Chiphell - 分享与交流用户体验

 找回密码
 加入我们
搜索
      
查看: 9376|回复: 22

[内存] 最详细DDR4超频指导+MemTestHelper (转自Github)

[复制链接]
发表于 2019-8-14 21:57 | 显示全部楼层 |阅读模式
本帖最后由 xiaxia686 于 2019-8-15 22:03 编辑

在Github发现一个很好的DDR4 超频指导+MemTestHelper(自动跑个Memtest进程)
链接:https://github.com/integralfx/MemTestHelper
向导链接:https://github.com/integralfx/MemTestHelper/blob/master/DDR4%20OC%20Guide.md

内容:(建议直接去Github上去看)
Table of ContentsSetup
  • Ensure your sticks are in the recommended DIMM slots (usually 2 and 4).
  • Make sure your CPU is fully stable before overclocking RAM, as an unstable CPU can lead to memory errors. When pushing high frequency with tight timings, it's possible that your CPU can become unstable.
  • Make sure your UEFI is up to date.
  • Thaiphoon to show what ICs (integrated circuits or RAM chips) your sticks use. This will give you an idea of what frequency and timings to expect.
  • MemTestHelper or your memory tester of choice. Karhu RAM Test (paid) is also a good choice. I wouldn't recommend AIDA64 memory test and Memtest64 as they are both not very good at finding memory errors.
  • To view timings in Windows:
  • Benchmarks:
    • AIDA64 - free 30 day trial. We'll be using the cache and memory benchmark (found under tools) to see how our memory is performing. You can right click the start benchmark button and run memory tests only to skip the cache tests.
    • MaxxMEM2 - free alternative to AIDA64, but bandwidth tests seem to be a lot lower so it isn't directly comparable to AIDA64.
    • Super Pi Mod v1.5 XS - another memory sensitive benchmark, but I haven't used it as much as AIDA64. 1M - 8M digits should be enough for a quick benchmark. You only need to look at the last (total) time, where lower is better.
    • HWBOT x265 Benchmark - I've heard that this benchmark is also sensitive to memory, but I haven't really tested it myself.

Expectations/LimitationsMotherboard
  • Motherboards with 2 DIMM slots will be able to achieve the highest frequencies.
  • For motherboards with 4 DIMM slots, the number of sticks installed will affect your maximum memory frequency.
    • On motherboards that use a daisy chain memory trace layout, 2 sticks are preferred. Using 4 sticks may significantly impact your maximum memory frequency.
    • On the other hand, motherboards that use T-topology will overclock the best with 4 sticks. Using 2 sticks won't impact your maximum memory frequency as much as using 4 sticks on a daisy chain motherboard (?).
    • Most vendors don't advertise what memory trace layout they use, but you can make an educated guess based on the QVL. For example, the Z390 Aorus Master probably uses a T-toplogy layout as its highest validated frequency is with 4 DIMMs. If the highest validated frequency were done with 2 DIMMs, it probably uses a daisy chain layout.
    • According to buildzoid, daisy chain VS T-topology only matters above 4000MHz. If you're on Ryzen 3000, this doesn't matter as 3800MHz is the typical max memory frequency when running MCLK:FCLK 1:1.
  • Lower end motherboard may not overclock as well, possibly due to the lower PCB quality and number of layers (?).
Integrated Circuits (ICs)Thaiphoon Report
  • Single rank 8Gb Hynix CJR.
  • Single rank 8Gb Micron Revision E (source: Coleh#4297).
    • SpecTek is supposedly lower binned Micron ICs.
    • Esoteric note: Many people have started calling this Micron E-die or even just E-die. The former is fine, but the latter can cause confusion as letter-die is typically used for Samsung ICs, i.e. 4Gbit Samsung E-die. Samsung is implied when you say E-die, but as people are calling Micron Rev. E E-die, it'd probably be a good idea to prefix the manufacturer.
  • Dual rank 8Gb Samsung B-die.
A Note on Ranks and Density
  • Single rank sticks can usually clock higher than dual rank sticks, but at the same frequency and timings, dual rank sticks can perform a bit better due to rank interleaving.
  • Density matters when determining how far your ICs can go. For example, 4Gb AFR and 8Gb AFR will not overclock the same despite being AFR.
Voltage Scaling
  • Voltage scaling simply means how the IC responds to voltage.
  • On many ICs, tCL scales with voltage, meaning giving it more voltage can allow you to drop tCL. Conversely, tRCD and tRP typically do not scale with voltage on many ICs, meaning no matter how much voltage you pump into it, it will not budge.
    As far as I know, tCL, tRCD, tRP and possibly tRFC can (or can not) see voltage scaling.
  • Similarly, if a timing scales with voltage that means you can increase the voltage to run the same timing at a higher frequency.

