Star Citizen: CPU Core Scaling (using 13900K with 4090)
HTML-код
- Опубликовано: 30 июл 2024
- How Many CPU Cores Do You Need for Star Citizen? Complete CPU scaling analysis using 13900k. Star Citizen Alpha 3.17.4
►Join the Game Room: www.game-day.xyz/
System Specs
CPU: 13900KF
RAM: Viper 32GB 4000MHZ CL16 @ 4100MHZ CL16 [Tuned]
GPU: Nvidia RTX 4090
OS M.2 NVMe: Crucial P2 1TB M.2
Game M.2 NVMe: WD_BLACK SN770 2TB
Methodology:
Benchmarking software: FrameCapX
In Game Settings: 1080P, Everything on Max, including motion blur
"Lorville run" from the apartments when you come out from the elevator running to the terminal
0:00 Into
0:10 Test setup & Methodology
1:37 Benchmark results 1
2:37 Benchmark results 2
4:01 Benchmark results 3
6:08 Summary of results
#starcitizen #4090 #13900k Игры
Thank you for this test. Keep it up!
Thank you!
@@gamedaytoday1 what cpu would you recommend upgrading to for star citizen? raptor lake or the upcoming ryzen 7000x3d chips?
@@AdrianPragacha My bet is 7000x3d will be better most likely, but a proper test need to be done with fully tuned ram. Will see soon
@@gamedaytoday1 Thanks for your input, would love to see a ram tuning guide :)
Well done!
Thank your :D
i always want more data. do more vids. subbed.
More data is coming, any requests?
If you could get your hands on some cheap Rev.B or DJR that could be interesting to see if the bandwidth can compensate the latency increase,your tuned 4100 vs a 6ghz with loose timmings.
Difference made by downclocking solely ram speed from 4100 to 3600 is insane. Looks like the game actually benefits a lot from 4000+ ram speed. Or did I miss something?
Ram timings. Frequency is not everything.
This would explain why the 5800x3D is destroying recent-gen CPUs in SC. The higher amount of cache means we can take advantage of the principal of locality - it doesn't have to do a slow transfer across the system bus, assuming how SC favours memory scaling as you mentioned in your other video.
Yes its definitely about lowering the latency, but could it also be bandwidth, I'll find out soon once I get my hands on the 5800X3D to do a full analysis.
How do you get the game to even run???¿
You got some quality content here. I'd suggest branching out to other games as well. People like to see specific reviews like this and the performance evaluation of diffrent games.
Might do that in the future just getting started, but for now main focus is star citizen
Very interesting result. Shame that the memory does not support. Please do more tests like these :)
If you make a video showing how you tuned your ram it would be helpful.
Once I can screen capture from the BIOS
13900KS and 5800X are only 8 cores cpus and some of your core are used by your os...
What about a 18c/36t Intel Xeon with quad channel memory ? Double the ram speed for the same timings and 18 cores vs your smalls 8 cores will be interesting. Thoose are really cheap now you can build a full system with 128gb ECC quad channel and 18c Xeon for less than the price of the 13900k but thoose Xeons don't go very high in frequences (3.5-3.7ghz max).
Logical data. Looks like I need to upgrade my mobo but at the limit of AM4 with 5800X3D. Need to get more ram speed, etc.
5800X3D is great and can benefit from faster ram tuning + 1:1 infinity band
how about 3.18? it seems that saying CIG solve this problem?
Did not solve it for me it actualy got worse, thats on win 10 tho might be different on win 11
Do a similar e core test , 4 vs 6 vs 8 vs 12 vs 16
think I'll go with red team this time
"trash and garbage" double!
Should be nice to compare with the AMD 5800x3d.
Will do soon subscribe so you don’t miss it
E cores will never be a benefit to gaming. Games are latency sensitive, so the very nature of those E cores having lower IPC and clocks, means they turn in work at a slower rate. The best games can do is be programmed to ignore them. They are only good in productivity, because most MT productivity won't care when it gets that work. The overall job gets done over a longer period of time than the milliseconds a GPU is expecting the next frame data. Every individual frame is the whole job.
E cores have their weaknesses, I don't like Intel's architecture but as a software developer, there's plenty of game features that could run on a slower but predictable core, for instance audio loves realtime scheduling to avoid drop outs and stutters.
Performance nowadays is fundamentally reliant on power efficiency, because the fast data flow driven part of action games can have more power to rip through that part when benefitting from offloading work to less power hungry processors.
@@RobBCactive There isn't a point to it though. Devs would have to go out the way to address them solely for the sake of using them, in lieu of P cores, to segment off operations that can afford to be wildly asynchronous. Game software audio solutions have been happy with like 1 thread and there are enough P threads to spare.
@@blkspade23 OK, you don't get soft realtime programming, there's no great effort to putting something like audio on a different core.
Games consoles have reserved cores and SDKs to provide media features.
Saying "just one thread", but it's the power consumption and the availability guarantee that us the benefit. Wasting energy is irrecoverable performance loss
Using words like never is unwise in software where yesterday's "tricky" becomes industry standard once a library solution gains traction.
Gamers often appear to have ADHD, while gaming they have video and chat running, browsers and perhaps some work.
@@RobBCactive It's not that it matters specifically for audio is what I'm saying. The way audio is already being handled isn't broken. My initial statement saying "never" is that they E-cores won't provide a speed-up in games, because what slows down games FPS on the CPU's end is making the GPU wait for data. Something that has been broken is render being last in line behind a main thread's collection of operations. Clearing that hurdle, you want to also leverage more of the available threads to send bigger draw calls to the GPU. There are still other elements that need to be able to complete somewhere in the time frame that frames are being rendered.
Currently any game that doesn't outright disregard the E-cores turns into a stutter fest, because they are slower and turning in work far later than the P cores. Star citizen is one such game that is capable of using more than 16 threads, but doesn't know to ignore the E-cores. Having them on causes big frame time swings. Their latency is too high. Outside of your singular audio example, we really would have to have games built solely to leverage (showcase) them, after maxing the P cores. Most game devs are probably more worried about leveraging the XBSX/PS5 APU.
Best case is a more indirect benefit of having all non-gaming related OS services offload to them. Windows 10 by itself on a SSD can fully occupy 4 threads. Since the 8700K, Intel has only felt the need to slap on more cores to stay relevant at MT productivity. E-core are just more of that, but also to make up for the P cores being way too power hungry. None of MT productivity stuff cares about latency to the degree that games do. Or the human's sensitivity to it.
Back before EasyAntiCheat was added to SC, I ran multiple instances of SC on one PC. One game instance running normally + others running in Sandboxie sandboxes so they can not see each other and prevent each other from running. Sandboxed instances ran worse than the non-sandboxed. I could get three instances running in a 'playable by SC standards' way, but the fourth instance would tank every other instance and be a slideshow as well. Messing with setting affinities of the different instances to not overlap only helped a little. I had fun trying other silly things, like seeing if installing the entire game to a massive RAMdrive made a difference in FPS (It didn't).
49" Samsung G9 LC49G97TSSNXDC
3x 32" Samsung G7 LC32G75TQSNXZA
AMD Threadripper 3970X 32-core
256GB 3600MHz RAM
RTX 3090
2x NVMe SSD (OS + Games)
robertsspaceindustries.com/spectrum/community/SC/forum/50264/thread/blueboyx-monolith-build