More information by another maintainer is at https://rentry.co/pcbg-more2
For Windows 10, you can download an installer to burn to a USB to install and add a product key at your leisure. Same for your Linux distro of choice, obviously.
To install Windows 7 on newer chipsets which Windows 7 no longer supports: https://pastebin.com/TUZvnmy1 Use this at your own risk as you have to manually download patches for security updates, and Win7 end-of-life comes Q1 2020.
Hackintosh is troublesome on Ryzen and Nvidia.
Also if doing KVM passthrough, an iGPU is useful, otherwise you need a second drive for Linux display as the display will turn off as it's displaying for the passed through display card.
Intel is your only choice for 6core+ CPUs with an iGPU.
Any X570 board will run a stock 3900X or 3950X CPU fine, but overclocking performance will vary.
Any X470 or B450 board will run a Ryzen 3000 6 or 8 core or lower fine, and many will OC them fine.
See: https://i.imgur.com/qm8xp5w.png for a motherboard VRM chart
Some software, pretty much exclusively games, sometimes has problems scheduling threads with user-mode scheduling. They can not be aware of Ryzen topology and schedule threads worse than Windows would itself.
Due to this, disabling SMT on a 12 core like the 3900X can drastically improve performance as much as 20%, and match the 9900k, in many games which poorly implement user-mode scheduling.
It is not recommended to disable SMT on a 6 or 8 core, but the 12 core has enough cores that 12 is plenty for any game and it has no drawback for pure gaming.
NVMe does not significantly increased boot time and application load times compared to a fast SATA SSD such as the MX500 or 860 Evo. They are not for gaming.
They can be useful if you have many VMs, so there is more read/write to split between them without them bottlenecking from that. It can also be useful if you're dealing with dozens or hundreds of Gb of new raw video files each week which you have to move around, organize, and work from. But your source much also have NVMe transfer speeds to truly take advantage of that.
Load times source: https://www.youtube.com/watch?v=ecCA0gx_eZk
Using Optane with StoreMi or another tiered storage solution can still see slightly noticeable improvements in responsiveness due to Optane having significantly higher iop/s and lower latency than NVMe. But this is mostly for an unlimited budget scenario.
A general guide to which SSD to choice is here: https://i.imgur.com/79MYtoE.png (as of July 2019)
In other words, a midrange NVMe like the 970 Evo is a complete waste for gaming and won't noticeably outperform a fast SATA SSD like the MX500 or 860 Evo.
As of mid-2019, SSDs are now cheap and HDDs are largely defunct outside of ultra budget <$500 builds and servers. You can now fit a 500GB to 1TB SSD in a reasonable $650 1080p@60 build.
The main storage configurations to consider, in range of budget, are:
HDD only for ultra budget secondary computers. Can add SSD later if possible. This is highly discouraged, as you can get a 500GB SSD for the cost of a decent 1TB HDD. HDDs are also worthwhile for cold storage ie home servers, servers, NAS.
1TB 3D-TLC for main SSD and the only drive in a budget to mid-range build. These can be found for $75-$110 or 90-115 euro from the low end SU635, to the higher end MX500/860Evo.
1TB-2TB Inland Premium, or Sabrent Rocket, or any of the slightly higher end models listed below for a mid-range to higher end build as the sole SSD.
256GB SX8200/EX950/Addlink S70 as main and/or scratch drive, with a secondary 2TB-4TB QLC drive (ie 660p, 860 EVO) for main storage for high-end
256GB SX8200/EX950/Addlink S70 as main and/or scratch drive, plus Optane and a 4TB QLC drive paired with StoreMi for ultra-high-end.
For workstation usage, it is recommended to have a second 256GB drive to have a dedicated scratch drive. If you make enough money from 8K video production with many terabytes or even petabytes of files, an even larger scratch drive can be worthwhile.
See the SSD choice guide link above for more considerations, though it is slightly outdated (ie missing the
QLC SSDs should not be used as a main drive, especially if 1TB or lower.
B350/B450 is the minimum chipset required to support fast memory on AM4 for AMD.
Z370 is the minimum chipset required to support fast memory on Coffeelake and the upcoming Cannonlake.
The formula to find the true latency of memory is:
2000 / Mhz * CL
You want a true latency of 10ns or lower, and generally want a speed of 2666MHz minimum and ideally 2933+.
ie 2000/3000*15 = 10ns
2000/3200*16 = 10ns
2000/3466*14 = 8.07ns (Samsung B-die overclock on The Stilt's timings)
Micron E-die modules will typically OC to at least 3700MHz@CL17 on *lake and Zen2 CPUs, if not 3800MHz+. Zen1/Zen+ has trouble running memory faster than 3466MHz.