    • You can see that tCL scales linearly with voltage on 8Gb CJR.
  • Some older Micron ICs (before Rev. E), are known to scale negatively with voltage. That is, they become unstable at the same frequency and timings just by increasing the voltage (usually above 1.35v).
  • Here are a table of common ICs and if the timing scales with voltage:
    [td]
    IC
    tCL
    tRCD
    tRP
    8Gb AFR
    Y
    N
    N
    8Gb CJR
    Y
    Y (?)
    N
    8Gb Rev. E
    Y
    N
    Y
    8Gb B-die
    Y
    Y
    Y

    • The timings that don't scale with voltage usually need to be increased as you increase frequency.

Expected Max Frequency
  • Below are the expected max frequency for some of the common ICs:
    [td]
    IC
    Expected Max Frequency (MHz)
    8Gb AFR
    3600
    8Gb CJR
    40001
    8Gb Rev. E
    4000+
    8Gb B-die
    4000+

    • 1CJR is a bit inconsistent in my testing. I've tested 3 RipJaws V 3600 CL19 8GB sticks. One of them was stuck at 3600MHz, another at 3800MHz but the last could do 4000MHz, all at CL16 with 1.45v.
    • Don't expect lower binned ICs to overclock nearly as well as higher binned ICs. This is especially true for B-die.

Binning
  • Binning is basically separating components based on their frequency.
    Manufacturers would separate ICs into different containers/bins depending on their frequency. Hence the term binning.
  • B-die binned for 2400 15-15-15 is significantly worse than good B-die binned for 3200 14-14-14 or even 3000 14-14-14. Don't expect it to have the same voltage scaling characteristics as good B-die.
  • To figure out which frequency and timings are a better (tighter) bin within the same IC at the same voltage, find out which timing doesn't scale from voltage.
    Simply divide the frequency by that timing and the higher value is the tighter bin.
    • For example, Crucial Ballistix 3000 15-16-16 and 3200 16-18-18 both use Micron Rev. E ICs. Simply dividing the frequency by tCL gives us the same value (200), so does that mean they're the same bin?
      No.
      tRCD doesn't scale with voltage, which means it needs to be increased as you increase frequency.
      3000 / 16 = 187.5 but 3200 / 18 = 177.78.
      As you can see, 3000 15-16-16 is a tighter bin than 3200 16-18-18. This means that a kit rated for 3000 15-16-16 will probably be able to do 3200 16-18-18 but a kit rated for 3200 16-18-18 might not be able to do 3000 15-16-16.

