What's shared memory in GPU for?

[4K_240hz_MLED]

I have like 7GB11GB SRAM in my GTX 1050 but only 3GB is for VRAM.

[Elucidator]

https://en.wikipedia.org/wiki/Static_random-access_memory

[4K_240hz_MLED]

Originally posted by Elucidator:

https://en.wikipedia.org/wiki/Static_random-access_memory

That's not what I meant. Also the recent consoles can use either SRAM or VRAM as both SRAM and VRAM, so just stop.

[Blaagh]

no you have 3GB VRAM in your card.. shared memory means it can use RAM on your mainboard too.

that helps so the computer does not crash once the VRAM is full.. but it is much slower so you should try to stay below 3gb

[Elucidator]

Originally posted by 20th Year Otaku:

That's not what I meant

The statement you gave was confusing.
Shared Memory is what it says it is. SRAM is different.

Shared Memory is not something that is in your GPU. WildHog explained it casually, but well.
https://en.wikipedia.org/wiki/Shared_memory
Shared Memory is System Memory that can be accessed by multiple processors that are the same memory blocks. (or, it can also be a memory block that can be accessed by multiple programs.)
A GPU, like the CPU is a processor that is on a different device, but this processor can, with Shared Memory enabled, now too access some of your installed RAM and Virtual Memory.
Considering that last part, don't use it too often.

Shared Memory is limited by limitations set by software. You can assign how much system memory can be accessed by both your CPU and GPU at the same time basically, or you should be able to.
GPU usually only needs a bit of VRAM, which is memory on the GPU's board (not the motherboard) that is specifically designed for graphics handling. Note that some software can change the default behavior.

Edit: Considering

Originally posted by 20th Year Otaku:

so just stop

part of your reply. Don't be so quickly offended. People aren't trying to attack you, but rather trying to figure things out and help you.

[4K_240hz_MLED]

Originally posted by WildHog McReady:

no you have 3GB VRAM in your card.. shared memory means it can use RAM on your mainboard too.

that helps so the computer does not crash once the VRAM is full.. but it is much slower so you should try to stay below 3gb

But since I upgraded to 16GB dual channel like almost a month ago, wouldn't it mitigate the slower bandwith of the SRAM? Theoretically going from 8GB DDR4 2667MHZ single channel to 16GB dual channel should make it perform as fast DDR5 5334MHZ and 128 Bit single channel right? I'm not sure about the GB/s that SRAM has but theoretically it should almost be on par with GDDR5 right?

[4K_240hz_MLED]

Originally posted by Elucidator:

Shared Memory is System Memory that can be accessed by multiple processors that are the same memory blocks. (or, it can also be a memory block that can be accessed by multiple programs.

I already know what shared memory is in terms of integrated graphics. I didn't know the same could apply to dedicated graphics (except on gaming consoles). Especially since in the Nvidia control panel it doesn't state if the shared memory was DDR4 or GDDR5 like AMD or Intel would.

[Elucidator]

Hm... I did some deeper digging. (I don't know everything either, so yeah.)

Shared Memory is basically decided by Windows. It measures how much installed System Memory you have available (which includes Paging (virtual memory) ), then takes half of this.
When you have a dedicated gpu (or graphics card), normally it will only start using shared memory when it runs out of dedicated video memory (vram).
Unfortunately, it seems I was wrong. You can't lower Shared Memory, unless you also lower System Memory and so you can't prevent the system from using part of your system memory for graphics data. (at least not in Windows)

Considering your question. GDDR5 is obviously a ton faster than DDR4. Not only is it optimized for graphics, but DDR5 is twice as fast as DDR4.
Dual Channel means the system's memory controller can send and receive data (of the same amount) twice to RAM at the same time, because it knows that the slots in at least one dual channel has the same RAM specifications. Therefore its twice as fast with handling bits. (You could see it as adding an extra lane, an extra road to the same destination.)
However dual channel doesn't overwrite the max speed at which bit processing is handled by the memory controller, which are described in the RAMs specs. It also means the MegaHertz (frequency, the speed per bit channel) remains the same.

So, it means with certainty, if you ever get to the point your GPU requires your system memory, it will slow down. Your FPS would at least drop to half. likely a lot more since your processor will suddenly be in wait till your GPU is done putting stuff in RAM before it can. Your memory controller will be very busy at least.

Also should paging be enabled, you can probably see the lifespan of your SSD drop very quickly.

edit: To me SRAM is still the installed RAM on your motherboard, although its more likely SDRAM, a.k.a. dedicated system memory. I suppose you're refering to shared memory with sram now. Anyway that is what got me confused before, so I send a link at first to show what sram is, in case you were reading it in some status window. I didn't intent to become someone you'd dislike at least.

[4K_240hz_MLED]

I saw on Crucial's website that multiple RAM sticks of lesser bus width in multi channel mode is faster than a single RAM stick with a larger bus width. I also already knew GDDR5 is up to twice as fast as DDR4 because of benchmarks I saw for GT 1030 DDR4 and GDDR5 variants. I highly doubt DDR5 SRAM is 2x faster than DDR4 though.