Comments by "Mikko Rantalainen" (@MikkoRantalainen) on "Linus Tech Tips" channel.

  1. I think they should bring the "M" suffix back when the mobile part isn't just undervolted and underclocked identical part. Calling this RTX 3080M and having performance between RTX 3070 and RTX 3080 would have been great. Calling this RTX 3080 while having totally different hardware than RTX 3080 is just lying. If they had RTX 3080M and RTX 3080 it would be obvious to everybody that those are NOT the same hardware.
    227
  2.  @Valcore123  My point is that if it is not the same part do not call it using identical name. If Nvidia had RTX 3080 and RTX 3080M, I wouldn't complain about the naming. However, when they currently have RTX 3080 and RTX 3080 and the latter is actually closer to RTX 3070 I do have a problem. And note that they used to have the "M" suffix for the mobile parts so this misdirection must be intentional!
    9
  3. The sync frequency to have sample accurate sync is only around 40 kHz so using that to syncronize clocks would get you to 25 µs accuracy. That's better than default NTP implementation but may not be accurate enough compared to what you can do with custom hardware.
    7
  4. I would love to have one but I currently have PS4 Pro and upgrading to PS5 + PSVR2 would be pretty expensive. Hopefully the prices will drop.
    4
  5. At least in theory the new Windows driver model could use IOMMU hardware which is often not available on Intel CPUs because Intel considered that feature only as server technology and only Xeon CPUs supported it historically. IOMMU supports preventing the hardware drivers overwriting memory and registers the driver hasn't requested from the OS beforehand. As such, it should prevent any random crashes because of driver overwriting wrong memory or registers. However, if the driver e.g. hangs the GPU, there's nothing the OS can do that. AMD supports IOMMU with all 64 bit CPUs. Just like ECC memory.
    4
  6. 12:30 I hope you guys are submitting test results to OpenBenchmarking.
    3
  7. I found this old video only today. Is the algorithm they use publicly known by now? The low latency alone is jaw dropping for processing of any kind, let alone this good.
    2
  8.  @Valcore123  It's not only about clocks and wattage. See https://en.wikipedia.org/wiki/GeForce_30_series for details. For example, "RTX 3080" on laptop has 6144 cores and "RTX 3080" on desktop has 8704 cores. On the other hand, "RTX 3070" on desktop has 6144 cores!
    2
  9. Poor Anthony had to use iDevices instead of Linux. Unsurprisingly he's upset. Your average Linux device is far from polished out of box but you can usually tweak it to perfection if you know tech well enough. That said, the stronghold of Apple has always been "not so much features but the features you get are really polished and everything works as promised out of box, even if you mix multiple Apple devices" – not anymore.
    2
  10. The hardware modifications to run those cameras in sync was really clever.
    1
  11. It almost seems like Intel has hired all the GPU drivers programmers that failed to make ATI GPUs to ever work. GPUs rely on drivers so much that it will be hard to create a working drivers for totally new hardware. Also, using the Intel GPU used to be the budget option. How this thing costs as much as RTX GPUs and is slower and buggy?
    1
  12. Watching this video about "ancient" motherboard using a system with 6 years older system than this "ancient" motherboard.
    1
  13. Are there any other kind of motherboard manufacturers? In my experience, ASUS has at least fixed some BIOS problems on their motherboards and that's already way better than majority of motherboard manufacturers. Tell me a modern motherboard that doesn't raise more than 1 BIOS/ACPI/UEFI warning/error with Linux kernel boot and I'll be a happy customer for that motherboard. My current ASUS board has about 14 "ACPI BIOS Error (bug): ..." or "ACPI Error: ..." problems on boot alone. The Linux kernel has automated workarounds for those but some of those workarounds are literally "lets not touch that feature at all" instead of "let's make it work by ourselves".
    1
  14. If those cats have any claws, the surface of the TV screen is going to suck after a while.
    1
  15. You need to create a new video with 400 Gbps QSFP-DD ports! If you run those in LACP/802.3ad you can have 800 Gbps link for your system. If you actually can find a card that has a QSFP-DD port for your PC.
