#1085
Nope, CPU isn't worth upgrading.

Not sure how high you want the settings, so GPU either a 380 or 390.
PCPartPicker part list / Price breakdown by merchant

Storage: Samsung 850 EVO-Series 250GB 2.5" Solid State Drive ($129.00 @ PCCaseGear)
Video Card: Asus Radeon R9 380 2GB Video Card ($288.00 @ CPL Online)
Power Supply: SeaSonic 450W 80+ Gold Certified Semi-Modular ATX Power Supply ($119.00 @ Mwave Australia)
Total: $536.00
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2016-01-30 23:05 AEDT+1100

http://au.pcpartpicker.com/part/asus-video-card-strixr9390dc3oc8gd5gaming

Apart from doing it properly and encoding it you could always just add more SSDs.

This is just a mess.

Return the mobo.
Z170 is a waste of money if you won't overclock and no offence, you specifically should not overclock. I just can't see that going well.
So you're already wasting money just by choosing that. Then you choose a mobo with way more features than you'll ever need. Even more money wasted. But on top of that you choose an extremely overpriced mobo.
Now you want to buy a new CPU but can't really afford a comparable one, so you want to buy a CPU that's cheaper than the overpriced mobo you got.
In case you haven't noticed yet your RAM won't fit either.
This isn't going anywhere.
Return the mobo.

You can get almost the same mobo that you had before for 60$. http://pcpartpicker.com/part/asus-motherboard-p8h61mlecsmr20

#12
I have to post because #13 summoned me, but basically what #17 said.
At 3 cores (or 2 cores with Hyperthreading) you're looking at 90-99% (depending on the clockrate) of the performance you'd get if you added infinite cores.

Due to scheduling the cores probably won't boost to the max so in the end everyone gets 4 cores with Hyperthreading at 3.1GHz. Even if it uses more cores that doesn't change anything so you're still looking at 3.1GHz with no core "bottleneck". An i3-4170 or 6100 (both 3.7GHz) would actually be faster. 2x 2706$ = 5412$ or 7x 117$ = 819$. Yeah, it's a terrible idea for TF2.

But I don't think this is about practicality.

The black ones are PCIe x1 slots.
Test something in those, if it works in those but not in the x16 slot it's dead.

x1 to x16 riser ftw? The bandwidth choke would be real.

1+3: There are definitely 3 PCIe slots on the mobo.

4. Working with something connected that actually uses PCIe power connectors?
5. No change at all? Takes a while usually. It's important that the monitor is connected before you boot. Did you use DVI?

Is it a DZ68DB? I'm pretty sure there is no DBZ68DB.

Which program?

PCI or PCIe?

There's a number of things you can/should check:
-Check the x1 slots with an x1 card if you've got one and the x16 with something other than a GPU like #2 already suggested. Anything goes as long as it's PCIe.
-Check the PSU. Wouldn't be the first time they forgot to connect the cables. One missing jumper and the GPU doesn't see a full 8pin and won't power on.
-Boot with a monitor connected (preferably DVI/VGA) to the GPU not the iGPU. I know for sure that the 7970 (and therefore the 280X, since it's just a rebrand) does this: Without a monitor connected it won't power on. Fits your behaviour just like faulty PCIe power connectors (see above). PCIe slot in use, but device not active.

I'm pretty sure starting at the highest Voltage that won't immediately kill the CPU is not how you get a "mild overclock".

There is no definite "max frequency" and it's trial and error. I can't tell what it'll be on your CPU.
Temps depend on the test, some are more intense than others. As long as it never starts throttling (e.g. AIDA64 shows that), which starts in the high 90s (°C of course), you'll be fine.

Manual, always manual.

Start by going for something reasonable. See if you can get 4.4 stable.
Change the voltage in small increments, don't start at 1.5V, start with 1.3V and 25mV=0.025V increments.
Go up if it crashes, down if it doesn't. Test at least half an hour with whatever you want (Prime95, AIDA64, linpack/LinX, IBT, x264 stresstest, whatever floats your boat). Watch for thermal throttling, if it's throttling the test doesn't count. You'll either have to up the fan speed or use a lower voltage (which probably means a lower clockrate as well).
When the behaviour changes after a step, if you went up and it crashed every time and now it doesn't, or if you went down and it didn't and now it does, change direction and use smaller increments (10mV). Repeat once more if you want (5mV). Depending on which direction your were going on the last step use the last/first voltage that it wasn't crashing with. This is your preliminary voltage. No, you're not done yet.

