Jump to content

DrMikeNZ

Member
 View Profile  See their activity

  • Posts

    1,559

  • Joined

  • Last visited

 Content Type 

Everything posted by DrMikeNZ

  1. DrMikeNZ
    I had an old H100i (not the v2) lying around which was out of the box compatible with AM4. All of the Cool-it based Corsair AIO's mount to AM4 using the clip method. The Asetek based coolers (half of the Corsair line, NZXT Kraken, etc) which were released prior to AM4 need a bracket, which can typically be ordered from the AIO manufacturer which may cause a small delay.
  2. DrMikeNZ
    You do not need PCIe gen 3 x16 to run GPUs, SLI in x8 x8 is fine. Just stick with a X370 board and an R7 1700 or 1600 if you are just gaming.
  3. DrMikeNZ
    RAM timings can be very important with a wide range of workloads, however for most people the cost increase of low latency high frequency RAM is not worth it. If RAM isn't a limiting factor in the workload you won't notice any difference. The rated frequency and timings on RAM can be an indication of the dies used in the manufacturing, which can be important with stability on some systems (particularly Ryzen). I would typically just get the most cost effective RAM that is on the motherboards qualified vendor list (QVL) to ensure compatibility, and not worry too much about it.
  4. DrMikeNZ
    Only the 95W TDP CPU's (1600X, 1700X and 1800X) have the 20°C offset, most programs were updated to be aware of this within the first month of Ryzen's release. Yes, it should be the package temp, that is the sensor on the CPU. The "CPU" temperature is typically the temperature sensor on the motherboard at the socket. Ryzen master should be considered the reference measurement. The temperature sensors are not calibrated or certified, so no program can be considered accurate. You can set adjustments in AIDA64 to get its measurements to more closely align with the Ryzen master readings. Thermal throttling starts at around 85°C with temperature shutdown occuring at 95°C. You shouldn't have any issues at or below 80°C.
  5. DrMikeNZ
    Not really necessary for consumer line-interactive UPS's where the system will just be briefly run on the stepped sine wave power while it is powered down. Strongly recommended if you are running sensitive equipment or Online UPS's where the inverter will be continuously powering the equipment.
  6. DrMikeNZ
    If you are using the system for gaming, then memory clock speed could be important. Some games have been reported to have an infinity fabric bottleneck and benefit from faster RAM. I would recommend looking for 2666 MHz if you can afford it as a balance between cost and speed. No worries, to be fair I expected a larger hit with single channel particularly at the low clock speed, although Cinebench isn't the best test for memory or infinity fabric bottlenecks. When I get time over the next few weeks I want to try to run experiments with single vs dual channel in some workloads which do see improvements with memory/infinity fabric clock speed. Yes, although I still got a 0.5% RSD over 8 replicate runs with a 1.5% difference between min and max, which is high (relatively speaking) compared to the differences in mean results I got when comparing 2133 to 3333 MHz memory. I have occasionally seen outlier results from cinebench which were 10% or so low, hence my OCD in running replicate analyses with it. It doesn't take long to run it a few times compared to the time between changing configurations anyway.
  7. DrMikeNZ
    Can't see a video, only an ad for amazon for some reason. Although, in Cinebench I get statistically the same performance from 2133 to 3333 MHz, and about a 7% improvement from single to dual channel. Cinebench can give unreliable results so I run multiple times and take the averages to ensure the differences are meaningful.
  8. DrMikeNZ
    Try a PS/2 port and keyboard if available, they are much more reliable at POST.
  9. DrMikeNZ
    It calculates random e-peens based on a 1990's workload.
  10. DrMikeNZ
    It really depends on your workload. If you use AVX2 heavily, you will notice about a 40-50% decrease in performance. Elsewhere, you will typically only notice a 0-10% decrease in performance. Power consumption would be lower with Ryzen.
  11. DrMikeNZ
    It might be the voltage. You can try overclocking again although I recommend doing it slowly and incrementally testing as you go taking notes on performance at each step. If you see the performance drop again you can reset the CMOS, keep a close eye on your actual processor clock speed (as read by HWmonitor or AIDA64, not task manager). If the clock speed is dropping down then it could be the overclock is unstable and the system dropping to a low power state to remain stable. You might be able to resolve this with some trial and error by using lower voltages and adjusting the load line calibration etc. If the performance is dropping and the clock speed isn't then there might be something else going on although it would be hard to identify the root cause without more data. Good luck.
  12. DrMikeNZ
    When you increase voltage and cinebench drops, what is your clock speed? There are some issues with overclocks dropping to 1.5Ghz with raised voltages due to a BIOS bug.
  13. DrMikeNZ
