Why Less CPU Cores are *Almost* Always Better.

preview_player
Показать описание
Spreadsheet with all the results:

System Specs:

CPU: r7 5800x3D / r9 5900x (depending on test)
MB: MSI B350 PC MATE (latest bios)
RAM: 2x16gb 3600 MHz c18
Boot: Samsung 980 pro 2tb
GPU: RX 7800xt / RTX 3080 (depending on test)

==JOIN THE DISCORD!==

I guess a bunch of CPU cores is just a flex now :/

It's really easy to get caught up and overspend on a CPU, but for the majority of people (even many enthusiasts) you will most likely not notice a difference. Spending your money towards fewer cores, but individually are higher quality, for most people, is a better move. It is faster in gaming, and in productivity, usually that is handled by the GPU now. It surprised me how little a bunch of cores effect the end result.

lmk what you think!

0:00- More core, More better... right?
1:50- The pressure
2:50- Disadvantages of getting extra CPU cores
4:32- Advantages of extra CPU cores
6:11- What this means
Рекомендации по теме
Комментарии
Автор

While I agree with the underlying point, I feel you did a bit of a disservice by comparing an X3D chip against a normal X chip, as that's more likely to be what's contributing to the performance difference (as you even said, higher quality cores)
Also, in terms of productivity, in the context where there is rendering going on (like blender), GPU's are most certaintly the better option, but most other productivity apps will and likely cannot benefit from GPU acceleration in the same way. For example code compiling, or running virtual machines/docker containers is a place where more CPU cores definitely still help

Once again though, I agree with your core point, as long as you've got 6-8 good cores you're plenty fine nowadays

TinkerTaverns
Автор

You should have used a regular 5800X when comparing that and the 5900X. The X3D has a bunch more cache so it will perform better in games. The core count isn't the deciding factor in performance in this instance.

chexmixkitty
Автор

Depends if someone needs a cpu intensive task

NubAnSkrub
Автор

The comparison between 5900x and 5800x3d is totally incorrect!
The 5800x3d has a 3d cache that is known to be particularly usefull in game, this isn't about cores numbers...

sergione
Автор

For majority of the games, single thread performance of the said CPU core(s) is more important than have X amount of cores a CPU have.

Kacheng_
Автор

This is a bit of a niave take, especially with respect to the "GPU vs CPU" point. It's ultimately just sort of complex; GPUs are better at solving _certain types of problems_ than CPUs, but you can cripple their performance by picking certain tasks. Generally, things that are well-suited to GPU acceleration are operations consisting of trivial parallelism, largely branchless code and pure math operations (ie. graphics programming). But, trying to do things outside of that problem space get slow quickly.

Operations that act serially or atomically on a resource are also very poorly suited to GPU acceleration (and may even be impossible to implement), while scaling well with many physical cores (this consists of many I/O bound things).

GPUs also get a leg up from a lot of fixed-function hardware, where some function is baked into the silicon itself on the card. This can result in some marked speed gains, but usually comes with the caveat of lower quality work (GPU video transcoding for instance, while fast, generally results in lower quality and larger file sizes than software transcoding. A lot of this comes down to fixed function hardware being set set at manufacture time, meaning you can't benefit from codec and encoding improvements. Optimisations made for performance generally also sacrifice quality).

TL;DR: There is a reason why Disney's rendering farm consists of roughly ~55, 000 Xeon cores, and not GPUs. To that extent, multicore CPUs have many uses among software developers, server operators and many other heavy workloads (including some specific games). While for most modern video game titles video card performance is usually more important than CPU performance, it's not a statement that should be made broadly about computing in general; it's very situation dependent.

mileswithau
Автор

Big one at work is code compilation - each C or C++ file is handled in one thread, so more cores = faster compilations. Even on smaller projects, the more compilations you can do, the more Jenkins jobs can be run overnight or by the Harbormaster when doing a GIT push

peterwstacey
Автор

My biggest use is multitasking, optimising media in davinci resolve for even faster playback and re-encoding videos with higher quality lower filesize formats like AV1. A 5900x performs the same as a 5800x or 5600x in terms of if only 2 or maybe 5 cores are being used. But the second you have anything like shader compilation or emulation (in games), high core count CPUs make way more sense.
Nintendo switch and PS3 emulation on a 5900x is a breeze as well as editing and compression.

