do the realbench 2.56 run.