    1
  16. When somebody is able to create OLED monitor than can display static image without fear of permanent damage I'm all in. Until that, IPS is the best long term solution. Of course, if you're rich enough, you just get a new OLED monitor every year or more often as the image detoriates and don't mind the costs.
    1
  17.  @Time.TraveI.0  VA display have acceptable color and great black level. The gamma shift on vertical movement is may be a problem depending on your usage pattern but the real killer is very bad pixel response time. VA pixel response from pretty dark to medium dark often takes about 30-50 ms whereas IPS and TN panels require 10 ms for bad panels and maybe 4-5 ms for good panels. The "1 ms pixel response" is just marketing lie on anything but OLED panels. If you can render more than 20 fps (as in your fingers and toes) you'll be limited by your panel's pixel response time on VA panels already!
    1
  18. ​ @Time.TraveI.0  See the RTINGS test results for those monitors instead of trusting the marketing material. The pixel response table shows about half the table in red for really bad overshoot on max refresh rate. The 60 Hz "Standard" (minimum) overdrive setting results in not-too-bad overshoot (still over 10%) and total response average time is 13-15 ms and worst case pixel response is way over 20 ms. And there's no overdrive setting that wouldn't result in heavy rendering artefacts; the scale for overdrive basically goes from mininum "too much" to "even more" and "extreme". So those have all the typical VA panel problems but Samsung tries to hide slow pixel response time issues with badly implemented overdrive. With a superbly implemented overdrive it might be possible to have an acceptable VA panel but Odyssey G9 (neither model) doesn't seem to be such a display. Note the worst pixel response time of 20 ms. That basically says that the max fps this display can render nearly correctly is 1000/20 or 50 fps. And even there you'll see heavy flickering because of the heavy overshoot. Compare that to LG 27GN800-B which has average pixel response time of 7.4 ms and worst case response time of 10.3 with ZERO overshoot (that is, overdrive totally disabled which is the setting that I use because I hate seeing those overshoot artefacts). Note that this display can correctly render all frames at 1000/10.3 or 97 fps and in most cases 1000/7.4 or 135 fps.
    1
  19. 15:00 Wow! Even that dynamic mic cannot hide the huge echo chamber you had. It nicely shows how much having enough acoustic panels will help. I guess you should have head mounted electret microphone very near your mouth if you have this problematic room. Think about how much better the sound in this video is than the "pro streamer" clips using the dynamic microphone! I think they used shotgut condenser mic for the actual video production so that's probably what you should use if you cannot improve acoustics.
    1
  20. Great intro!
    1
  21. Running three or more identical nodes on consumer hardware connected with GlusterFS is going to provide you with all the redudancy you need. No need for special hardware.
    1
  22. Trying to glue big metal parts with hot glue is not going to work because the glue is going to get too cold before you get the pieces together. It also seems that the white part probably has some contaminants (grease?) on it which prevents glue from sticking into it.
    1
  23.  @Valcore123  Misleading advertising is illegal at least here in Finland. Advertising may be required even for a solid product because there's a difference between making a new product known vs lying about the features of new product. The former is obviously okay, the latter depends on jurisdiction. And latter should probably be considered morally wrong everywhere.
    1
  24.  @kennythawsh  Mobile "RTX 3080" is practically same as undervolted "RTX 3070" on desktop. Calling mobile "RTX 3080" as "RTX 3080M" or "RTX 3070" would have been honest. I think the M suffix would have been best because it's not exactly identical to desktop 3070 either but much closer to it than desktop 3080 part.
    1
  25.  @jimpiunti  The problem with labeling different hardware identically when they perform "at the same level" is that you can play with the benchmarks too much. I guess you can find some benchmark where 2080 and 3080 ti get the same results (e.g. actually limited by CPU). Those GPUs are not at the same level, though.
