CPU and overclock: i5 2500 @ 3.3 GHz (Not OC'd)
Graphics Card: GTX 760

Driver version: 347.52
dxlevel (default is 90): 81
Resolution: 1920x1080
Full-screen or windowed: Full
FPS configs enabled: r4ndom cfg with mat_antialias 4
Shadows enabled/disabled: enabled

2639 frames 16.705 seconds 157.98 fps ( 6.33 ms/f) 10.025 fps variability
2639 frames 16.745 seconds 157.60 fps ( 6.35 ms/f) 10.144 fps variability
2639 frames 16.735 seconds 157.48 fps ( 6.33 ms/f) 10.025 fps variability

Driver version: 347.52
dxlevel (default is 90): 81
Resolution: 1920x1080
Full-screen or windowed: Full
FPS configs enabled: http://pastebin.com/UdTH4uDr
Shadows enabled/disabled: enabled

2639 frames 15.836 seconds 166.64 fps ( 6.00 ms/f) 11.730 fps variability
2639 frames 15.833 seconds 166.68 fps ( 6.00 ms/f) 10.829 fps variability
2639 frames 15.870 seconds 166.29 fps ( 6.01 ms/f) 10.730 fps variability

This is by far the best result i've been getting with any graphical cfg
Using Rhapsody's cfg, comanglia cfg I've been barely getting 155 fps.

It's a heavy edited maxframes.

CPU and overclock: A8-5600k @ 4.0 GHz (Not OC'd)
Graphics Card: ASUS R9 270x Direct CU II

Driver version: 15.200.1062.1003
dxlevel (default is 90): 81
Resolution: 1152x864 (4:3)
Full-screen or windowed: Full
FPS configs enabled: slightly modified comanglias
Shadows enabled/disabled: disabled

additional notes:
AVG PC TuneUp fully optimized on turbo mode, all graphics settings lowest in-game (except filtering mode is Trilinear), only necessary background programs enabled during benchmark.

Anyone have this demo still? The OP link is down

benchmark1.dem

Thank you sir

2639 frames 16.435 seconds 160.57 fps ( 6.23 ms/f) 12.727 fps variability
2639 frames 16.313 seconds 161.77 fps ( 6.18 ms/f) 12.357 fps variability

CPU and overclock: i7 4790k @ 4.6Ghz
Graphics Card: 2x GTX 680s

Driver version: 355.82
dxlevel (default is 90): 90
Resolution: 1920x1080
Full-screen or windowed: Full
FPS configs enabled: Personal high graphics cfg
Shadows enabled/disabled: enabled

Additional notes: With second monitor hooked up

Edit: Interestingly enough, when I changed the -dxlevel to 95 as well as turned on HDR (was off previously for some reason) and world reflections (which don't apply for this particular demo) I got these results:

2639 frames 16.120 seconds 163.70 fps ( 6.11 ms/f) 13.546 fps variability
2639 frames 15.991 seconds 165.03 fps ( 6.06 ms/f) 11.919 fps variability

My frames went up slightly with higher settings, although it is probably due to dx 95 being better compatible with my hardware. For anyone with a high end computer that doesn't feel like they get the proper framerates should maybe try tweaking settings and dxlevels to higher values.

Another mirror for the demo.

So I did some tests.

infinity redid his tests on windows 8.1 so I could compare the G3258 with an i7-4790K at the same speed with 2 cores and HT disabled.

CPU and overclock: i7 4790k @ 4.4GHz
Graphics Card: 780 Ti

Driver version: 340.52
dxlevel (default is 90): 81
Resolution: 1920x1080
Config: http://pastebin.com/NNwtc9f7

The goal was to find out if and why the G3258 was significantly slower than an i5.
I had a hunch that the bigger L3 cache wouldn't do much and I was right. I was getting 1-2% less fps. At first I chalked it up to the other two monitors and running a text editor and the task manager during the benchmark but that might not have been the cause, more on that later.

After I had finished all the tests I noticed that infinity's cfg used mat_queue_mode 1. So I redid all the tests and used it as an opportunity to see how much fps multithreading actually yields.