Maximum Recommended Daily Voltage
  • JEDEC (p.174) specifies that the absolute maximum is 1.50v.
    Stresses greater than those listed under “Absolute Maximum Ratings” may cause permanent damage to the device. This is a stress rating only and functional operation of the device at these or any other conditions above those indicated in the operational sections of this specification is not implied. Exposure to absolute maximum rating conditions for extended periods may affect reliability.
  • That being said, I'd only recommend running 1.50v on B-die as it's known to have high voltage tolerance. At least for the common ICs (4/8Gb AFR, 8Gb CJR, 8Gb Rev. E, 4/8Gb MFR), the max recommended voltage is 1.45v. Some of the lesser known ICs like 8Gb C-die have been reported to scale negatively or even die above 1.20v, though YMMV.
Integrated Memory Controller (IMC)Intel - LGA1151
  • Intel's IMC is pretty strong, so it shouldn't be the bottleneck when overclocking.
    What would you expect from 14+++++?
  • There are 2 voltages you need to change if overclocking RAM: system agent (VCCSA) and IO (VCCIO).
    DO NOT leave these on auto, as they can pump dangerous levels of voltage into your IMC, potentially degrading or even killing it. Most of the time you can keep VCCSA and VCCIO the same, but sometimes too much can harm stability(credits: Silent_Scone). I wouldn't recommend going above 1.25v on each.
    Below are my suggested VCCSA and VCCIO for 2 single rank DIMMs:
    [td]
    Frequency (MHz)
    VCCSA/VCCIO (v)
    3000 - 3600
    1.10 - 1.15
    3600 - 4000
    1.15 - 1.20
    4000+
    1.20 - 1.25

    • With more DIMMs and/or dual rank DIMMs, you may need higher VCCSA and VCCIO than suggested.
  • tRCD and tRP are linked, meaning if you set tRCD 16 but tRP 17, both will run at the higher timing (17). This limitation is why many ICs don't do as well on Intel and why B-die is a good match for Intel.

    • On Asrock and EVGA UEFIs, they're combined into tRCDtRP. On ASUS UEFIs, tRP is hidden. On MSI and Gigabyte UEFIs, tRCD and tRP are visible but setting them to different values just sets both of them to the higher value.
  • Expected memory latency range: 40ns - 50ns.

AMD - AM4
  • Ryzen 1000 and 2000's IMC can be a bit finnicky when overclocking and can't hit as high frequencies as Intel can. Ryzen 3000's IMC is much better and is more or less on par with Intel.
  • SOC voltage is the voltage to the IMC and like with Intel, it's not recommended to leave it on auto. You typically want 1.0 - 1.1v as above 1.1v doesn't help much if at all.
    On Ryzen 2000 (possibly 1000 and 3000 as well), above 1.15v can negatively impact overclocking.
    There are clear differences in how the memory controller behaves on the different CPU specimens. The majority of the CPUs will do 3466MHz or higher at 1.050V SoC voltage, however the difference lies in how the different specimens react to the voltage. Some of the specimens seem scale with the increased SoC voltage, while the others simply refuse to scale at all or in some cases even illustrate negative scaling. All of the tested samples illustrated negative scaling (i.e. more errors or failures to train) when higher than 1.150V SoC was used. In all cases the maximum memory frequency was achieved at =< 1.100V SoC voltage.
    ~ The Stilt

    • On Ryzen 3000, there's also CLDO_VDDG (not to be confused with CLDO_VDDP), which is the voltage to the Infinity Fabric. I've read that SOC voltage should be 40mV above CLDO_VDDG, but other than that there's not much information about it.
  • Below are the expected frequency ranges for 2 single rank DIMMs, provided your motherboard and ICs are capable:
    [td]
    Ryzen
    Expected Frequency (MHz)
    1000
    3000 - 3600
    2000
    3400 - 38001
    3000
    3600 - 3800 (1:1 MCLK:FCLK)
    3800+ (2:1 MCLK:FCLK)

    • With more DIMMs and/or dual rank DIMMs, the expected frequency can be lower.
    • 13600+ is typically achieved on a 1 DIMM per channel (DPC)/2 DIMM slot motherboard and with a very good IMC.
    • 13400MHz - 3533MHz is what most, if not all, Ryzen 2000 IMCs should be able to hit.
      On the tested samples, the distribution of the maximum achievable memory frequency was following:
      3400MHz – 12.5% of the samples
      3466MHz – 25.0% of the samples
      3533MHz – 62.5% of the samples
      ~ The Stilt
    • 2 CCD Ryzen 3000 CPUs (3900X and 3950X) seem to prefer 4 single rank sticks over 2 dual rank sticks.
      For 2 CCD SKUs, 2 DPC SR configuration seems to be the way to go. Both the 3600 and 3700X did 1800MHz UCLK on 1 DPC DR config, but most likely due to the discrepancy of the two CCDs in 3900X, it barely does 1733MHz on those DIMMs. Meanwhile with 2 DPC SR config there is no issue in reaching 1866MHz FCLK/UCLK.
      ~ The Stilt

