>2000+18
>still having explain yourself
2000+18
Other urls found in this thread:
en.m.wikipedia.org
twitter.com
I haven't bought new hardware since 2014.
Also, for a single application having more than 4 threads (2 cores can handle 4 threads) has huge amounts of diminished returns. You get like a 4% boost from going from 4 to 8 cores in an extremely optimized multithreaded application.
4 to 8 threads*
640 cores ought to be enough for everybody
I'm a poorfag
if you think you need hyperthreading for gayming you're deluding yourself
because i do actual work and don't play gaymes. Plus I have access to high power servers which I could use if I needed to compile anything big (or i just let my computer do it over night or something)
>tfw even my phone has 8 cores
I have four cores per CCX.
I've never HAD a 4 core cpu. 2 > 8 > 16
Most the software I use barley utilizes 2.
I'm pretty content with my Core2 Duo. Only reason I'm upgrading is to get USB 3.0 support, probably going for an i7 iMac. An i3 would be more than enough but it'll be nice having 8 threads
what's with the old guys mustache
is he some kind of stalin/hitler afficionado
wow, come to think of it, almost naan of the software i use uses 2 either
There are 100% parallelized tasks so not sure where you got your retarded numbers.
No there aren't.
let's say near 100%
encoding, transcoding, rendering, etc
He's right about diminishing returns existing but it depends on how parallelized the application is
Can't you just like encode 1/16th of a movie in one thread for 16x speed increase compared to one thread? You will get slightly bigger files because the edge frames won't be optimized, but whatever. Obviously not scalable infinitely because one frame is encoded in relation to the ones next to it.
I bring around with me a shit Chromebook. If something happens to it. No big deal. If I had a shiny gaming computer or an unholy Mac, some nigga if gonna take it. (Or more realistically I might drop it and break it)
More like near 30%
It depends on the circumstance dumbass. Read a book. Try not to get a papercut.
>book
How about a Wikipedia article, boomer.
en.m.wikipedia.org
>doesn't include the CUDA example
But the article contradicts completely.
they truly are lesser humans
>6 seconds
kek
Almost all programs don't go above 40% parallelization.
>You get like a 4% boost from going from 4 to 8 cores in an extremely optimized multithreaded application.
SMP scaling is hard to do well, but a 4% improvement going from 4 --> 8 can hardly classify as "extremely optimized multithreaded" anything
100% scaling is simply possible. Simple things like throttling and boost clocks already make that impossible.
The memory connection may not be able to fully saturate all cores at full load either, even if they are completetly independant of eachother.
>Simple things like throttling and boost clocks already make that impossible.
What about a processor with constant overclock and with adequate cooling (no throttling)?
>özil mit erdogan
Was soll es 1fach
Those that do can be crunched on a gpu as - almost by definition - gpus exist to scale to N threads. That said there are still some inherent advantages to scaling on the cpu (mostly maturity of software) but fundamentally wit the rise of dedicated graphics anything that can scale to 4+ threads can scale to thousands of threads and falls intro GPU territory.
These days CPU computation is either by design or an inherent limitation of mathematics and as such limited in a way that ALWAYS will demand absolute serial performance.
DDR4 prices are Jewish tier otherwise I would've gotten a Ryzen already to replace my shitty Haslel.
Because my X201 is comfy.
And I don't have money right now.
why would I need more?
All my compiling/training is offloaded to a build server anyway.
>not knowing amdahl's law
3570k stil werk gud
Threadlets, lol
>CPUs started
>JustWerks™ Still Edition
You should be selling your cpus wiht a vps service
This is why. Also, a lot of older machines from 10 years ago will run very fast if you just put an SSD in them.
My phone has 8 cores
My desktop has 4, but you don't really need more, because x64 is more powerful than ARM
How are GPUs with 3000 cuda cores able to do anything then?
I don't I have 8
Based 2700x saving me from Intel's yearly chipset thermal jizz 5ghz corelet overpriced high temp spectre meltdown hell
I bought gskill flare-x Ryzen ram 16gb 3200 at 350aud it was worth it
By feeding them massively parallel tasks - thats what they are good at. When you've got millions of pixels, graphics cards can shine when doing tasks on them
/thread
>dat GPU though
Thats a result of shitty programming not the technology itself
Most programs weren't even designed with parallel concurrent processing in mind