Then I tested processor/core affinity and RAM speeds. I had used 1333MHz CL9 RAM as opposed to infinity's 1600MHz CL9. Of course I could've spent time manually matching the timings, but I thought it shouldn't make much of a difference. I tested with 2400MHz CL10 later to confirm that. I was wrong. RAM speed might actually explain why I got less fps than infinity with his G3258. Going from 1333MHz CL9 to 2400MHz CL10 increased the fps by
8% with 4 cores (didn't matter if 4 physical (i5) or 4 logical, 2 physical cores (i3)), by
11% on 2 logical / 2 physical (Pentium) and by
12% on 8 logical / 4 physical (i7).

Setting affinity only hurt the performance, although I have only tested it with all cores and Hyperthreading enabled and not all possible configurations. I highly doubt that confining TF2 to 3 or less cores would help though.

On a sidenote, the second batch of tests I ran with -threads 2, since it gave the best results. The difference is minor though.

Now for the interesting and very strange part of the results.
The bulk of the gains is single threaded.
Multithreading does benefit from more cores, but not much. It gets you 18%-32% (one number inflated to 33.54% due to a bad single threaded result). Logical cores help almost as much as physical cores.
Beyond 3 physical or 4 logical and 2 physical cores diminishing returns hit hard. Combined with the fact that physical cores give only a minor advantage over logical cores that means an i3 gets you 94% of the performance of an i5. Since i3s with the same clockrate are far cheaper you could in theory even make up for the difference with faster RAM.
As expected multithreading didn't help with 1 core / 1 thread. Surprisingly it didn't help with 2C/2T either even though it helped with 1C/2T. While the second physical core helped single threaded since multithreading didn't really help (+1-2%) a true dual core with two physical cores is only 5% faster than a single core with hyperthreading. I am rather upset about this. So the answer to question if and why the G3258 was significantly slower than an i5 is:
Yes, because multithreading doesn't seem to work with 2 physical cores. Why that is, I have absolutely no idea.

Here are the statistics: http://www64.zippyshare.com/v/u1InD0Vt/file.html
And the raw data: http://www64.zippyshare.com/v/1sg9PANj/file.html

Could anyone verify these results?

tl;dr
Only Source 2 can save us now.

It makes sense that the main worker thread is what's bounded by the single threaded performance of the cpu. Enabling multicore will offload some of that but the added threads don't have as much to do as the primary engine, so a hyperthreaded core has plenty of time to keep up. This also explains why over overclocking improves performance linearly.

Performance always increases linearly with clockrate for a deterministic algorithm.

Even if -threads 2 does nothing at all, it doesn't explain why single threaded performance increases with the number of cores.
If -threads 2 works it becomes even more mysterious. How can HT increase performance when there's 3 physical cores and only 2 threads?
Why does multithreading not increase performance on 2 cores but it does on every other possible configuration (except 1 core obviously). It works with 1 core + HT, why not with 2 cores? It works on 3 cores without HT aswell. Only on 2 cores without HT it doesn't.

You'll probably have to check what clock rate the CPU is actually running at throughout the test. Because of dynamic frequency scaling depending on how much load is on each core in addition to how temperature affects frequency merely counting cores and looking at its rated clock rate is not enough.

It's possible that one of the CPUs is just reaching a higher frequency because of one of the above reasons. Also keep in mind that every chip is different (due to part binning for example).

So basically what you're trying to tell me is that:
1. "Dynamic frequency scaling" increased the clockrate with 3 or 4 cores enabled, but not with only 1 or 2 cores enabled.
2. An i7-4790K at stock voltage, 4.4GHz and <30% load is thermal throttling. I doubt that would happen even with the Intel stock cooler and I'm not exactly running stock cooling.
Even if I hadn't monitored the clockrate I'd have a hard time believing that.

Of course I could redo the tests with Turbo, C1E, C3, C6/7 and EIST all disabled and monitor all cores seperately instead of just the average via task manager, but for some reason I doubt it would change anything.