  • tRCD is split into tRCDRD (read) and tRCDWR (write). Usually, tRCDWR can go lower than tRCDRD, but I haven't noticed any performance improvements from lowering tRCDWR. It's best to keep them the same.
  • Geardown mode (GDM) is automatically enabled above 2666MHz, which forces even tCL, even tCWL and CR 1T. If you want to run odd tCL, disable GDM. If you're unstable try running CR 2T, but that may negate the performance gain from dropping tCL.

    • For example, if you try to run 3000 CL15 with GDM enabled, CL will be rounded up to 16.
    • In terms of performance: GDM disabled CR 1T > GDM enabled CR 1T > GDM disabled CR 2T.
  • On single CCD Ryzen 3000 CPUs (CPUs below 3900X), write bandwidth is halved.
    In memory bandwidth, we see something odd, the write speed of AMD's 3700X, and that's because of the CDD to IOD connection, where the writes are 16B/cycle on the 3700X, but it's double that on the 3900X. AMD said this let them conserve power, which accounts for part of the lower TDP AMD aimed for. AMD says applications rarely do pure writes, but it did hurt the 3700X's performance in one of our benchmarks on the next page.
    ~ TweakTown
  • Expected memory latency range:
    [td]
    Ryzen
    Latency (ns)
    1000
    65 - 75
    2000
    60 - 70
    3000
    65 - 75 (1:1 MCLK:FCLK)
    75+ (2:1 MCLK:FCLK)
  • On Ryzen 3000, high enough FCLK can overcome the penalties from desynchronising MCLK and FCLK, provided that you can lock your UCLK to MCLK.


Overclocking
  • Disclaimer: The silicon lottery will affect your overclocking potential so there may be some deviation from my suggestions.
Finding the Maximum Frequency
  • On Intel, start off with 1.15v VCCSA and VCCIO. On AMD, start off with 1.10v SOC.
  • Set DRAM voltage to 1.40v. If you're using Micron/SpecTek ICs, exluding Rev. E, set 1.35v.
  • Set primary timings to 16-20-20-40 (tCL-tRCD-tRP-tRAS).
  • Increase the DRAM frequency until it doesn't boot into Windows any more. Keep in mind the expectations detailed above.
    • If you're on Intel, a quick way of knowing if you're unstable is to examine the RTLs and IOLs. Each group of RTLs and IOLs correspond to a channel. Within each group, there are 2 values which correspond to each DIMM.
      Asrock Timing Configurator
      As I have my sticks installed in channel A slot 2 and channel B slot 2, I need to look at D1 within each group of RTLs and IOLs.
      RTLs should be no more than 2 apart and IOLs should be no more than 1 apart.
      In my case, RTLs are 53 and 55 which are exactly 2 apart and IOLs are both 7. Note that having RTLs and IOLs within those ranges doesn't mean you're stable.
    • If you're on Ryzen 3000, make sure that the Infinity Fabric frequency (FCLK) is set to half your effective DRAM frequency.
  • Run a memory tester of your choice.
    • Windows will use ~2000MB so make sure to account for that when entering the amount of RAM to test. I have 16GB of RAM and usually test 14000MB.
    • Minimum recommended coverage:
      • MemTestHelper (HCI MemTest): 200% per thread.
      • Karhu RAMTest: 5000%.
        • In the advanced tab, make sure CPU cache is set to enabled. This will speed up testing by ~20%.


