During its GDC 2017 event, AMD showcased the benefits of its High-Bandwidth Cache Controller. According to the red team, this new memory architecture allows Vega GPUs to do a number of exciting new things that its predecessors can’t. High-Bandwidth Controller Cache can handle memory traffic in a vastly more efficient fashion, but also significantly cuts back on wasteful memory allocations.

In order to showcase the benefits of High-Bandwidth Controller Cache, AMD showcased Deus Ex: Mankind Divided’s in-game benchmark running with and without High-Bandwidth Controller Cache. AMD limited the amount of VRAM that was available for this application to 2GB (so that it can better showcase the benefits of this new memory architecture).

As the red team claimed, Deus Ex: Mankind Divided offers 50% performance gains in average framerates and 100% performance gains in minimum framerates.

It’s also worth noting that the tech demo was running on a Ryzen + Vega combo PC system.

This new memory architecture will definitely help gamers overcome VRAM limitations, especially when gaming at really high resolutions (such as 4K resolutions). Do note the GPUs should be powerful enough to handle games in 4K, and this feature won’t magically improve your experience if you are not VRAM limited.

Source: DSOG - http://www.dsogaming.com/news/amds-vega-10-high-bandwidth-cache-controller-aims-to-improve-performance-by-almost-100-minimum-fps/

Seems pretty insane on how they went about accomplishing control of the cache as well as being able to give eprf gains that big, if anything it had me going:

Amd is certainly ticking all the right checkboxes. Hopefully it all turns out to be legit

AMD are doing alot of smart things lately.

Ryzen and Vega are def. looking pretty good.

Aye, I'm hoping so myself, because if those gains are good around the libraries out there, then I can easily see myself going for a full AMD build before years end.

If they had to limit the VRAM to show the effect I'm gonna assume it will have little to no effect with a good amount of RAM.

edit: yeah it already says it in the text.

So in what kind of fringe cases will this work? GPUs should run into the computing limit way faster than the memory limit in games. So it's probably good for video editing and stuff?

They did mention some time back that they were aiming for gaming enthusiasts as well as streamers and video editors. I guess they want to aim for at least 3 of those bases which isn't that bad to aim for. 

Probably pretty neat if your gaming GPU is also great at graphic stuff.

Also I forgot that those who stream are very likely to record audio+video while streaming their gameplay, so it's be considered a bonus to those people like Jacksepticeye for example. 

If only my ISP had decent upload speed, the combination of Ryzen+vega would make for a good gaming and video editing/streaming setup. 

I read somewhere else that Vega will be priced around GTX 1080 levels. Someone that buys such expensive cards is more than likely to also have an Ultrawide or 4K monitor. Or an Eyefinity/Surround setup.

On topic: What AMD is promising is very impressive, but it remains to be seen if those gains will happen with every game (after all, Deux Ex is an AMD game). Still, more stable framerates are always welcome.

God dammit. Why can't the explain how they achieved what they did with this cache? I want details, not claims.
What kind of  cache is it? Is it using existing DRAM on the GPU? A proper high-speed cache on the GPU itself? Is it powered by combustible kittens from outer space?

***

It's use however is going to be extremely limited.

Low-End GPU's already have more memory than they could ever possibly hope to use...
The mid-range could see some benefit, especially when manufacturers sell a version of a GPU with less memory to save on costs. (Like the Radeon RX 480.)

The High-End typically always has enough memory to make such a feature worthless anyway, with a few-edge case exceptions like Fury and Fury X, but that was because it was using such a new memory technology. (HBM/Stacked memory connected via an Interposer.)

If this "cache" takes up transistors and thus drives up costs and it's benefit is only in low-memory situations, then I would rather they didn't bother and just threw more GCN pipelines in.