|
ate all the Oreos posted:how practical would it be to just make them like, a standard socket like the CPU? It's kind of been covered sort of, but I have bad news about how 'standard' CPU sockets are.
|
#
?
Oct 4, 2017 09:46
|
|
|
|
| # ? Jun 8, 2024 13:55 |
|
|
Doc Block posted:isn't part of the reason that opencl never caught on was because apple was the only one really invested in it? it was basically an apple invention, iirc, until they lost interest in favour of Metal. nvidia were interested exactly as far as making sure opencl support wasn't a reason to not buy an nvidia card, amd were interested because they needed to compete with cuda but they've put more effort into various hokey run-cuda-by-parsing solutions for the last few years. and a whole host of other companies (especially mobiles) were briefly interested in it as a selling point, but google shot that down by not supporting opencl on nexuses in favour of something called renderscript as part of an internal google turf war
|
|
#
?
Oct 4, 2017 10:55
|
|
|
pvr is tiled-based, but internally uses a scanline renderer. Tris within a tile are z-sorted, and then only the topmost pixel is drawn in lines. the actual fill rate is pretty poor compared to amd/nvidia as it is using more general processors instead of dedicated vertex/pixel pipelines which are considerably simpler in terms of instructions they can do. pvr really falls down on multiple layered transparencies and full screen effects which the normal gpus are chomping through with greater performance. deferred rendering is becoming ubiquitous, which for those who don't follow this stuff is where you draw everything with no textures to just work out the z-buffer. you have to run all the vertex shaders, but the pixels just render depth. then you render everything again with full pixel shaders, so only the visible pixels incur the cost of the pixel shader. but you don't write out final colours, you just store the normal/base color/roughness/conductivity/etc. then you do a full screen pass that applies shadows/lighting/reflections. based on the normals/physical parameters once for each pixel in the final image. then you do the fullscreen post process effects (colour balance, tone mapping, antialiasing, depth of field, fog, etc). as the hw is designed to do multiple passes on the scene, it's optimised to do these passes very efficiently. recent hardware has introduced 'tiles' to these buffers, but it is fundamentally different. pvr draws each tile full of final colours from z-sorted lists of prims. regular cards use tiles as cache units, and to track simple state flags to accelerate the composite/clear passes. there's no storage of prims/sorting them like pvr does.
|
|
#
?
Oct 4, 2017 11:35
|
|
|
i'm irrationally happy that powerVR came out of the 3dfx days as the eventual victor. probably not the same company any more i know
|
|
#
?
Oct 4, 2017 12:15
|
|
|
feedmegin posted:It's kind of been covered sort of, but I have bad news about how 'standard' CPU sockets are. eh i was kinda thinking it wouldn't be that different from how PCIe gets a new revision every couple years it seems but then i remembered that that's actually a proper standard and it's backwards compatible, and good luck doing that with a chip lol
|
|
#
?
Oct 4, 2017 12:27
|
|
|
Trimson Grondag 3 posted:i'm irrationally happy that powerVR came out of the 3dfx days as the eventual victor. probably not the same company any more i know Well, now it's Imagination, which is p much going to go broke now Apple has decided to use their own GPU instead of licensing PowerVR.
|
|
#
?
Oct 4, 2017 13:23
|
|
|
ate all the Oreos posted:eh i was kinda thinking it wouldn't be that different from how PCIe gets a new revision every couple years it seems but then i remembered that that's actually a proper standard and it's backwards compatible, and good luck doing that with a chip lol tbh, it wouldn't be too big a stretch with hbm coming down the pipeline into consumer poo poo soon, have a big slot for the gpu on the mobo, call it accelerated graphics slot or whatever, have gpus be basically socs, like where amd and intel are moving with their desktop cpus already (amd even has the efi/bios in the chip now, apart from ram everything on the motherboard is optional and that's going to be changed by hbm in the future). that'd probably work. what won't work though, is getting 250+ watts of heat off of that chip 
|
|
#
?
Oct 4, 2017 13:30
|
|
|
in the hpc space nvidia is pushing nvlink which can do gpu-gpu and gpu-cpu; you can get it on a power8 or power9 cpu from ibm but intel isnt really interested in that obvs; anything that needs more than gen3 x16 is a competitor still a lot slower than on-die hbm tho
|
|
#
?
Oct 4, 2017 17:52
|
|
|
josh04 posted:it was basically an apple invention, iirc, until they lost interest in favour of Metal. there's quite a bit of time between the two: OpenCL was announced at WWDC 2008 and introduced in Snow Leopard in 2009 Metal was announced at WWDC 2014 and shipped that fall on iOS and the following fall on macOS
|
|
#
?
Oct 4, 2017 18:14
|
|
|
i keep working with amd engineers working on opencl, actually, so presumably they haven’t completely lost interest
|
|
#
?
