Reinstalled Windows 10 for testing, so might as well test different TF2 tweaks

Specs:
5800X PBO (4.85GHz all cores)
RX 580 8GB stock
4x8GB 3200MHz CL14
Windows 10 IoT Enterprise LTSC 2021
benchmark_test

As a reference, from a completely stock TF2 I scored 339FPS. 0 tweaks or custom files.

After testing, this was what I settled on:

4812 frames 8.643 seconds 556.74 fps ( 1.80 ms/f) 68.202 fps variability

This was with DX 8.1, CleanTF2+, 1024x768 fullscreen, mastercoms low with viewmodels and match hud off, lighthud.

Some interesting things I found:

• XMP on or off made quite literally 0 difference. 2133 JEDEC vs 3200 CL14 netted me 0.14FPS. Not sure why
  
• -dxlevel 100 reduced my performance by 6%
  
• Turning off both match HUD and viewmodels gave me the most FPS (other than using a config)

If I wanted to +200MHz to PBO and test CO again, disable chat, maybe overclock my GPU a bit, i'm sure I could get 580FPS+, but I'm not that desperate.

Not bad for Zen 3.

Finally able to oblige.
Third run of benchmark1 on mastercomfig medium preset w/ customizations (spoiler below), 1080p. Hardware: RTX 3070 Ti, 32 GB DDR4-3600 CL18

3700X (old CPU):

2639 frames 14.540 seconds 181.50 fps ( 5.51 ms/f) 14.491 fps variability

5800X3D (new CPU):

2639 frames 11.418 seconds 231.13 fps ( 4.33 ms/f) 17.377 fps variability

I run these mastercomfig settings because I'm a filthy pubber, but if you'd like a run with more comp-like settings, I can bench it for you (as long as it's still mastercomfig). Either way, for just a single CPU generation gap I'm pretty happy with the result, now I just have to check if my frames can still drop to 100s in pubs, but I'm not getting my hopes up...

EDIT: So far, so good! From dropping to 100-110 on medium preset, to dropping to 130 on medium high. That alone made the upgrade worth it for me, tbh

CPU: AMD Ryzen 7600X
Graphics Card: RTX 3060 Ti 8GB
Memory: 2x16GB DDR5 6000 CL36

OS: Windows 10 22H2
GPU driver version: 531.68
Resolution: 1920x1080 (full-screen)
FPS config: modified mastercomfig low
dxlevel: 81
Mods: nohats, modified explosions

Ryzen 5 3600 @Mastercoms medium-low
4812 frames 17.177 seconds 280.14 fps ( 3.57 ms/f) 52.686 fps variability
Ryzen 5 5600x @Mastercoms medium-low
4812 frames 12.537 seconds 383.82 fps ( 2.61 ms/f) 81.909 fps variability
Ryzen 5 5600X @Mastercoms high
4812 frames 14.710 seconds 327.13 fps ( 3.06 ms/f) 48.189 fps variability

CPU: AMD Ryzen 5600X
Graphics Card: RX 580
Memory: 2x16GB DDR4 @3200

OS: Windows 10
GPU driver version: 23.4.1
Resolution: 1920x1080 (full-screen)
dxlevel: 81
Mods:modified explosions

Benchmark_test.dem

4812 frames 15.532 seconds 309.80 fps ( 3.23 ms/f) 44.535 fps variability
4812 frames 15.189 seconds 316.82 fps ( 3.16 ms/f) 40.040 fps variability
4812 frames 15.201 seconds 316.56 fps ( 3.16 ms/f) 39.707 fps variability

Driver version: 22.11.2
dxlevel (default is 95): 81
Resolution: 1920x1080
Full-screen or windowed: Full
FPS configs enabled: J_4rb1's TF2 FPS config
Shadows enabled/disabled: disabled

GPU : AMD radeon rx5500xt oc 8000mb @ 1500-1600MHz
CPU : Intel core i5-10400f @ 4.1 - 4.2GHz
Memory : DDR4 @ 2133MHz
OS : Windows 10 pro

HUD: M0rehud gunblade

Mods : CleanTF2+ (no hats, no hands&feet, no weapon gibs, no gibs)

Launch options
-novid -reuse -noforcemaccel -noforcemspd -noforcemparms -nohltv -refresh 240 -full -w 1920 -h 1080 -nojoy -nosteamcontroller -particles 1 -noquicktime -precachefontchars -nostartupsound -softparticlesdefaultoff

Additional notes: host_thread_mode 1 on the benchmarks with 316+ fps

I haven't touched tf2 in years but it's still interesting to compare just for the sake of it.