Sorry for the double post, but I think this is important.

nohats might explain why he got higher fps than me, but more importantly:
mat_queue_mode 2 still barely does anything.

@Setsul, did you see if the game prefers higher frequency on the rams or tighter timings? Just wondering really

No, not yet. I didn't want to put in even more work before the previous results were verified. Also manually setting RAM timings is a pain in the ass.

It does say "TODO: speed vs timings" in the results file for that reason.

I might just do some quick testing tomorrow, along the lines of 2200 MHz CL11, 2000 CL10 and 1866 CL9

This probably has been done but I though I would compare win10 to win8.1 While I'm at this.
Proper cfg this time.

Windows 10

2639 frames 14.492 seconds 182.10 fps ( 5.49 ms/f) 21.980 fps variability
2639 frames 14.299 seconds 184.56 fps ( 5.42 ms/f) 16.347 fps variability
2639 frames 14.099 seconds 187.18 fps ( 5.34 ms/f) 16.515 fps variability

Windows 8.1

2639 frames 14.889 seconds 177.25 fps ( 5.64 ms/f) 20.688 fps variability
2639 frames 14.429 seconds 182.89 fps ( 5.47 ms/f) 15.897 fps variability
2639 frames 14.669 seconds 179.90 fps ( 5.56 ms/f) 15.580 fps variability

Just a quick question, for some reason the demo doesn't work right for me anymore? About half way through it changes to the POV of an engineer until he dies and then the camera just stays where he was.

CPU and overclock: i7 2700k @ 4.2Ghz
Graphics Card: GTX 660ti
Driver version: 355.82
dxlevel: 90
Resolution: 1920x1080
Full-screen or windowed: Windowed Borderless
Config: Personal config emphasizing fps

Extras: Have 2 monitors in total hooked up to the same graphics card
Running Windows 7.

2639 frames 13.608 seconds 193.92 fps ( 5.16 ms/f) 14.320 fps variability
2639 frames 13.599 seconds 194.06 fps ( 5.15 ms/f) 13.640 fps variability
2639 frames 13.610 seconds 193.90 fps ( 5.16 ms/f) 14.000 fps variability

I updated the OP, but I feel like this demo is getting fairly long in the tooth and could use a replacement and a new thread to accompany it in the near future

Yeah, it does need an update. Also, do you know why post #202 happens?

In case anyone else was curious about performance in Windows 10 (I decided to check it out after infinity's post), I just upgraded and ran the demo again.

Windows 8.1:
2639 frames 16.120 seconds 163.70 fps ( 6.11 ms/f) 13.546 fps variability
2639 frames 15.991 seconds 165.03 fps ( 6.06 ms/f) 11.919 fps variability

Windows 10:
2639 frames 15.981 seconds 165.14 fps ( 6.06 ms/f) 12.658 fps variability
2639 frames 16.009 seconds 164.85 fps ( 6.07 ms/f) 12.931 fps variability

Negligible difference, but at least the frames didn't go down.

windows 10 is just a visual revamp of windows 8. No core elements have changed.

This is just the first link I found on the matter but based on the benchmarks here: http://www.dualshockers.com/2015/07/30/windows-10-vs-windows-8-1-gaming-benchmark-comparison-which-os-grants-better-performance/

It implies more than a visual revamp. Boot times and OS size have also decreased so at the very least there's less bullshit getting in the way. Windows 10 is still free until July 2016 if anyone else wanted to check it out.

yo can people do this again and see if they lost any fps? mine is down about 35 from last week

I just tested with a Skylake i5 6600k @ 4.5 GHz and got 160 fps avg with 6 fps variability @ 1080p

Comanglia CFG, dxlevel 81

This with the integrated GPU. Just thought you guys would fancy knowing how the latest generation of cpu performs.

Post discrete GFX results as well please.

Meh currently I only have a r7 250x, but I'll do it anyway :)

Update. with r7 250x, I got around 230 fps with variability of more or less 11 fps.

Bonus pic of testing rig : https://dl.dropboxusercontent.com/u/8661606/Not%C3%ADcias/IMG_20151024_190412.jpg