Oct 4, 2017 18:23
|
|
|
rjmccall posted:i keep working with amd engineers working on opencl, actually, so presumably they haven’t completely lost interest opencl is their only hope of dealing with Cuda which the is the golden ticket to hpc and data center riches for example virtually none of the big dnn frameworks have decent if any opencl support they do have amazing Cuda support
|
|
#
?
Oct 4, 2017 23:48
|
|
|
I wanna know the bizarre aechitectures that must be used on the third tier GPUs like vivante
|
|
#
?
Oct 4, 2017 23:51
|
|
|
also I stumbled on a Verilog implementation of gcn on GitHub which is p cool
|
|
#
?
Oct 4, 2017 23:51
|
|
|
as i understand nvidia and other companies dont manufacture their physical chips. what prevents the manufacturers from copying the design themselves?
|
|
#
?
Oct 5, 2017 00:11
|
|
|
patent/copyright laws.
|
|
#
?
Oct 5, 2017 00:14
|
|
|
Malcolm XML posted:third tier GPUs you're talking embedded but this got me thinking about old desktop cards matrox parhelia SiS xabre XGI volari (apparently they had a dual-GPU card  )i still wonder who bought any of that crap because as i recall they were all total garbage compared to ATI and nvidia
|
|
#
?
Oct 5, 2017 00:26
|
|
|
so i dabbled in a tiny bit of opencl a while ago and eventually i'd like to come back around to it and play around with it a bit because i'm a turbodork who finds this kinda thing interesting, should i even bother with it if it's going out of style or should i look into something else instead?
|
|
#
?
Oct 5, 2017 00:51
|
|
|
SO DEMANDING posted:you're talking embedded but this got me thinking about old desktop cards
|
|
#
?
Oct 5, 2017 00:54
|
|
|
the hottest take
|
|
#
?
Oct 5, 2017 01:00
|
|
|
them bit boys in trouble again!
|
|
#
?
Oct 5, 2017 01:32
|
|
|
SO DEMANDING posted:you're talking embedded but this got me thinking about old desktop cards wasn't matrox notable for having video cards with three outputs back when most still had two at max?
|
|
#
?
Oct 5, 2017 02:03
|
|
|
somewhere in my boxes of boards i have what was an extremely expensive PCI 3D accelerator made by evans and sutherland. i oughta build a 90s pc workstation someday freedom graphics or realimage 2000 or something? it has tiny fans on the main chips, a daughterboard, and at least one simm slot looked somewhat similar to this AccelGraphics Eclipse, which is E&S based:  I also have some prototype of some "realizm" card or other IIRC. Intergraph, not the later 3DLabs ones atomicthumbs fucked around with this message at 02:27 on Oct 5, 2017 |
|
#
?
Oct 5, 2017 02:22
|
|
|
Farmer Crack-rear end posted:wasn't matrox notable for having video cards with three outputs back when most still had two at max? idk we have a whole rack of cpu-heavy 2016 servers at the office and they have like the matrox dorito "gpu" in 'em
|
|
#
?
Oct 5, 2017 02:27
|
|
|
atomicthumbs posted:somewhere in my boxes of boards i have what was an extremely expensive PCI 3D accelerator made by evans and sutherland. i oughta build a 90s pc workstation someday bitchinfast 3d 2000
|
|
#
?
Oct 5, 2017 03:09
|
|
|
remember 3dfx and their crazy 4(+?) core cards that never made production iirc
|
|
#
?
Oct 5, 2017 03:12
|
|
|
the voodoo5 6000 was supposed to have four separate discrete processors, not just cores
|
|
#
?
Oct 5, 2017 03:18
|
|
|
there's a few dozen prototype 6000s floating around apparently
|
|
#
?
Oct 5, 2017 03:30
|
|
|
atomicthumbs posted:somewhere in my boxes of boards i have what was an extremely expensive PCI 3D accelerator made by evans and sutherland. i oughta build a 90s pc workstation someday kind of like this one on eBay but PCI?
|
|
#
?
Oct 5, 2017 08:26
|
|
|
ate all the Oreos posted:so i dabbled in a tiny bit of opencl a while ago and eventually i'd like to come back around to it and play around with it a bit because i'm a turbodork who finds this kinda thing interesting, should i even bother with it if it's going out of style or should i look into something else instead? it still has the advantage of being multi-platform and multi-manufacturer. on my macbook pro i can run tasks on both the internal gpus etc. and nvidia added opencl 1.2 support to their driver in the last two years so they all support roughly the same featureset. it's just sad reading about cl 2.0 and not being able to get any of the nice features unless you have a specific AMD card and do the magic driver rain dance.
|
|
#
?