Anyway, current laptop specs
RTX 3060 6GB 115W
i5 12500H
16GB DDR4 3200
Hardware @ stock settings, no overclocks etc, same config/settings as before, just win11 this time

benchmark1.dem
2639 frames 11.447 seconds 230.55 fps ( 4.34 ms/f) 23.853 fps variability

benchmark_test.dem
4812 frames 21.397 seconds 224.89 fps ( 4.45 ms/f) 39.755 fps variability

timedemo benchmark1

2639 frames 10.890 seconds 242.34 fps ( 4.13 ms/f) 21.129 fps variability
2639 frames 10.851 seconds 243.21 fps ( 4.11 ms/f) 19.558 fps variability
2639 frames 10.884 seconds 242.47 fps ( 4.12 ms/f) 19.167 fps variability

dxlevel (default is 95): 95
Resolution: 2560x1440
Full-screen or windowed: Full
FPS configs enabled: Mastercomfig High Preset
Shadows enabled/disabled: enabled

GPU : AMD radeon rx 7900 xtx
CPU : AMD ryzen 9 7950x3d @ 4.2GHz
Memory : 32Gb (16Gb x 2) DDR5 @ 6000MHz
OS : Windows 11 Home

AMD Driver version: 23.5.2

HUD: broeselhud blue-master

Launch options
-novid -nojoy -nosteamcontroller -nohltv -particles 1 -precachefontchars -noquicktime

Upgraded to ddr5 and did some tweaking.
Don't read too much into it being the RAM - I only got about 10-15 extra fps in each benchmark from ddr4->ddr5, the rest of the improvement is from optimizing the core configuration (disabling ecores and hyperthreading gives you more thermal headroom to crank the pcores).
Even with ecores enabled, manually changing the core affinity in hl2.exe gives me an enormous increase in fps, which tells me it's not worth having them on for purely tf2.
I'm on windows 10 though and I've heard the w11 scheduler is much better with ecores.

As a sidenote, benchmark_test doesn't seem to care much what I do to it anymore. The results vary wildly based on seemingly nothing. I suspect the death cam zoom in it is causing problems

old benchmarks
i5 13600KF 5.4GHz Pcores + 3.8GHz Ecores
DDR4 3600MHz C16 stock XMP
GTX1660Ti

dx8

new benchmarks
i5 13600KF (5.6GHz Pcores / 4.7GHz ring / Ecores off / SMT off) @~1.35V
DDR5 6600MHz 34-40-40 (XMP with some tightened subtimings)
GTX1660Ti

dx8

dx98

The RAM isn't running anywhere near optimal but I don't know how many more reboots to bios I have in me.
I had it running at 7400c34 stable and the fps didn't improve by much at all so I'd rather not worry about it

This thread made me realize I need a new PC. Some of these are basically triple the frames of my 4770k

Anyone got a download link for the benchmark demo that works? I just get captchas over and over.

https://cdn.discordapp.com/attachments/686311458237317142/773714064593518622/benchmark1.dem

Did benchmark1 stop working for anyone else after the latest update?

I'm getting this error message when I try to run it:

Host_EndGame: CL_ParseClassInfo_EndClasses: CreateDecoders failed.

Host_EndGame: CL_ParseClassInfo_EndClasses: CreateDecoders failed.

https://www.teamfortress.tv/post/1086297/cant-load-in-demos

until valve fix it the workaround is..

https://i.imgur.com/9x1ylpI.png

Test of Ryzen 7800x3d on Tf2

Resolution: 2560x1440
Config: Mastercomfig High Preset
Hud: Improved Default Hud

GPU: Nvidia RTX 3080 OC to 1865 MHz
CPU: AMD Ryzen 7 7800x3d OC to 5.05 GHz all core 5.3 GHz single core
[quick note: fclock is also set to 2100 MHz as opposed to its default setting of 2000MHz for more synchronous operation with the memory controller]
Memory: 32Gb (2x16) DDR5 @ 6200MHz cl28-36-36-36 (tightened secondary timings as well)

[This overclock was achieved by the use of the PBO2 tuner with a negative offset of 50. A manual clock speed override was also set to +250 MHz in order to achieve 5.3 GHz in single core as the maximum frequency from the factory for this chip is 5.05 GHz]

OS: Windows 10 Home

I've conducted these tests first using the provided demo file from the benchmarking forum, then 4 other demos I recorded myself to compare with my old system detailed below.

