Jow Forums comes up with a better design for a Modular Graphics Card than pic related

do not* do as good of a job

>modular-graphics-card-nvidia-radeon-design-2018-dave-delisle-davesgeekyideas-1.jpg
>nvidia-radeon
Wut

>how are you gonna mount the heatsinks, fans and shroud properly?
with screws, like it's done on today's GPUs
>Usb to hdmi/dp causes latency issues so its much better to just solder the hdmi and dp ports straight into the pcb
who said it's going to use USB?

>memory modules
Silly, the HBM memory is on chip, next to the GPU die

>Usb to hdmi/dp
Modern cards that have a USB port use USB-C, it's native DP, no conversion or extra latency.

you're a dumbshit. standard DRAM sticks do not reach the necessary bandwidth targets for a GPU to function at full performance, not to mention the massive hits to latency you're going to take with adding more complexity to the GPU's memory controller to support whatever weird fucking configurations consumers want to put in the slots.

>memory sticks
Just fuck my latency up

This a million times. Model-specific fans are worse than the holocaust.

>meanwhile in freedomland, use whatever fucking fan you want, and replace it when it rattles or dies

Attached: morpheus-13-1280x1024.jpg (1280x800, 183K)

If I were to design my own Modual Graphics Card, I would use PCie 5.0 x12 (47.256 GB/s) for the VRAM models instead of DIMM, let alone SO-DIMM.

If you don't mind the vram and gpu being on the same board and just want cooler and i/o to always stay the same, consider MXM to PCI-E adapters.

Attached: 7fb202cfe3.jpg (500x383, 55K)

The "central SoC" IS the GPU. Everything around it, except for the BIOS (cheapo SPI flash) and memory modules (which are embedded with the GPU if it's HBM), is there to convert and regulate voltages, or just to interface with something else.
Why do you think even motherboards, which ARE modular, only support 2 or 3 generations of CPUs and you still get a heavy bottleneck with the later gens?

Outdated as fuck

>Why do you think even motherboards, which ARE modular, only support 2 or 3 generations of CPUs and you still get a heavy bottleneck with the later gens?
Thank God Ryzen exists, you know. Literally SoCs with power in, SPI and PCIe lanes. Chipset is only use for extra shit like networking and audio. Directly connects the GPU and storage to the CPU. No problems with upgrading or bottlenecks latter on.

>x8 + x8 SLI
kek

Seems like you forgot the entire radiator block.

Also modulatority increases size and cost due to neef for connectors and shit.

Well first off
>pic related
Secondly
>how are you gonna mount the heatsinks, fans and shroud properly?
The same way it already works for CPUs, GPUs, etc.
>gpu fans are different than case fans, (as shown in the picture) case fans do as good of a good job as gpu fans
I guess you also expect the LGA socket and SO-DIMM slots to not be only a mockup... Now let me blow your mind: Take a decent, branded case fan, break the thin supports between the center and outer armature, and put it next to a current non-blower GPU fan. Can you see the similarity now?
>complications with the heatsinks, heatpipes, fans and shroud/cover would cause overheating
The same would go for CPUs, just isn't true.
>thirdly you cant just mount gpu dies on a pcb like its a cpu
Then I guess they've been doing it wrong all this time, the die should be floating in a vacuum? If you mean it can't be socketed, I bet you've never seen a BGA CPU. It works both ways, but sockets are bulkier, dissipate less heat and create more resistance.
>Usb to hdmi/dp causes latency issues so its much better to just solder the hdmi and dp ports straight into the pcb
Literally wut? The ports are all soldered into the PCB, but anyways both HDMI and DP already interface via external ASICs.

Attached: 1510700527241.jpg (720x687, 67K)

>t. seething

Seems like a fine idea to me.

Attached: 1484624512296.jpg (194x259, 5K)

Whoever made this picture clearly knows nothing about GPU architecture. Bus width makes it impossible to use memory dims in GPUs.Memory speed and power delivery requirements vary DRASTICALLY gen to gen.

>Bus width makes it impossible to use memory dims in GPUs.
Even if DIMMs were practical for VRAM, you are going to need quite the wide load.

Attached: graphics-card-components.jpg (680x458, 106K)

Is that supposed to be for pic related?

Attached: 0082_01_l.jpg (800x800, 87K)

good sir please send morpherus 2 to me as a gift, do the needful

>no heatsink
>fans blow DIRECTLY ONTO PROCESSER
you must be 18 years old to post here.

Attached: bitch.png (124x216, 29K)

umm sweaty where are the rgbs?

What's the point of modular GPUs? I'd rather have a modular cellphone. Now THAT would be undeniably based.
>pick from plastic, metal, ceramic or glass
>pick your color
>pick your storage capacity
>pick the option for a jack or expandable storage
>pick the size of the battery
>pick the cameras
>pick your choice of display; flat or curved, notched or with bezels
>pick if you want front firing speakers
woooooo boy I want this NOW