That should take you a while. If you have to go above 1.4V for 4.4GHz you are probably doing something wrong or other voltages need tweaking.

Either way it's time to move to more specific tests to figure that out, but I'll post that in a few days when you actually need it.

#1081
What kind of usage are we talking about? For games it never has been the best. For rendering it was good 3 years ago. Now? Forget it.

#1083
You are aware that the CPU must fit the socket on the motherboard?

Yes, to elaborate:
Most older models already have LODs, the lvl 2 & 3 dispenser, which coincidentally also have the absolute highest polygon count (14/15k) aka most details, being the obvious exception. Have you ever noticed how your fps drop when you approach a huge sentry nest (3 or 4 engies)? It's not even the sentries firing, the dispenser are worse.
Now the problem with source engine is that these LODs have to be manually created. Now Valve has the choice: Add the missing LODs and release new weapons/miscs/hats/skins with LODs or instead spend all of that time on making new weapons/miscs/hats/skins without LODs, which means way more new w/m/h/s and therefore way more money. I think you know what Valve chose.
This is exaggerating a bit, Valve does add some LODs every now and then, but some stuff never gets any (dispenser, dead ringer, lots of hats), and some is still way worse than the stock options even with LODs, for example festive weapons. Highest/lowest polygon count for the stock minigun are 8,405 and 770. The festive one is at 13,719 and 1,707. While 1,707 is still far better than having to render all 13,719 the ratio (1,707/707 vs 13,719/8,405) is actually worse on lower detail. So all the new stuff even with LODs (without it's even more pronounced) has a bigger impact on lower settings. Exaggerated example: On high you might get 100 fps and lose 10% -> 90fps, on low you'd get 200 but lose 20% -> 160 fps.

What I'm trying to say is: We are fucked and it will only get worse.
Considering that so far after every "optimization update" I got lower fps in every benchmark a port to Source 2 seems like the only way we'll ever get more fps. I am 99% sure though that this will only happen after the "final" version has been completed, which will include Vulkan, which hasn't even been released yet. I always expected at least one game with the "full" engine to be release and ports coming after that. If I had to guess maybe 2017 for L4D3 if we're lucky, then depending on how much they expect to break either CS:GO or TF2 being ported, then the other one. The problem is that Valve doesn't really do "release dates". Vulkan pushes the engine back, add Valve time and it could be easily another year or two on top of that. The other Source 2 game which shall not be named is pretty much on backburner at this point according to rumours. And even if they made 2017 at that point there's really no point in calling it a port anymore, you might as well release it as TF3, if you bother at all. CS:GO being ported is far more likely.
But never give up. TF2 on Source 2 in 2020? Get Hype!

EDIT: I'm not sure why I wrote this, but there's no point in deleting it.

EDIT2: #8
Low model detail:
2639 frames 12.926 seconds 204.16 fps ( 4.90 ms/f) 17.867 fps variability
High model detail:
2639 frames 13.714 seconds 192.43 fps ( 5.20 ms/f) 14.727 fps variability

No and that's one of the reason why TF2 runs like dogshit.
It's also the reason why turning down model details gets you higher fps even though your GPU definitely isn't maxed out.

First of all there is no 6 hours oc. It's either 24/7 stable or not. Sure there's bench ocs that might last a few hours, but that's not what you should be going for. With a bench oc you can just rerun the benchmark until it finishes before it crashes. Rerunning an official or cast? Much more difficult.

What part of the guide did you get stuck on though?

Sandy Bridge is extremely easy to overclock. You shouldn't touch the BCLK anyway, no straps for now, ring bus/cache is synced to the core so you've only got one voltage and multiplier that matter. Later you can change the other voltages a bit for stability and power consumption/temps, but up until you get to that point where you can start tweaking the oc it's literally two settings.

#1079
390.

No.
You want to change which component does which part of the work without changing anything in the engine? Not possible.

And then there's another problem, even if they could, would they do it? No.

They could lessen the load simply by adding LODs. It's been almost 9 years and there's still no lvl2/3 dispenser LOD.

#1077
1. Haven't heard about that yet. Link(s)? On 1080p 4GB aren't necessary, with a 380/960 it'd just be waste of money. Turn up the settings high enough to even get close to 2GB and you're looking at <30 fps average. http://www.teamfortress.tv/post/514892/pc-build-thread

2. Depends on your headphones. Not sure how much research you've done regarding that. Although given that you want a mic as well the input pre amp (or rather the lack thereof) of onboard might not cut it. The audio setup is a whole different can of worms though.

3. Depends on the game and settings. In TF2 also on the CPU clockrate.