Why do NVIDIA's mobile GPUs have comical levels of GDDR5 VRAM?

[sk3tch]

I was just looking at the Clevo ad in the latest PC Gamer mag - 8GB of GDDR5 for the GTX 880M...which is what...a desktop GTX 770/680 in terms of power? You see it all over the place in the mobile side - yet with desktop GPUs (where it could actually be utilized for 4K and more as there's more power and capability due to more space/power/better cooling) breaking 4GB is a rarity.

I understand a lot of it is marketing - but then why not toss that on the desktop side, as well? AMD doesn't seem as guilty as NVIDIA on the mobile side of overdoing VRAM, too...which is interesting.

 

[harmattan]

Historically, extra VRAM was included on mobile cards since the thought was the card would need to have a longer lifespan than a desktop card (which can be replaced often... mobile cards, not so much). Now it's added since the same chip is refreshed several times for mobile use and nV and AMD need another draw and justification to rename the card than just a small mhz bump -- regardless of whether that extra VRAM will ever be useful.

In the case of 880m, nV "refreshed" the 780m using the exact same chip, but only clocking it in bios a bit higher. It would be as if, in the desktop world, nV took a GTX 770, increased the core speed 75mhz and called it "GTX 870". They needed some other differentiator to rename the card (and remain relevant) - useful or not - so they added another 4GB VRAM.

It's worthy noting what was probably the most egregious example of this mobile card rebranding scheme: the 8800m GTX to 9800m GT. Both were the same exact card and chip, but renamed. Dell and Alienware (back when they were separate companies) were even selling both cards at the same time, but the 9800 GT at a $100 premium -- rebranding and marketing at their finest At least nowadays, they are adding more memory to charge you more...

 

[Unknown-One]

You generally see "comically large" banks of video RAM on cards that are designed to operate on limited buses. For example, it's fairly common to see modern cards, designed for the legacy PCI bus, pile on the RAM.

This is done so that as much data as possible is available to the GPU in real-time. The goal is to reduce cache-misses (which result in data being pulled across the bus from system RAM), thus avoiding the largest bottleneck. You end up with longer loading times, but less stuttering and hitching.

It's possible these mobile GPU's are connected with only one or two PCIe lanes, in which case the additional memory could come in handy.

 

[misterbobby]

Well the gtx880m has 4gb of vram standard but I guess any laptop company can go with an 8gb version.

 

[xorbe]

Because noob buyers equate big vram numbers as an indicator of laptop graphics performance.

 

[evilsofa]

I understand a lot of it is marketing - but then why not toss that on the desktop side, as well? AMD doesn't seem as guilty as NVIDIA on the mobile side of overdoing VRAM, too...which is interesting.

You mean, like the 4GB GT 640? (EVGA, MSI) The really cheap cards with 2GB or 4GB of VRAM are for certain content creation apps like Mudbox, Mari, Maya, and MAX, where having lots of VRAM to hold large models and/or textures is more important than the speed of the card. They are irrelevant for gaming.

 

[zandor]

Another use for lots of vram is certain GPGPU applications. A friend of mine worked on physics simulations in grad school and at one point ported one over to CUDA, but he couldn't run full size simulations on it since the card didn't have enough vram. That's a pretty niche application of course, and doesn't justify gobs of vram on a gaming laptop.