  • If you crash/freeze/BSOD or get an error, drop the DRAM frequency by a notch and test again.
  • Save your overclock profile in your UEFI.
  • From this point on you can either: try to go for a higher frequency or work on tightening the timings.
    • Keep in mind the expectations detailed above. If you're at the limit of your ICs and/or IMC it's best just to tighten the timings.

Trying Higher Frequencies
  • This section is applicable if you're not at the limit of your motherboard, ICs and IMC.
    This section is not for those who are having trouble stabilising frequencies within the expected range.
  • Intel:

    • Increase VCCSA and VCCIO to 1.25v.
    • Set command rate (CR) to 2T if it isn't already.
    • Set tCCDL to 8. Asus UEFIs don't expose this timing.
    Ryzen 3000:
    • Desynchronising MCLK and FCLK can incur a massive latency penalty, so you're better off tightening timings to keep your MCLK:FCLK 1:1.
    • Otherwise, set FCLK to whatever is stable (1600MHz if you're unsure).
  • Loosen primary timings to 18-22-22-42.
  • Increase DRAM voltage to 1.45v.
  • Follow steps 4-7 from Finding the Maximum Frequency.
  • Proceed to Tightening Timings.

Tightening Timings
  • Make sure to run a memory test and benchmark after each change to ensure performance is improving.
    • I would recommend to benchmark 3 to 5 times and average the results, as memory benchmarks can have a bit of variance.
    • Thereotical maximum bandwidth (MB/s) = ddr_freq * num_channels * 64 / 8.
      [td]
      Frequency (MHz)
      Max Dual Channel Bandwidth (MB/s)
      3000
      48000
      3200
      51200
      3400
      54440
      3466
      55456
      3600
      57600
      3733
      59728
      3800
      60800
      4000
      64000

      • Your read and write bandwidth should be 90% - 95% of the theoretical maximum bandwidth.
        • On single CCD Ryzen 3000 CPUs, write bandwidth should be 90% - 95% of half of the theoretical maximum bandwidth.
          It is possible to hit half of the theoretical maximum write bandwidth. See here.



  • AMD:

    • Try disabling GDM and setting CR to 1T. If that doesn't work, leave GDM enabled.
    Intel:
    • Try setting CR to 1T. If that doesn't work, leave CR on 2T.
  • I would recommend to tighten some of the secondary timings first, as they can speed up memory testing.
    My suggestions:
    [td]
    TimingSafeTightExtreme
    tRRDS tRRDL tFAW6 6 244 6 164 4 16
    tWR161210

    • Minimum tFAW can be is tRRDS * 4.
    • You don't have to run all of the timings at one preset. You might only be able to run tRRDS tRRDL tFAW at the tight preset, but you may be able to run tWR at the extreme preset.
  • Next are the primary timings (tCL, tRCD, tRP).

    • Start with tCL and drop that by 1 until you get instability.
    • Do the same with tRCD and tRP.
    • After the above timings are as tight as they can go, set tRAS = tCL + tRCD(RD) + 2 and tRC = tRP + tRAS.
      • Setting tRAS lower than this can incur a performance penalty.
      • tRC is only available on AMD and some Intel UEFIs.

  • Next is tRFC. Default for 8Gb ICs is 350ns (note the units).

    • To convert to ns: 2000 * timing / ddr_freq.
      For example, tRFC 250 at 3200MHz is 2000 * 250 / 3200 = 156.25ns.
    • To convert from ns: ns * ddr_freq / 2000.
      For example, 180ns at 3600MHz is 180 * 3600 / 2000 = 324.
    • Below are the typical tRFC in ns for the common ICs:
      [td]
      IC
      tRFC (ns)
      8Gb AFR
      260 - 280
      8Gb CJR
      260 - 280
      8Gb Rev. E
      300 - 350
      8Gb B-die
      160 - 180

  • Here are my suggestions for the rest of the secondaries:
    [td]
    Timing
    Safe
    Tight
    Extreme
    tWTRS tWTRL
    4 12
    4 8
    -
    tRTP
    12
    10
    8
    tCWL
    tCL
    tCL - 1
    tCL - 2