The cost of Samsung B-die for Zen1/Zen2+ is not worth it. You get better performance running cheaper memory like Micron E-die on Zen2 instead.
Graphics cards continued#
Nvidia Pascal and earlier GPUs lose ~10% FPS when using proper 10bit HDR. Keep that in mind if you plan to get a new or upcoming HDR monitor. Source (Main source is in Germa)
AMD GPUs tend to be a very sub-optimal choice for console emulators which only support OpenGL, such as Cemu. Emulators which use Vulkan or which are older and much less intensive such as Dolphin are fine.
Avoid cheap MODELS ie MSI Armor (Mk2 is ok) or Ventus XS, Gigabyte G1/Wf, ASUS dual, and others which have small heatsinks and low quality fans
To expand on this, many cards are basically meant to die soon after their warranty expires. Heat can kill components on the card besides the GPU itself. So not only are poor quality coolers louder, they will typically have a worse longevity.
The difference in cards for most consumers come down to cooler quality, and you can somewhat tell the quality by looking at the pictures if you can't find good comparison reviews of performance vs noise & heat.
Though for more extreme overclockers, PCB(printed circuit board and the components soldered to it) quality also matters. I'll just go over the cooler quality here, as that is something you can learn to largely identify from pictures.
Those listed are some of the lowest quality models, but it can be hard to say what is average vs good vs great by pictures.
You can at least see what's bad. The MSI Armor is one which is really easy to see. Its cooler and heatsink are so thin that it's not even the full 2 slots.
The ASUS Dual is a more difficult one. Its shroud covers the heatsink, and it does fill 2 slots.
But if you look closely... you can see the heatsink is very thin
Also from the front, you can see the heatsink doesn't extend to cover behind a good deal of the fans. there's big gaps.
There's also only 2 heatpipes. One each goes behind each fan.
The Gigabyte G1/Windforce 1070, 1070ti, and 1080 is also difficult to see that it's bad at first glance, being that it's a full 2 slot and triple fan card. But it is if you've heard it in person. It mostly comes down to poor quality and small diameter fans which are not only noisy, but the pitch of their noise is good at penetrating the noise insulation which your case would usually provide. Another reason it's bad is that much of the heatsink is just a milled aluminum block, which requires those fans to spin faster to keep it cool.
A dual fan cooler like the EVGA ACX 3.0 or Sapphire's Vega 56 Pulse cooler are both far, far better than this triple fan one from Gigabyte due to these factors. They both use far more heatpipes and use them much more intelligently. The Sapphire Vega 56 Pulse is also a 2.5 slot card.
However, more metal is only better to a point. The Strix cooler is very very good. It both has a good deal of metal, heatpipes, and good quality fans.
There are some really overkill looking coolers out there for super high end cards that despite all their metal and large fans only perform a few % at most better than the Strix cooler. So there is a happy medium somewhere for the power output of any given card.
For <65 watt cards, a extruded metal aluminum heatsink with a copper core is fine. Don't get a 1060 with that, though. Or even a 1050Ti.
A extruded metal heatsink as you can see on https://www.msi.com/Graphics-card/GeForce-RTX-2060-VENTUS-XS-6G-OC is not okay for a 160W card like the 2060.
Heatsinks like those can be equally poor, or worse, than a "good" blower.
Nvidia GPUs with newer drivers now support Adaptive Sync over Displayport, but not Freesync over HDMI. Compatibility is reportedly spotty and input latency is increased by 10ms over an AMD GPU when using this.
When buying a VA panel, be especially mindful of the response time. Good ones can match a good IPS, or they can also be much worse.
Also be mindful that the response time listed is not always entirely accurate, and is for grey-to-grey pixel changes and not color-to-color. The real world response time is slower than the g-t-g would imply.
For monitor distance vs viewing size see: https://imgur.com/a/RTbKAxD
In order to have little-to-no microstutters and tearing without adaptive sync you must:
Run the game at 2x the framerate (ie 120fps on a 60hz monitor, or 288fps on a 144hz montir, though around +80% can be sufficient to not see many microstutters and eliminate tearing)
Enable fast sync (Nvidia) or enhanced sync (AMD)
Disable VSync and frame buffering, as these add 30-100ms of input latency depending on the game
That 60fps on a 60hz screen will be smooth without Vsync is a common misconception stemming from many people having frame buffering enabled in borderless window mode and not realizing it, and/or simply having bad eyes and not knowing what a smooth display looks like.
https://www.youtube.com/watch?v=xsXFUVYPIx4 explains the impact of Vsync and frame buffering on input latency, how they eliminate frame tearing and stuttering at the cost of latency, and how Freesync/Gsync also eliminate these but without a greatly increased latency penalty.