    I expect there to be another issue causing the majority of the poor performance. I have just run some tests on my Ryzen system with single channel memory to determine how significant the impact might be, and while running single channel on Ryzen will result in less direct communication with memory to the CPU cores reducing performance somewhat, the reduced performance running single channel in my tests is not as low performance as you are observing in cinebench. Single Channel 2133 MHz, CPU 3.8GHz: 1545cb Single Channel 3200 MHz, CPU 3.8GHz: 1550cb Dual Channel 2133 MHz, CPU 3.8GHz: 1660cb Dual Channel 3200 MHz, CPU 3.8GHz: 1680cb If you clear the CMOS, reset back to stock, what cinebench score do you get? What clock speed does HWmonitor report the CPU running at under load?
  14. DrMikeNZ
    Yes, I have 2x8GB in my Ryzen system.
  15. DrMikeNZ
    RAM speed on its own shouldn't be an issue with cinebench. I got 1660cb at 3.8GHz with 2133MHz RAM, and 1680cb at 3.8GHz with 3200MHz RAM (dual channel) on my R7 1700. However, if there is only a single ram stick, then there could be issues. With the cost of RAM, I would typically only recommend 2666MHz as a "cost effective" option for Ryzen (as if there is any cost effective option for RAM) as 3000+MHz overclocks are not guaranteed with many of the memory controllers limiting it and most infinity fabric bottlenecks are alleviated by 2666Mhz. Is this a single stick or 2x 4GB sticks?
  16. DrMikeNZ
    I am not sure what videos you are looking at, although they are most likely using an all-in-one closed loop cooler. These are pre-filled at the factory and typically have the pump built into the CPU mounting block and fairly straight forward to use.
  17. DrMikeNZ
    At 6MB/s I doubt the drive is actually at 99% usage, it might be a misread, task manager can be pretty useless like that some times. Does Resource Monitor provide any additional information?
  18. DrMikeNZ
    Have you tried clearing the CMOS?
  19. DrMikeNZ
    If your memory is not being saturated resulting in application crashes and system hangs, then there isn't much to worry about. However, Windows does have a lot of issues with the way it handles memory usage and page files. Use the details tab and look at the commit size or use Resource Monitor.
  20. DrMikeNZ
    I have only had 40% of my AIO's fail in the first 11 months of use (excluding the ones that were DoA). There are better than even odds that you won't have issues. The warranty was a bit of a joke as it costs a fortune to send it back for RMA.
  21. DrMikeNZ
    If you are unable to achieve 3200MHz on that board it will likely be due to bad luck with the CPU's IMC more than anything else. While Asus haven't tested and validated that specific kit, they have tested other G.Skill 3200C16 kits which would use the same dies.
  22. DrMikeNZ
    It is not really that the VRM is currently at 87°C that concerns me, it is more the temperature trend that is an issue, I have only had this system for 5 months and can already observe an upward trend in VRM temperature (ambient room temperature decreased going from summer to winter). I should not be able to see a trend after such a short period of time. The high starting temperatures of the VRM is likely accelerating the decomposition of the (possibly faulty) thermal pads, and decreasing VRM efficiency. Scheduled replacement of thermal pads should not need to be something that is performed regularly. Although with higher initial temperatures on the VRM, low quality thermal pads insulating them, this almost appears as if the motherboard manufacturers are intentionally designing the boards to fail within 2-3 years or they are just woefully incompetent. I have no idea what the f&%! Asrock were thinking with that board, the only reason I would consider buying one would be to study the VRM failure rate on it.
  23. DrMikeNZ
    I expect a node shrink and a little reprieve on power draw to a point. After that we might see a few different forks in technology competing as CPU manufacturers experiment with exotic transistor materials to overcome clock speed hurdles. For an AIO, the Kraken X62 is pretty good, a TR4 bracket is included for asetek coolers (which the kraken x62 is) with the processor. For a custom kit, I am not sure if there are any kits with included TR4 bracket yet. Ouch, I think I might need another power point and breaker to run one of those.
  24. DrMikeNZ
    I am less concerned about the CPU and more about the motherboards. I ran my 3930K for 3.5 years at 80°C, and the thermal paste degraded a few times. The VRM on the motherboard eventually degraded to the point where the system power throttles unless underclocked 30%. I ran my 6900K for a few weeks at 100°C for some urgent rendering tasks when the CPU cooler failed, I wasn't too concerned as it hadn't reached thermal throttle. While I run my R7 1700 at around 60°C, I am a little concerned about the motherboard VRM, as it is at a constant 86-87°C, and know that that will cause issues down the line.
  25. DrMikeNZ
    If you are afraid that AIO cooler pumps are no good then don't get them. Excluding DoA, I have only had about 40% of my AIO coolers fail in the first 11 months so you are more likely than not to be fine. If you want to keep the CPU cool at 4.0GHz, you will likely need to water cool. Just set aside the cost of the system in a contingency fund to replace any dead components in the rare case that there is a leak. Many of the X399 boards are ATX and should fit in most cases, some are E-ATX (eg Asus Zenith) and might be a bit harder to fit in standard cases, although the Cosmos is a behemoth and shouldn't be an issue (it supports ATX and E-ATX). it wasn't long ago that 1000W+ power supplies were excessive and a bit of a joke with no real need for them in the consumer space and it was expected that power consumption was only going to get lower. The jokes on us, and the PSU manufacturers must have had a bloody amazing crystal ball.
×