    1
  26.  @TabalugaDragon  Yeah, that's technically true if you consider undervolting literally. However, in real world you have to reduce clocks to be able to undervolt any meaningful amount because the chip cannot work equally fast with lower voltages (this is caused by transistor switching time being dependant on the voltage - the lower the voltage, the slower the transistor). The power limits force the chip to reduce both clocks and voltage because the voltage cannot be reduced alone without making the chip unstable. However, the undervolting is the only way to reduce power use unless you can magically reduce current somehow without adjusting voltage. (For most circuits, the resistance over the circuit is constant and power usage is voltage squared divided by resistance.)
    1
  27.  @TabalugaDragon  Yeah, I agree that many chips can be undervolted without lowering the clock. However, you should always consider yourself lucky when that succeeds. Manufacturers are trying to bin their chips so that better quality chips are sold as more expensive parts. It's not called silicon lottery for no reason.
    1
  28.  @TabalugaDragon  Manufacturers do not put those higher voltages for fun. They know that using too high voltage increases power usage and noise needed cool things down. However, they only have limited amount of bins to put the chips and maybe you don't run the software that's the most demanding for the chip so it can deal with slightly less voltage than it could handle in the worst case for that chip. The problem with silicon chips is that you cannot run them borderline the working voltage and back off once you start seeing problems. Once the chip miscalculates because of having too little voltage, it may hang and the user has to hit reset button for the whole rig. If you as a user are willing to optimize the voltage for your specific workload and chips, you can often find that you can undervolt parts quite a bit. If all of CPU, GPU and RAM actually needed spec voltages to run at the spec speed, you should consider yourself really unlucky.
    1
  29.  @TabalugaDragon  If you undervolt your GPU, sure, usually nothing bad happens if you have already saved all changes before the hang. However, if you use your computer as a general purpose machine, undervolting your CPU to the limit may end up trashing your filesystem. Hopefully you had up to date backups. For a pure gaming rig where you don't have any data, sure, undervolting is the way to go.
    1
  30.  @TabalugaDragon  Yes, it totally depends on your system workload. If the system doesn't need to update any data for the top level folders because of your background tasks, the changes of losing lots of files is really small. If you have something running in the background doing changes to the filesystem and the CPU gets unstable, all bets are off. If CPU makes even a single bit error while updating the tree data for filesystem structure, you can potentially lose huge amount of files at once.
    1
  31.  @TabalugaDragon  I wouldn't claim one in million but I agree that the probability is not that high. I would guess more like 0.1% per hard reboot. It really boils down to how much risk you're willing to take. No hardware is perfectly stable no matter if you use spec clocks and voltage, though, so it's always probability game.
    1
  32.  @TabalugaDragon  Yeah, for laptops you usually have to replace the motherboard in case of hardware failure. And unless you run something like Framework laptop, it's not usually cost effective to replace the motherboard only. And the fact that your system has been stable is great but it's a statistics with sample size of 1. Manufacturers select clocks and voltage so that they get approximately 1-3% warranty cases. If your hardware can cope with significantly lower voltage and/or higher clocks, consider yourself lucky.
    1
  33. ​ @TabalugaDragon  Okay, how big is the sample size for the statistics you think of? If you consider chips included in carefully selected review products, you'll end up with seriously different statistics compared to actual products purchased from random store. I'm saying that if you buy 1000 "identical" CPUs, you'll find out that you cannot seriously undervolt majority of those without issues. If you have a good source of trustworthy statistics with a big sample size (at least 1000 identical parts included in the statistics), please, share a link. I'd be happy to be pointed wrong.
    1
  34. If you get only 100 ms accuracy with NTP, it's time to replace the software. Any computer running NTP should be able to get time accuracy within 1 ms from true time. If you need more accurate time than 100 µs, then you probably want something better than NTP.
    1
  35. Great video but I think you guys should get better mics for noisy environments. Maybe something like DPA 4088 with a wind screen?