Ryzen 7 3700x PBO Enabled
RTX 3080 OC to 1865 MHz
Memory: 32Gb DDR4 (4x8gb) @ 3200 MHz cl16-18-18-36

All tests were run 3 times then averaged out to eliminate any margin of error

Benchmark1

2639 frames 9.358 seconds 282.01 fps ( 3.55 ms/f) 21.609 fps variability
2639 frames 9.347 seconds 282.33 fps (3.54ms/f) 21.123 fps variability
2639 frames 9.365 seconds 281.78 fps (3.55ms/f) 20.866 fps variability

Average

2639 frames 9.357 282.04 fps (3.55ms/f)  21.199 fps variability

-------------------------------------------------------------------------------------------------------------------------------------------------------------
[These are now my own tests using 3 different demos on 3 different maps, Badwater, Harvest, and Dustbowl. I have recorded and averaged the time the benchmark took to complete as well as the average fps during the benchmark]

Badwater

3700x: 42 seconds 180 fps
7800x3d: 24 seconds 316 fps

76 percent improvement

Harvest

3700x: 32 seconds 140 fps
7800x3d: 18 seconds 246 fps

76 percent improvement

Dustbowl

3700x: 101 seconds 149 fps
7800x3d: 56 seconds 266 fps

80 percent improvement

Overall: the Ryzen 7 7800x3d is 77 percent faster than the Ryzen 7 3700x in Team Fortress 2 according to my testing.

This confirms my assumption that Ryzen 7000x3d would be very well suited to tf2 due to its sizeable L2 cache and enormous L3 cache as well as its high clock speeds which help diminish the lackluster code optimization and therefore performance.

Anyone with a 7700x? Considering between it and the 7800x3d, and saw in a hardware unboxed cs:go(f) benchmark the 7700x was 20% faster.

you should get 7800x3d, it's better for sure

Windows 11
CPU: 7700X
GPU: 1070Ti
RAM: 32GB 5400MHz
Resolution 1440p
Config: Very similar config to mastercoms high preset
Dx9
Benchmark1.dem

Note that now to record on anything except pre_ji_demos you have to use offline mode on steam or disable internet or the game crashes on boot. Theres this addon that ficool2 made you can add so you don't have to do that though I don't know if that would bias testing in some way.
https://github.com/ficool2/custom_items_games

So, according to data available here, the 7800x3d is 16,62% faster than the 7700x, but according to prices on geizhals.eu, it's 25,17% more expensive. Still, thanks for your output fellas.

outside of tf2 its gonna be a very small diff and if all ur looking for is consistent 360hz experience a 7700 will do. imo not worth the price bump unless u got a lot of use out of the extra features. but if u wanna future proof more deffos spend the extra

Did some benchmarks comparings Linux vs Windows out of curiosity.

Setup:
CPU: Ryzen 5 5600x stock
GPU: 3060 Ti
RAM: 32GB 3600MHz
Resolution 1080p
Config: Mastercomfig Medium-low
Dx9

benchmark1 Win 11

benchmark1 Arch Linux

While the raw performance seems to be better on Linux, the actual gameplay is more stable on Windows 11 on my machine. Now and then the frames break into the 140-160s for a few seconds in actual gameplay on my linux installation.

new link: https://www.dropbox.com/s/efb16cqkfe0sv3j/benchmark1.dem?e=1&dl=1

lenovo laptop 14alc05
ryzen 5700u
16gb, 7.7gb shared, 512mb dedicated to i-gpu
driving 1440p external monitor via hdmi.
benched day 1 of 64bit

did a bunch of testing after joining skial harvest @ 22 players and with "low" avgs of 60-70 and shooting up to 170 when not looking at a lot of players (same as test #1). Seemingly can't alt-tab with -vulkan being used, not sure if igpu drivers are being weird or what.

Will say, very pleased with how this is turning out. I've slowly seen this laptop over the past... 2? 3? years become more and more viable for the random stuff i want to do occasionally. Went back to harvest, 15-20 players on the server lowest i saw with dx81, mastercom very low was like 80-90 on a mid.

I ran nerd shit for dx81, 90, and 95 progressively working up to the final 2 states i benched. like 2 dozen other tests if you want to check it out here: https://pastebin.com/b3d0bqfH

Used shounics benchmarks: https://drive.google.com/file/d/1xbuiBnZb301EXFyYuQP17KS21AWM_v5o/view

///dx95 (no config, no launch options, all low adv options 2560x1440p)
sbmrk----11795 frames 252.900 seconds 46.64 fps (21.44 ms/f) 2.911 fps variability
sbmrk24--5032 frames 79.008 seconds 63.69 fps (15.70 ms/f) 5.114 fps variability

