Why VRAM will never stack in SLI - regardless of DX12

[Cheddle]

VRAM will never stack on PCIe 3.0 - running out of VRAM on one card in DX12 will be exactly the same as running out of VRAM on one card in DX11 - swapping over the PCIe bus will happen

 

The short story:

RAM needs to be fast, the thing connecting the SLI cards together is not fast....

 

The long story:

 

The VRAM on a GTX980 has a bandwidth of 224 GB/s

 

even with PCIE 3.0x16 (the fastest possible slot for a video card today) the RAM access speeds between card 1 GPU and card 2 RAM will be capped at 16GB/s - This is four times slower than an Nvidia 8800gtx from 2006 and 14 times slower than card 1 GPU accessing card 1 RAM.

 

The point behind these wild claims of DX12 stacking ram is to prove that DX12 is very flexible, and enables developers to do more things.... it will NOT make your SLI configuration act like a card with double the ram EVER 

 

(p.s: the above applies unless NVlink is real life, but its not until Pascal and even then its only 60 GB/s with a claimed max of 192 GB/s, OK by todays standards but likley a bottleneck with cards in 2016/2017 with HBM and the likes)

[Lolzious]

Ehhhhhhhhhhhhhhhhhhhhhhhhhhhh

don't kill it for me man

I DREAM

THAT ONE DAY

I CAN HAVE 6GBs OF VRAM FROM A 280X CROSSFIRE

[1Scotty1]

Ehhhhhhhhhhhhhhhhhhhhhhhhhhhh

don't kill it for me man

I DREAM

THAT ONE DAY

I CAN HAVE 6GBs OF VRAM FROM A 280X CROSSFIRE

 

Same here with my 970s, 7GB 8GB for me!

[Swndlr]

man you aint even gonna hope

thats dark

real dark fam

[Godlygamer23]

Same here with my 970s, 7GB for me!

Don't start.

[Godlygamer23]

man you aint even gonna hope

thats dark

real dark fam

Reality is superior to hope.

[Sauron]

You are missing the point... with hypothetical split ram there would be no need for direct data sharing between the cards. Each card would load only what it needs, with SLI only making sure they aren't processing the same part of the scene. If anything, a hard drive could start becoming a bottleneck soon, but that's another matter entirely.

[FruitBasketSilex]

You are missing the point... with hypothetical split ram there would be no need for direct data sharing between the cards. Each card would load only what it needs, with SLI only making sure they aren't processing the same part of the scene. If anything, a hard drive could start becoming a bottleneck soon, but that's another matter entirely.

A hard drive is already a bottleneck to me, and all I do is limited multitasking around windows and booting (already into OS, waiting for programs and services to load)

[Guest]

Don't start.

Are you serious?

[1Scotty1]

Don't start.

Sorry, bored at work

[Godlygamer23]

Are you serious?

Dead serious.

[The Crazed Child]

Same here with my 970s, 7GB for me!

ayyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyyy

[Guest]

Dead serious.

People shit on AMD, but they can't on Nvidia?

Calm down, you need it.

[FruitBasketSilex]

People shit on AMD, but they can't on Nvidia?

Calm down, you need it.

Pretty sure if someone shit on AMD in this thread he would've said the same thing.

Don't be an ass just to shove someone down.

[Guest]

Pretty sure if someone shit on AMD in this thread he would've said the same thing.

Don't be an ass just to shove someone down.

No, he wouldn't.

How would you even know.

Also, I am not an ass. Told him to calm down.

[Stefan1024]

VRAM will never stack on PCIe 3.0 - running out of VRAM on one card in DX12 will be exactly the same as running out of VRAM on one card in DX11 - swapping over the PCIe bus will happen

 

The short story:

RAM needs to be fast, the thing connecting the SLI cards together is not fast....

 

The long story:

 

The VRAM on a GTX980 has a bandwidth of 224 GB/s

 

even with PCIE 3.0x16 (the fastest possible slot for a video card today) the RAM access speeds between card 1 GPU and card 2 RAM will be capped at 16GB/s - This is four times slower than an Nvidia 8800gtx from 2006 and 14 times slower than card 1 GPU accessing card 1 RAM.

 

The point behind these wild claims of DX12 stacking ram is to prove that DX12 is very flexible, and enables developers to do more things.... it will NOT make your SLI configuration act like a card with double the ram EVER 

 

(p.s: the above applies unless NVlink is real life, but its not until Pascal and even then its only 60 GB/s with a claimed max of 192 GB/s, OK by todays standards but likley a bottleneck with cards in 2016/2017 with HBM and the likes)

 

As long as atlernative frame rendering is used your statement is true (AMD and Nvidia use this today). But if they switch to split screen rendering, probably not all textures needs to be stored in both GPUs....

[FruitBasketSilex]

No, he wouldn't.

How would you even know.

I don't know. And neither do you.

I do however, have good faith that he'd do the right thing.

[Godlygamer23]

People shit on AMD, but they can't on Nvidia?

Calm down, you need it.

The 3.5GB VRAM thing has been on here for a while. Thread after thread, joke after joke and for the record, I criticize all companies, including NVIDIA. How can I calm down if I'm already calm?

[Guest]

I don't know. And neither do you.

I do however, have good faith that he'd do the right thing.

Your faith is misplaced.

Maybe if this forum was unbiased. But the majority likes Nvidia way more. And that is okay, but don't take jokes dead serious.

You're defending weird behaviour.

[Delusional Phil]

What has been seen cannot be unseen ... 

You broke a lot of people's dream there ...

[Guest]

The 3.5GB VRAM thing has been on here for a while. Thread after thread, joke after joke and for the record, I criticize all companies, including NVIDIA. How can I calm down if I'm already calm?

Yeah so?

AMD gets heat jokes. Ubisoft gets digging jokes.

Just ignore them if you don't like them.

[Godlygamer23]

AMD gets heat jokes. Ubisoft gets digging jokes.

Those jokes are old as well.

[FruitBasketSilex]

Your faith is misplaced.

Maybe if this forum was unbiased. But the majority likes Nvidia way more. And that is okay, but don't take jokes dead serious.

You're defending weird behaviour.

I've seen plenty of people that like AMD GPUs. I've seen plenty of people reccomend AMD GPUs instead of Nvidia ones.

However, nvidia does have some pretty nice shit and quite recent, with low power draw, and AMDs big thing right now is "Hey, throw money at us while you wait for us to finally release a new product" (in reference to the (possibly) new FX chips, zen, and 300 series GPUs)

[Imabigmac]

Okay, but what if the access to the memory was done through a proprietary connector, I'm thinking of a wider, much faster SLI bridge acting as the the memory bus. 

[Unlikelyjoker]

Your faith is misplaced.

Maybe if this forum was unbiased. But the majority likes Nvidia way more. And that is okay, but don't take jokes dead serious.

You're defending weird behaviour.

People make me sick.

70% of the forum are just biased Nvidia owners.