Its about time gpus made the switch from pcie to a dedicated

File: 1494331810675.png (760KB, 960x640px) Image search: [Google]

760KB, 960x640px

>>60580310
You mean like the dedicated socket they used to have over 9000 years ago?

>>60580310
AGP sucked.

>>60580310
>dedicated socket
>A
>G
>P

how young are you? we moved /away/ from gpu-specific slots to pci-e

>>60580430
what issues did they face? i suppose pcie fixes the issue when using multiple gpus in a sense. If thats the case why havent cpus switched to a similar interface? Also why dont we have better gpu cooling solutions yet? that seems pretty funny that if pcie fixed socket issues it didnt account for cooling issues

>>60580463
PCI was getting old, and there needed to be a replacement, but AGP was too gpu-specific to be used for anything else
so they made the PCI replacement suitable for gpus as well, so now we just have PCI-E

File: [HorribleSubs] Gabriel DropOut - 05 [720p].mkv_snapshot_17.59_[2017.02.06_14.16.09].jpg (138KB, 803x720px) Image search: [Google]

138KB, 803x720px

>>60580310
We already have meme M.2 slots, we don't need more memes on our memeboards.

File: a.jpg (101KB, 1484x1030px) Image search: [Google]

101KB, 1484x1030px

>>60580463
>seems pretty funny that if pcie fixed socket issues it didnt account for cooling issues
PCI-E came out in 2004, gpu's in 2004 didn't have "cooling issues"
pic related is what a top end graphics card looked like around when PCI-E was released

>>60580310
>CPU
>cooling one single die
>GPU
>covers the whole chip which is far longer than a CPU
GPUs are pretty much a whole computer on their own, so cooling would still be expensive as fuck (presumably)

>>60580556
m2 is actually a useful thing though.

>>60580577
fair enough, so its time for a better solution.

>>60580580
not expensive, a cpu cooler (pick any, even a 212) will cool better than a stock cooler by a large margin. Its mounting thats an issue, and using more than one gpu thats also an issue. Thats why i though a dual gpu socket would be better for cooling.

>>60580654
but there's more stuff to cool. remember the EVGA 1070s that exploded from poor VRM cooling? maybe i'm not grasping what i need to but if you only cool the chip, you're kinda fucked on the rest so standard CPU coolers wouldn't work and we'd be in the same shit from a different set of assholes

>>60580654
>so its time for a better solution.
maybe
perhaps it's not such a crazy idea these days to consider a dedicated spot on the motherboard again for the gpu, like a cpu
they're used for much more now than ever before

i say "again", because for a good while integrated gpu's were integrated into the northbridge, before being moved into the cpu

>>60580463
>what issues did they face
well for one your AGP graphics card only fit one specific slot so it better not interfere with your northbridge cooler, your network card or your sound card.

>>60580556
The question is, when do we get m2 gpus?

File: a.jpg (38KB, 500x417px) Image search: [Google]

38KB, 500x417px

>>60580707

File: LOL.gif (168KB, 300x168px) Image search: [Google]

168KB, 300x168px

>>60580707
>>60580732

File: ASRock-Z270M-STX-MXM-angle1_678x452.jpg (78KB, 678x453px) Image search: [Google]

78KB, 678x453px

>>60580732
MXM is the future

>>60580687
yes theres more to cool, but have you seen the size of a cpu? its tiny compared to the coolers area of effective cooling, which fits over a full gpu

>>60580704
so we create a standard interface in the same way pcie is. a situation where the limits of gddram and bandwidth are so ridiculously high we wont be bottlenecked by the lanes for some time, at which point we release v2

>>60580808
What's the point of creating a new pcie? GDDR5 bandwidth is a problem, pcie bandwidth is not.

>>60580844
so that the gpu has direct access to the ram, that lets you offload easier and encourages devs to focus on a single ram standard to develop for. The main thing is gpu computing plays a signicant role in almost everything now days, to the point where the cpu is usually an afterthough.

>>60580895
The GPU has its own RAM and its far faster than the CPUs RAM. A GPU wouldn't gain anything by having access to the RAM. And I'm pretty sure a one RAM standard wouldn't work because then you'd have your CPU stalling while your GPU fetches data, or vice versa.

>>60580808
fair enough, never thought of that

File: a.jpg (72KB, 640x480px) Image search: [Google]

72KB, 640x480px

soon someone's gonna ask why cpus aren't on slots--
>>60580463
>why havent cpus switched to a similar interface?

ah, nevermind

pcie is fine, we just need 4 slot cooler designs at this point. kinda like blower style cooling, but with axial fans moving the heat out of the case - like tower CPU coolers.

>>60581071
it would, in fact i believe thats what the ps3 and 4 do.

>>60581256
and make it so theres a flippable gpu so that theres a card in slot 1 and 4 and between them is a large heatsink with a fan on each end.

>>60580310
Uhh they have access to the RAM as it is now. They can even communicate peer2peer over PCIe.

>>60580798
I would be cool with this, if PCIe was on the other side of the board.

>>60581636
>it would, in fact i believe thats what the ps3 and 4 do.
So you want to bring PCs back to console tier performance? Why not buy a CPU with integrated graphics and enjoy your compute?

>>60581722
wasnt it a trick done to increase performance with such shit hardware?

>>60581712
is it direct access? i thought it was via cpu

>>60581985
Oh, yeah it is through CPU. It's a graphics API feature.

Vega may have direct access, though, it looks like. It looks like they'll have access to other GPUs VRAM in other PCIe slots as well.

>>60581722
Consoles perform well for their given hardware.
Besides targeted optimizations to specific hardware, a big reason why they perform better than you'd like they do on paper is that they can share RAM via the CPU and GPU.