    • On Intel, tWTRS/L should be left on auto and controlled with tWRRD_dg/sg respectively. Dropping tWRRD_dg by 1 will drop tWTRS by 1. Likewise with tWRRD_sg. Once they're as low as you can go, manually set tWTRS/L.
  • Now for the tertiaries:

    • If you're on AMD, refer to this post.
      My suggestion:
      [td]
      TimingSafeTightExtreme
      tRDRDSCL tWRWRSCL4 43 32 2
    • If you're on Intel, tune the tertiaries one group at a time.
      My suggestions:
      [td]
      TimingSafeTightExtreme
      tRDRD_sg/dg/dr/dd8/4/8/87/4/7/76/4/6/6
      tWRWR_sg/dg/dr/dd8/4/8/87/4/7/76/4/6/6

      • For tWRRD_sg/dg, see step 5.
      • For tRDWR_sg/dg/dr/dd, drop them all by 1 until you get instability. You can usually run them all the same e.g. 9/9/9/9.
      • Note that dr only affects dual rank sticks, so if you have single rank sticks you can ignore this timing.
        These are my timings on B-die, for reference.
      • tREFI is also a timing that can help with performance. Unlike all the other timings, higher is better for tREFI.
        It's typically not a good idea to increase tREFI too much as ambient temperature changes (e.g. winter to summer) can be enough to cause instability.

  • You can also increase DRAM voltage to drop timings even more. Keep in mind the voltage scaling characteristics of your ICs and the maximum recommended daily voltage.

Miscellaneous Tips
  • Usually a 200MHz increase in DRAM frequency negates the latency penalty of loosening tCL, tRCD and tRP by 1, but has the benefit of higher bandwidth.
    For example, 3000 15-17-17 has the same latency as 3200 16-18-18, but 3200 16-18-18 has higher bandwidth.
  • Secondary and tertiary timings (except for tRFC) don't really change much, if at all, across the frequency range. If you have stable secondary and tertiary timings at 3200MHz, you could probably run them at 3600MHz, even 4000MHz, provided your ICs, IMC and motherboard are capable.
Intel
  • Loosening tCCDL to 8 may help with stability, especially above 3600MHz.
  • Higher cache (aka uncore, ring) frequency can increase bandwidth and reduce latency.
  • Increase IOL offsets to reduce RTLs and IOLs. Make sure to run a memory test after.
    More info here.
  • If you have an Asus Maximus motherboard and you can't boot, you can try tweaking the skew control values.
    More info here.
AMD
  • Try playing around with ProcODT if you can't boot. On Ryzen 1000 and 2000, you should try values between 40Ω and 68.6Ω.
    On Ryzen 3000, 1usmus suggests 28Ω - 40Ω.
    This seems to line up with The Stilt's settings.
    Phy at AGESA defaults, except ProcODT of 40.0Ohm, which is an ASUS auto-rule for Optimem III.
  • Lower SOC voltage may help with stability.
  • On Ryzen 3000, higher CLDO_VDDP can help with stability above 3600MHz.
    Increasing cLDO_VDDP seems beneficial > 3600MHz MEMCLKs, as increasing it seems to improve the margins and hence help with potential training issues.
    ~ The Stilt

Useful Information

评分

1

查看全部评分

发表于 2019-8-14 21:58 | 显示全部楼层
头好晕。。。快来扶一扶哀家
发表于 2019-8-14 23:45 | 显示全部楼层
这个也太棒了吧
发表于 2019-8-15 15:40 | 显示全部楼层
看了一下,这段有点问题
In terms of performance: GDM disabled CR 1T > GDM enabled CR 1T > GDM disabled CR 2T.

应该是GDM disabled CR 1T > GDM disabled CR 2T > GDM enabled CR 1T  
 楼主| 发表于 2019-8-15 20:37 | 显示全部楼层
wesleyxy 发表于 2019-8-15 15:40
看了一下,这段有点问题
In terms of performance: GDM disabled CR 1T > GDM enabled CR 1T > GDM disable ...