///dx81 mastercoms very low latest headsfeet (+outlines,killfeed,msgs,killstreaks)
sbmrk----11795 frames 169.904 seconds 69.42 fps (14.40 ms/f) 4.937 fps variability
sbmrk24--5032 frames 45.972 seconds 109.46 fps ( 9.14 ms/f) 7.284 fps variability

///dx90 -vulkan mastercoms very low latest headsfeet (+outlines,killfeed,msgs,killstreaks)
sbmrk----11795 frames 201.871 seconds 58.43 fps (17.11 ms/f) 3.812 fps variability
sbmrk24--5032 frames 56.345 seconds 89.31 fps (11.20 ms/f) 6.224 fps variability

///dx95 (no config, no launch options, all low adv options 2560x1440p)
sbmrk----11795 frames 252.900 seconds 46.64 fps (21.44 ms/f) 2.911 fps variability
sbmrk24--5032 frames 79.008 seconds 63.69 fps (15.70 ms/f) 5.114 fps variability

///dx81 mastercoms very low latest headsfeet (+outlines,killfeed,msgs,killstreaks)
sbmrk----11795 frames 169.904 seconds 69.42 fps (14.40 ms/f) 4.937 fps variability
sbmrk24--5032 frames 45.972 seconds 109.46 fps ( 9.14 ms/f) 7.284 fps variability

///dx90 -vulkan mastercoms very low latest headsfeet (+outlines,killfeed,msgs,killstreaks)
sbmrk----11795 frames 201.871 seconds 58.43 fps (17.11 ms/f) 3.812 fps variability
sbmrk24--5032 frames 56.345 seconds 89.31 fps (11.20 ms/f) 6.224 fps variability

12600K + 1080ti + 32GB 3200mhz DDR4

sbmrk - 11795 frames 94.795 seconds 124.43 fps ( 8.04 ms/f) 11.176 fps variability
sbmrk24 - 5032 frames 113.019 seconds 44.52 fps (22.46 ms/f) 34.364 fps variability

idk why these results are so low :/

This benchmark just straight up doesn't work for me. A bunch of red error messages with "missing recvprop" and
"Cannot figure out which search path".
Edit: So apparently an update in mid 2023 caused all pre 2023-ish demos to stop playing correctly, and the fix is to load a beta version of tf2 that does support the playback; HOWEVER, since the 64 bit update, the beta doesn't work anymore.

with the launch option "-no_texture_stream". I got like 380 fps now on dx95 sbmrk24.

You're saying the "-no_texture_stream" launch option affected performance for you by over 25%?

Scratch that.
????
mat_picmip 3 was the issue. I thought worse texture quality was suppose to help with FPS but I guess not.

w/ mat_picmip 3:
5032 frames 15.258 seconds 329.80 fps ( 3.03 ms/f) 41.008 fps variability

mat_picmip -1:
5032 frames 12.974 seconds 387.86 fps ( 2.58 ms/f) 42.280 fps variability

can confirm has been the case for me since even before 64bit, mat_picmip -10 gives the most FPS by far on my benchmarks

CPU: Intel i3-12100f
Graphics Card: GTX 1660 Ti
Memory: 2x8GB DDR4 3200CL14 (F4-3200C14-8GFX)
Motherboard: Asus Prime H610M-A D4

OS: Windows 10 22H2
GPU driver version: 551.23
Resolution: 1920x1080
FPS config: recent mastercomfig low
dxlevel: 90
Mods: no explosion smoke script

Decent cpu for the price (~90eur). In 6s, outside of midfights it can get over 240 fps pretty stable. And changing picmip gave a surprisingly large boost for this setup

I had the opportunity to show how CPU limited the game is while upgrading my GPU.

CPU: 5800X3D (PBO -30 all cores)
Memory: 80Gb 3200Mhz DDR4 (2x16Gb 2x32Gb)(i know weird config)
GPU: see benchmark
Storage: 2x 990 PRO 2Tb in BTRFS raid 0

OS: Nobara 40 (kernel 6.10.3-201)
Game version: Aug. 2 2024 patch
Config: MasterComfig Ultra without Anti-Aliasing (due to AMD GPU linux quirk which degrades FPS)
Launch options: MasterComfig recommended w/ no texture stream
Resolution: 3840 x 2160

As the benchmarks shows the performance is basically the same despite a significant difference in GPU power which points towards a bottleneck either elsewhere in my system or with the game itself. I am planning on upgrading my CPU, memory, and motherboard once x870e motherboards are released. After i upgrade ill run this test again to see if the GPUs preform the same.

Just some other thoughts, does anyone know of a consistent way of testing networking latency in game? I have a few different 2.5Gbe, 10Gbe, and 25Gbe NICs from a few different brands and would love to quantify the effect of different network driver stacks within the game.