If you ONLY game, and you ONLY play native games (no emulation), yeah 6-8 cores are enough. 12-16 cores are for anyone going into the stuff I just talked about

Physuo
Автор

Depends on the variables and how the arhitecture of said cpu is (non x3d vs 3d from amd as an example) and what do you have in mind when you make a system, etc.

apostleoffate
Автор

For gaming, I think you really should be comparing a 5900x not to the 5800x3d because obviously that is designed to be amazing for gaming with the extra cache, but compare the 5900x to the standard 5800x. Those results would be more interesting imo. I don't think more cores are better for higher fps per se, but for higher 1% & 0.1% lows.

Also, my 12 cores would like to have a word with you about video rendering, because yes, while GPU encoding is much quicker, there is a lot of debate between video/PC nerds on whether or not GPU video rendering outputs the same visual quality as CPU software encoding. I always CPU software encode videos for myself and clients for that bit of higher visual qualtiy, so that I can more safely lower the bitrate by a little bit. Until it's proven than GPU encoding for video is the same or better than software encoding, I will continue to own high core count CPU's and will continue to CPU software encode videos.

Besides all of that, it simply is nice to do multiple intensive things at the same time for a pure smooth user experince. I'm going to buy whatever is the final halo CPU for AM5 years later once that comes out.

nicksterba
Автор

I'm a software engineer, so I bought myself 5700X instead of 5600X when I was building an AM4 system, and I knew it's not gonna be much better in games. So after all it really depends

maxxoft
Автор

I think it generally matters in some cases such as, as you mentioned, encoding and rendering, but for gaming it kinda just boils down to the single core performance.

X_irtz
Автор

Something you missed. For productive 3d workloads like blender, having more cores/threads is actually beneficial because then the GPU can render multiple "tiles" at once.

RawfunRahman
Автор

Downvoted. Comparing across different manufacturers and archs is straight up misleading. So I have to conclude that you either "don't know what you're talking about" or do know and intentionally chose to misrepresent the issue for greater effect.

The issue of a 14900K vs. a 7800X3D sipping more power does not follow from the number of cores but rather the architecture, process node, power limits / clock speeds and the X3D's cache.

If you wanted to get your point across you could have pitted two non-3D cache Zen 3 CPUs against each other and matched clocks or, alternatively, disabled some of the cores on your 5900X and compared results and show that more than 6 cores made little to no difference in gaming.

Innosos
Автор

The thing that almost no one ever mentions is LATENCY and IO performance. Lower core count cpus, with one ccd, have better windows response times, so the overall usage feels snappier. It matters for productivity too.


Hybrid architecture seems to have latency issues, and dual ccd/infinity fabric struggle too. Intel's older ring bus still seems to have the best io performance with many people claiming to feel a snappier pc with 10700K vs 5800x3d or 12900K. I wish people brought it up more actively.

HeartOfAdel
Автор

Great video. All I do is game so super happy with my R5 5600.
Maybe you shouldn't use the 5800x3d in your comparison without mentioning the 3d memory.
Someone looking to buy may get the 5800 non 3d and think that they will be getting the same performance.

justincowans
Автор

In most cases for gaming, the architecture and single core speed of the cpu makes up for the difference in core count.

dragonman
Автор

for me ffmpeg file compression makes high core count cpu very appealing, because hardware encoders are very very fast but but compromise in quality or in file dimention (at the same bitrate quality is worse / at the same quality bitrate is much higher).
Is a niche use case and won't apply to most peoples but is still a use case.

lorenzo
Автор

I fully agree, was kinda hypnotized by the core count when I got a 3900X, upgraded to a 5900X, but it's overkill for what I do, next one will definitely be 8800X3D ^^

pascaldifolco
Автор

Emulation is a great use case for more cores. Virtual machines also generally require lots of cores and memory if you're multi tasking. I noticed that Reshade uses all 12 cores of my r9 5900x doing different effects on each. Clock speed/IPC is king in gaming, single thread single core is why u can have a fine experience with a 7600x/13600k system. More cores come in very specific use cases such as art production, streaming, editing, programming, etc which is why I went with the 5900x and will upgrade likely to a 8950x3d or similar when the 50 series or 8000 series comes.

arghpee