Oct 5, 2017 11:48
|
|
|
josh04 posted:it still has the advantage of being multi-platform and multi-manufacturer. on my macbook pro i can run tasks on both the internal gpus etc. and nvidia added opencl 1.2 support to their driver in the last two years so they all support roughly the same featureset. yeah the thing that appealed to me was supposedly being able to compile it to run on an FPGA somehow??? which seems real cool since I have a few FPGA dev boards I'd like more excuses to play with someone above mentioned they made some translation layer (or are making?) to translate to the new thing (vulkan? i have no idea what these things are) so i guess it's reasonably future-proof to go with for at least a little while then?
|
|
#
?
Oct 5, 2017 12:00
|
|
|
altera sell fpgas which run opencl, i've only had a cursory look at them though, and i suspect you'd still be doing a lot of work specific to the fpga to get decent results.
|
|
#
?
Oct 5, 2017 12:29
|
|
|
josh04 posted:altera sell fpgas which run opencl, i've only had a cursory look at them though, and i suspect you'd still be doing a lot of work specific to the fpga to get decent results. That is a pretty safe bet, just like with Xeon Phi. You can theoretically run unchanged CPU code there, but it will have hilariously bad perf.
|
|
#
?
Oct 5, 2017 13:49
|
|
|
atomicthumbs posted:somewhere in my boxes of boards i have what was an extremely expensive PCI 3D accelerator made by evans and sutherland. i oughta build a 90s pc workstation someday i have a sun workstation that came with a 3dlabs wildcat card, as a cheaper alternative to Sun's own graphics. it is very, very slow, even by circa 2002 standards. i'm not sure how intergraph or 3dlabs stayed in business so long was OpenGL CAD/CAM just that fuckin impossible on consumer cards?
|
|
#
?
Oct 5, 2017 15:14
|
|
|
in that era i think the intersection between what cad/cam needed and what was accelerated on a consumer graphics card was almost empty yeah. hopefully someone knows this better than i do though, i don't dare effortpost on the matter, but believe it comes down to the consumer option of a simple chip that can have (poorly) textured and lit screenspaced triangles thrown at it by cpu at brisk (in total pixels/texels getting drawn) pace, versus cad/cam chips dealing with abuse of huge displaylists of stippled lines and simplistically rendered but hugely detailed meshes that need gpu support for the transforms and somewhat specialized rendering otoh i learned opengl on one of those wildcat suns back in the day, and as i recall they were indeed pathetically slow, but i think they served entirely as a mostly pointless entry-level thing
|
|
#
?
Oct 5, 2017 15:30
|
|
|
Shinku ABOOKEN posted:as i understand nvidia and other companies dont manufacture their physical chips. what prevents the manufacturers from copying the design themselves? now that we're at like 10 nm fabs the foundries have to be spend more money and be bigger, so if TSMC ever gets caught passing poo poo around everyone will immediately pull their business and they'll be bankrupt in the span of about 7 hours.
|
|
#
?
Oct 5, 2017 15:40
|
|
|
Notorious b.s.d. posted:i have a sun workstation that came with a 3dlabs wildcat card, as a cheaper alternative to Sun's own graphics. it is very, very slow, even by circa 2002 standards. i think i still have a 3dlabs oxygen card circa 1998 or 1999 somewhere that i got when a university was throwing them out. it had two glint chips and dual monitor support in 1999 so it probably more than doubled the price of the machine it was installed in. by 2001 they were already headed for the dumpster because they were so outdated and slow.
|
|
#
?
Oct 5, 2017 15:42
|
|
|
Farmer Crack-rear end posted:wasn't matrox notable for having video cards with three outputs back when most still had two at max? I was legit excited for getting the parahelia for exactly this reason, except then the marquee game with support for it got cancelled (and then much much later resurrected and rebranded), and the reviews called it garbage. I could have had multiple monitor life over a decade ago. Alas.
|
|
#
?
Oct 5, 2017 15:46
|
|
|
Volmarias posted:I was legit excited for getting the parahelia for exactly this reason, except then the marquee game with support for it got cancelled (and then much much later resurrected and rebranded), and the reviews called it garbage. the big question is: did you cast your head?
|
|
#
?
Oct 5, 2017 15:55
|
|
|
Cybernetic Vermin posted:the big question is: did you cast your head? No because ultimately I never got it. Maybe it would have.
|
|
#
?
Oct 5, 2017 16:02
|
|
|
|
| # ? Jun 8, 2024 13:55 |
|
|
i have an original Ageia PhysX card that i just found lying in a desk at my last job. i took the fan off and polished up the shiny chip to a mirror finish just for fun but idk what else to do with it i guess i could try very carefully desoldering it and putting it in a little frame?
|
|
#
?
Oct 5, 2017 17:50
|
|