It is in the config.

You do get better FPS, probably because of your bad graphics card, but you should also play a while with dx9 and dx8 because average FPS is only part of the story. See which one feels smoother and has less stutter. If that is dx8 for you, then use it, but if not, I would sacrifice those few extra frames for more stability.

DirectX 9 is much more efficient renderer that uses your graphics card more effectively. And about skins, they shouldn't affect FPS too much except for the initial generation.

I've made a very basic new benchmark that should be better at testing FPS on modern TF2 for now.

Don't worry, he made a 10h first impressions video for 7.07. https://www.youtube.com/watch?v=x6zJ-G2syho

What does net_graph 4 look like?

r_worldlightmin 0.004

me too

I'm currently investigating possible fixes to this issue, with some work on it done in 5.0.0.

5.0.0 released with revamped folder structure, formatting improvements, comment improvements, sound fixes, performance optimizations and new stripped quality addon.

download game_sounds_weapons.txt

put it here: tf/custom/<custom_folder>/scripts/game_sounds_weapons.txt

open it, go to this line

modify "wave" for Weapon_Winger.Single to ")weapons/pistol_shoot.wav"

and "wave" for Weapon_Winger.SingleCrit to ")weapons/pistol_shoot_crit.wav"

Thank you for the suggestions.

1) I will add a note to check the top of the config in the init section and installation instructions. I want it at the top of comfig so it is more likely for people to see they need to change -dxlevel and mat_dxlevel.

2) I'll have to think a bit more about how a better exec system would be implemented to solve this issue.

I'm sure Valve knows they should make TF2 into an esport, I just hope they execute it well.

Not sure what's there to confirm, if you can't change cl_interp_ratio while in spectating, then the server is restricting it and you'll have to use cl_interp.

How is it pointless if it works? The default is unlimited, so on a lot of servers, it will use 170016. If a server does happen to limit it, then it will be set to 100000. I don't see the problem.

It still affects network feel and gets rid of some stutters/jumping view from incorrect client prediction.

Are you using the -primarysound -snoforceformat launch options?

Also, you can try raising your snd_mixahead cvar to 0.08 or 0.1.

cl_cmdrate 66 // Send to server at this many times per second
cl_updaterate 66 // Get from server at this many times per seconds
cl_interp_ratio 1 // Reduce entity interpolation times to minimum
net_queued_packet_thread 1 // Work around bad Windows networking
cl_interp 0 // Use cl_interp_ratio for a more accurate value
net_compresspackets 1 // Compress packets to prevent split packets
net_compresspackets_minsize 1260 // Only compress packets if we would send a splitpacket
rate 170016 // High rate to reduce choke and scale to packet rate
            // Might want to lower rate if you have <1.4Mbps stable upload
net_maxcleartime 0.015 // Balance between reducing network delay and choke
net_maxpacketdrop 5000 // Use default packet drop threshold
net_splitrate 4 // Split 4 packets per frame, reduces choke but is more CPU intensive
net_splitpacket_maxrate 170016 // Scale splitpacket to rate
                               // Might want to lower rate if you have <1.4Mbps stable upload
cl_timeout 70 // Wait the maximum time before automatically timing out on client
cl_pred_optimize 2 // Skip repredicting if we do not need to do it
cl_lagcompensation 1 // Ensure lag compensation is turned on
cl_predictweapons 1 // Ensure weapon firing prediction is turned on
cl_smooth 1 // Smooth out prediction errors
cl_smoothtime 0.1 // Smooth out view for 0.1 seconds

try these net settings

Sorry, I meant that comment for CPUs with less than 4 cores with HDDs. Though, are you having any problems with snd_async_minsize 0, like any stuttering? If so, use a higher setting for it.