[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/sci/ - Science & Math

View post   
File: 190 KB, 1024x768, Supermicro-H8QGi-F-Quad-6000-Opteron-CPU-Socket-G34-Server-Motherboard-1024x768.jpg [View same] [iqdb] [saucenao] [google]
14717802 No.14717802 [Reply] [Original]

So many important research tasks depend on GPUs yet the main selling point for GPU companies is "it plays games so good!!". When is it time for comp sci to reclaim what gamers have taken for granted all this time
Also as another note why not just multithread multiple processors stacked all along the motherboard to fit the architecture and use that instead of rbg shittery

>> No.14717816

>>14717802
Non-gaming GPUs have existed for a while now.

>> No.14717886

>>14717802
>So many important research tasks
name one thats genuinely important

>> No.14718722

>>14717886
Figuring out how to play games so good

>> No.14718727

>>14717802
> What are GPGPUs
Read nigga Read

>> No.14718763

>>14717802
how about actually educating yourself instead of making retarded statements on 4gay?

>> No.14718777

>>14717886
Simulations of all kinds.
>>14717802
Multi-socket processors are a selling point, and most pc tasks aren't that processor intensive.

>> No.14718779

>>14717802
Don't worry, they are building AI into GPUs to monitor everyone now. The panopticon is the peak of pooter science.

>> No.14718788

>>14717802
>When is it time for comp sci to reclaim what gamers have taken for granted all this time
so you want to start your own ponzicryptoscheme because your triggered at me looking at pixels? others are way ahead of you and they werent that successful actually

>> No.14718792

>>14717802
>GPUs made for graphics.
just build better CPU architecture for computations.

>> No.14718815

>>14717802
Nobody cares what you think or ask. Even if you spam it to people. Fuck you thats what.

>> No.14718819

Because it's easy to make a big supercomputer work with a lot of GPUs. If you're a research scientist, you can quickly knock out a -nix configuration that uses all those GPUs efficiently in a small time span. Doing it "correctly" with lots of normal CPUs means installing z/OS and learning COBOL or make an equally autistic language like ADPAC. Tell me OP, how many CS grads do you know what to write software using 1970s era lexicon and conventions that'll take weeks to properly set up? Sure they could get a 20-30% performance boost right off the bat but it means fucking with firmware and creating computationally sound solutions. Few have the will to power for it versus an off-the-shelf solution.

Chalk it up to the commercialization of computing. Most /sci/ence computing benefits from a mainframe. But it's not done because mainframes are big, expensive, and leased exclusively by one company who is notorious for not providing freeware versions of their software for academic use. Whereas you can go into Best Buy right now and get 10 GPUs, and set them all up doing some task within a few hours.

>> No.14718860

>>14717802
GPUs began as 3D accelerators for playing games that did a few preset tasks that took a lot of CPU time in games. Only over time they became programmable enough to become useful for other tasks.

>> No.14718867

>>14718777
>Simulations of all kinds.
name one

>> No.14718869

>>14718867
the cum dripping from my cock

>> No.14718875

>>14718819
They have linux on ibm supercomputers on z
https://en.m.wikipedia.org/wiki/Linux_on_IBM_Z

OP just buy one of these dude you’re all set
Start writing your grant for funding and kissing ass like most “scientists” do
Oh wait you’re just another poo with a shitcoin
Guess you’re SOL

>> No.14718887

>>14717802
>So many important research tasks depend on GPUs

GPUs are designed to handle spreadsheets really well. That's it.

>> No.14718903

>>14718875

Yes, but that's my point. IBM does not sell Z machines anymore. They lease them. At least the high-preformance ones a scientist would want. Getting a normal Z machine is hard enough and requires registering with IBM and doing all IT through them. This means use of third-party software is strictly controlled. And of the available third party software, it's in ancient languages nobody wants to learn. Or assembly.

>> No.14718953

>>14718903
If you can’t pony up the cash to lease it then you can pound sand
There are other supercomputer vendors
You need loot
I don’t see a problem here
They did the hard work so you can run redhat or ubuntu on it, pay up