    1
  36. Jake should probably read about basic electronics like voltage dividers or something similar. I guess you could have used some adjustable resistors to calibrate the pedals instead of cutting pieces off. In addition, you could have probably calibrated / programmed the transmitter, too. Unless you have really crappy transmitter it should support adjusting the curve from the controller to the actual behavior. That would include reversing the direction so you wouldn't have needed to swap the pedals. I'm pretty sure that the original 10K pot would have been just fine, too. And if you want even lower latency for the head mounted display, go with full analog setup with rapidFIRE. It will have worse image quality and maximum transmit distance is slightly less but the latency is even better. I'd suggest to try contacting Joshua Bardwell if you need more info about sending video stream over video link with low latency.
    1
  37. The fiber optic HDMI is a good idea, RGB lighting for this HDMI cable is a stupid idea which only increases costs. Manufacturing HDMI ends which can take SFP-25G-SR cable would be a very good idea. That way you can change the fiber length without re-paying for both ends and when (not if!) the fiber breaks, you can easily change it without re-paying for both ends. That would also lack the RGB which could be considered as an extra bonus!
    1
  38. I wish Dell followed standards for motherboard layout and PSU. I put zero emphasis on the looks of the box but the whole point of a desktop computer is ability to upgrade and Dell definitely fails here because of proprieatry parts.
    1
  39. 14:50 when the following ad has worse sounding overall audio quality than your workshop...
    1
  40. I think you should have added one recording with actual high quality audio setup to compare the results. Other than that, funny video.
    1
  41. That metal printer setup looks sick! Can you tell what kind of dimensional accuracy are you getting (X, Y, Z) – I would expect that the accuracy is not that good with such a huge shrinkage during the process but maybe the software can compute insanely good correction?
    1
  42. Unfortunately, full kit of PiKVM seems to be sold out everywhere.
    1
  43. If they attached the side panel with strong magnets instead "aerospace" glue, it would be much better idea. Or I would perfer a couple of metric screws to be honest.
    1
  44. Honestly, it was a good decision to use a better mic and even that wasn't really good enough for that echo chamber they had.
    1
  45. If you have money to waste, cable management arms is the right solution if you want to pull servers without cables ever being an issue. (The arm has minimum of 3 hinges to make it rotate correctly behind the server and one end is connected to the server and one to the rack.)
    1
  46. If you have background tasks, you should get more RAM instead of getting a faster CPU. Disclaimer: I'm still running i5-3570K with Samsung SSD and 32 GB of fast RAM.
    1
  47. 11:20 Why on earth all power bricks have the pins that way? If they rotated the plug 90 degrees in the power brick, you could put those next to other power bricks much more easily.
    1
  48. 2:04 Maybe I've misunderstood something but I thought use of registered memory instead of non-registered memory causes extra latency for the registration operation. Isn't that like +1 CL for the registered part?
    1
  49. You should underline that fiber will permanently break if this cable ever gets too much stress.
    1
  50. In addition to redudant servers (that is any given server is redundant, not redundant part in a single server such as RAID or multiple PSUs), you should also have redundant network. Hopefully something that supports LACP over parallel connections so in normal case you get multiplied network bandwidth and in case of hardware failure, your network still works but it might lose 25-50% of it's total bandwidth. 8:10 Nice, you explained that you also have redundant networking. I would hate to see the list prices for that networking gear, though.
    1
  51. Why didn't you bolt the wheel column, pedals and the monitor to the moving platform, too? If you've ever driven a real car, you know that the distance from the seat to wheel or pedals doesn't change while you go over a bump.
    1
  52. Honestly, I'd have rather see that highly expensive keyboard as a keycap set with Cherry style stem mounts. That would make much more sense because then you would be paying for the craftmanship for the part that's actually hard to manufacture accurately. The fact that the whole set is carved out of a single block of wood to have continuous wood grain means that the whole set is going to trash if even a single keycap cracks during the manufacturing. Combined with highly expensive wood, that would make a keyset expensive, no matter what.
    1
  53. Couldn't you just put a huge heatsink like that on the normal side of the motherboard and use PCIe raiser cables for the extension cards? That would allow using fully standard parts and still have the same cooling performance.
    1
  54. 7:20 This is why you did get the milling machine, right?
    1