[ / / / / / / / / / / / / / ] [ dir / acme / animu / arepa / fast / sunny / tacos / vg / vichan ][Options][ watchlist ]

/tech/ - Technology

You can now write text to your AI-generated image at https://aiproto.com It is currently free to use for Proto members.
Email
Comment *
File
Select/drop/paste files here
Password (Randomized for file and post deletion; you may also set your own.)
* = required field[▶ Show post options & limits]
Confused? See the FAQ.
Expand all images

File (hide): b113005d2759cd1⋯.jpg (3.78 MB, 4288x2848, 134:89, shoulder_shrug.jpg) (h) (u)

[–]

 No.953282>>953288 >>953451 >>953768 >>953967 >>954172 [Watch Thread][Show All Posts]

>People still call coprocessors where graphics rendering is typically offloaded to with fully programmable shader models "Graphical Processing Units"

Why are we still doing this? The term GPU comes from a time when everything used fixed-function rendering pipelines. Modern GPUs are fully programmable. Even calling the programs that are designed to take advantage of modern GPU execution units "shaders" are a misnomer because even in the context of graphics they're being used for more than just shading shit now.

We should start calling GPUs "Manycore coprocessing units" (MCUs" instead. That would be a more fitting name honestly. It would certainly be less retarded than the absolute abomination of a buzzword "GPGPU" I mean seriously who the fuck thought that was a good idea?

 No.953286>>953287 >>954161

They can be called "GP"GPU when they start publishing full specs on how to program them.


 No.953287>>953296 >>953797 >>953970

File (hide): c986a707c3ef6aa⋯.jpg (9.06 KB, 259x194, 259:194, wizard2309823.jpg) (h) (u)

>>953286

openacc

nvidia cuda

openmp


 No.953288>>953294

>>953282 (OP)

>typically offloaded

Literally answered your own question. Shit thread.


 No.953294

>>953288

GPUs have taken on a more general purpose role sense the rise of machine learning and shit


 No.953296>>953797 >>954164

>>953287

Also OpenCL and DirectCompute


 No.953390>>953393

Also why do we call computers computers when they do more than compute stuff


 No.953393>>953451

>>953390

such as?


 No.953451>>953452 >>953621 >>954163

>>953282 (OP)

GPUs are also distinguished by their focus on vector operations, so "manycore vector coprocessor" might be the best description. Another distinction held by GPUs is their being the only application where VLIW significantly outperformed RISC or CISC.

>>953393

Storage and retrieval?


 No.953452

>>953451

>implying you don't have to compute the location of the file fragments


 No.953621>>953714

>>953451

>manycore vector coprocessor

MVC? Sounds like the most appropriate name for them. Now we just have to autistically start calling them that every time the latest GTX or Polaris cards come up and hopefully it sticks with normalfags eventually


 No.953714>>954162

>>953621

Hoping this becomes reality. How to meme this into existence?


 No.953736>>953737

Holy fuck you people are autistic.


 No.953737


 No.953768>>953789

File (hide): 1c8a711b23c90ef⋯.jpg (4.06 KB, 500x500, 1:1, mbry.jpg) (h) (u)

>>953282 (OP)

member PPUs?

I member.


 No.953785

guise my x86 abacus stopped working


 No.953789>>953790

>>953768

The name PPU was appropriate though since it could literally only do picture processing


 No.953790>>953795

>>953789

I was referring to physics processing units.


 No.953795>>953796

>>953790

Didn't only PhysiX make them? And then nvidia bought them out really quick and killed off the tech?


 No.953796>>954165

>>953795

Nvidia never killed off the tech. PhysX still lives inside of every new GTX card and driver. The PPU is integrated into the die and while few software/games have dedicated PhysX libraries anymore, the drivers still use the integrated PPU to offload general physics computation from the CPU


 No.953797>>953827

>>953287

>>953296

+Vulkan compute shaders


 No.953827

>>953797

*SPIR-V


 No.953967>>954166 >>954283

>>953282 (OP)

Because they're opaque, undocumented piles of shit that do fuck all else.


 No.953970>>954161

>>953287

These are APIs exposing features, but they don't actually document what the GPU is doing, how and in what order, or provide lower level access to its internals.


 No.954161>>954166


 No.954162

>>953714

be autistic about it but dont overdo the copypasta pls

hopes that mememagic exists


 No.954163

>>953451

No modern GPU is VLIW since the Radeon HD 6xxx days.


 No.954164

>>953296

OpenCL is dead, thanks to a consolidated effort by Apple and Nvidia.


 No.954165>>954254

>>953796

There is no "integrated PPU", it's all done in the shader cores.


 No.954166

>>953967

>undocumented

>>954161


 No.954172>>954283

>>953282 (OP)

>Why are we still doing this?

Because for an average user the GPU is doing fuck all besides graphics. GPU offloading haven't really caught on outside some very specialised domains and GPU-accelerated programs are still a very rare breed - unless they're doing graphics of course.

>from a time when everything used fixed-function rendering pipelines.

There's still lots of fixed-function hardware in a modern GPU. Rasterisation, texture sampling and filtering, ROPs...

>Even calling the programs that are designed to take advantage of modern GPU execution units "shaders" are a misnomer because even in the context of graphics they're being used for more than just shading shit now.

This is pure DirectX faggotry. OpenGL has called them "programs" since the very beginning. Of course, since normalfags are buying GPUs purely for DirectX gaymes, DirectX terminology is what vendors end up using to describe their HW in marketing materials.


 No.954254

>>954165

There's still a Physx controller that uses the cores though


 No.954283>>954326

>>953967

Isn't writing directly on the metal for individual GPU dice a fucking terrible idea in principle? Having a better idea of how a badly written driver might be fucking up is fine, but shipping binaries optimized for particular GPUs absolutely isn't.

>>954172

Pretty sure even the likes of Excel are exploiting GPGPU for parallel compute.


 No.954320>>954325

>People call portable computing devices "phones"


 No.954325

>>954320

>people still call devices for looking at memes and shitposting "computers"


 No.954326

>>954283

Pretty sure only LibreOffice Calc is doing that, and then only for very large sheets a normalfag will never encounter.




[Return][Go to top][Catalog][Screencap][Nerve Center][Cancer][Update] ( Scroll to new posts) ( Auto) 5
35 replies | 2 images | Page ?
[Post a Reply]
[ / / / / / / / / / / / / / ] [ dir / acme / animu / arepa / fast / sunny / tacos / vg / vichan ][ watchlist ]