Geardown mode (GDM) 对性能影响这么大吗?
发表于 2019-8-15 20:57 | 显示全部楼层
看不懂英语
发表于 2019-8-15 21:29 | 显示全部楼层
xiaxia686 发表于 2019-8-15 20:37
Geardown mode (GDM) 对性能影响这么大吗?

虽然影响没那么大,但也不小,GDM disabled 1T才是正1t,但FCLK 1866以后就开始比较难开正1t了。GDM enabled CR 1T约等于2t-2.5t之间的性能,没有正2t性能好
发表于 2019-8-15 23:30 来自手机 | 显示全部楼层
本帖最后由 547737657 于 2019-8-15 23:44 编辑

初期的MesTestHelper多开脚本效率不如噹4.0,不知现在如何

HCI Memtest 200%,Karhu RAMTest居然要求5000%,难度系数差几个数量级好么!
发表于 2019-8-15 23:30 来自手机 | 显示全部楼层
xiaxia686 发表于 2019-8-15 20:37
Geardown mode (GDM) 对性能影响这么大吗?

初代ZEN关闭GDM基本歇菜
发表于 2019-9-27 20:15 来自手机 | 显示全部楼层
好帖没人顶。
发表于 2019-9-27 20:53 | 显示全部楼层
文章很好,可惜没翻译。
发表于 2019-9-27 21:47 来自手机 | 显示全部楼层
psone 发表于 2019-9-27 20:53
文章很好,可惜没翻译。

我正在翻译,发现drunk大大推荐的那篇intel平台的超频指导有部分内容就是这里面的,的确是好文章
发表于 2019-9-28 09:05 来自手机 | 显示全部楼层
547737657 发表于 2019-8-15 23:30
初代ZEN关闭GDM基本歇菜

是吗。。难道我1200死活只能3200原因就在于关了gdm?
发表于 2019-9-30 10:00 | 显示全部楼层
有没有大佬能翻译一下啊。
发表于 2019-9-30 13:14 来自手机 | 显示全部楼层
未名 发表于 2019-9-28 09:05
是吗。。难道我1200死活只能3200原因就在于关了gdm?

关了GDM还能到3200算好的了
发表于 2019-9-30 16:02 | 显示全部楼层
ngqhc 发表于 2019-9-27 21:47
我正在翻译,发现drunk大大推荐的那篇intel平台的超频指导有部分内容就是这里面的,的确是好文章 ...

辛苦了
发表于 2019-9-30 16:51 | 显示全部楼层
真的很想学习一下但是看不懂,俺的幻光戟3200C14只能XMP使用也不知道为啥,稍微改一下到3400都看不到BIOS
发表于 2019-9-30 17:07 | 显示全部楼层
正在苦修英语,争取能看懂。
发表于 2019-9-30 20:26 | 显示全部楼层

争取在国庆假期翻译完咯
发表于 2019-9-30 20:37 | 显示全部楼层
ngqhc 发表于 2019-9-30 20:26
争取在国庆假期翻译完咯

辛苦了。。。。。。
发表于 2019-9-30 20:41 | 显示全部楼层
坐等大神翻译
发表于 2019-9-30 22:09 | 显示全部楼层
马克一下。。。
发表于 2019-10-7 17:28 | 显示全部楼层
[内存]内存超频翻译转载
https://www.chiphell.com/thread-2112162-1-1.html
(出处: Chiphell - 分享与交流用户体验)
已经翻译,感谢大佬的分享
您需要登录后才可以回帖 登录 | 加入我们

本版积分规则

小黑屋|手机版|Archiver|Chiphell ( 沪ICP备12027953号-5 )沪公网备310112100042806

GMT+8, 2020-7-11 19:16 , Processed in 0.015687 second(s), 20 queries , Gzip On, Redis On.

Powered by Discuz! X3.2

© 2007-2019 Chiphell.com All rights reserved.

快速回复 返回顶部 返回列表