VRAM on new GPU's?

[t-scene]

I just got a GTX 980Ti. I was wondering why AMD decided to with 8GB on the 390x and only 4GB on their flag ship Fury X.

Seems the 980Ti rests in the sweet spot between the two.  Does the 390x really have the horsepower to make good use of the 8GB?

With the main stream AMD cards having 8GB does that mean developers will make my 6GB obsolete soon?

[kipper]

HBM

[SkilledRebuilds]

I just got a GTX 980Ti. I was wondering why AMD decided to with 8GB on the 390x and only 4GB on their flag ship Fury X.

Seems the 980Ti rests in the sweet spot between the two.  Does the 390x really have the horsepower to make good use of the 8GB?

With the main stream AMD cards having 8GB does that mean developers will make my 6GB obsolete soon?

HBM ver1 is limited to 4GB, ver2 will address this.

[Mohenjo]

4gb essentially equals 6gb with hbm 

[Duskie]

I just got a GTX 980Ti. I was wondering why AMD decided to with 8GB on the 390x and only 4GB on their flag ship Fury X.

Seems the 980Ti rests in the sweet spot between the two.  Does the 390x really have the horsepower to make good use of the 8GB?

With the main stream AMD cards having 8GB does that mean developers will make my 6GB obsolete soon?

Because the Fury cards use HBM memory, which currently has a limit of 4GB. The 390 and the 390x use GDDR5 though.

[ThE_MarD]

Heyyo,

 

4gb essentially equals 6gb with hbm

hahaha... you realize that makes absolutely no sense right?

That's like saying a 100GB SSD M.2 has more storage than a 100GB 7200RPM HDD. No it doesn't. It just moves data with a lot better power and time efficiency.

The only way the 4GB HBM would equal 6GB GDDR5 is if AMD has some magical on-the-fly data-compression.

I honestly don't see how people think a wider/higher bandwith memory makes storage act like it has more storage unless there's some magic trick I'm missing.

http://www.digitaltrends.com/computing/amd-explains-how-high-bandwidth-memory-translates-to-fury/

Please show me where it says "makes storage magically bigger than what we say it is." It says "performance well beyond" and NOT "storage well beyond."

[Mohenjo]

Heyyo,

hahaha... you realize that makes absolutely no sense right?

That's like saying a 100GB SSD M.2 has more storage than a 100GB 7200RPM HDD. No it doesn't. It just moves data a lot more power and time efficiency.

The only way the 4GB HBM would equal 6GB GDDR5 is if AMD has some magical on-the-fly data-compression.

well i meant it in a sense that the memory is so efficient that what gddr5 might need 6gb for hbm would only need 4

 

not literately 4gb equals 6

[Casperian]

well i meant it in a sense that the memory is so efficient that what gddr5 might need 6gb for hbm would only need 4

 

not literately 4gb equals 6

While I see your point, it might be good to clarify that just because HBM is faster doesn't mean that if you need to use 4.5GB there will be no difference, because at that point you have to grab from disk and I don't care how fast HBM is, the disk ain't that fast.

 

EDIT:

What HBM will do however, is improve memory IO. Post processing effects, such as a bloom, use a lot of memory IO so it is likely we will see PostFX not hitting the Fury as hard as it hits other cards, however it might not matter if other bottlenecks might come into play first.

[Zulkkis]

The HBM's bandwidth doesn't really make it equal to higher quantities. The color compression and other tech help in that regard, but nVidia has something similar as well.

 

I've heard that higher versions of directX and just proper compression would bring the memory usage in the ballpark of ~2GB if done properly, even in open-world games, but naturally not everyone is using the newest dX functionalities which even dx 11 has... Generally it's sort of alright, and not alright - It can't be any other way since it is a limitation of the technology, preferably you'd always have more, but on the other hand, the launching prices are not outrageous if the performance is there.

[ThE_MarD]

Heyyo,

 

While I see your point, it might be good to clarify that just because HBM is faster doesn't mean that if you need to use 4.5GB there will be no difference, because at that point you have to grab from disk and I don't care how fast HBM is, the disk ain't that fast.

Exactly! That's the problem I forsee with the AMD Fury potentially having the performance of a GTX 980 ti but with 2GB less of storage. We've seen how good the GTX 980 ti flows through GTA V... but it fills up to 5GB of VRAM with the graphics cranked up on 4K resolutions.

Essentially, the AMD Fury will run into a VRAM limitation wall before the GPU even reaches full load. That is the EXACT same problem I'm running into with my GTX 680 2GB SLI setup. On GTA V? I could easily pass my settings past High with MSAA x4 as my GPUs sit at 70% load... but I'm stuck at 1900MB of VRAM. As soon as I push my graphics higher? My framerate drops to about 10fps at times when my VRAM useage hits 2048 MB, so the 2GB of VRAM my GPU comes with sigh.

Of course, 4GB of VRAM is an issue at 4K resolutions and DirectX 11 / OpenGL... but that's also banking on DirectX 12 and Vulkan delivering on their promise of VRAM sharing and hopefully it working in all titles along-side native support for Split-Frame Rendering.... which may or may not work as intended. There isn't even a Multi-GPU driver for DirectX 12 out yet from both NVIDIA and AMD so I can't just plop in a test of 3DMark's 3D API Overhead test and see what happens... it crashes as soon as I try with SLI on.. so only one GPU works on that test... I'm banking on it working though since I'm running into VRAM limitations already and I'm at 1920x1080 resolution.

[EpicAdom]

I just got a GTX 980Ti. I was wondering why AMD decided to with 8GB on the 390x and only 4GB on their flag ship Fury X.

Seems the 980Ti rests in the sweet spot between the two. Does the 390x really have the horsepower to make good use of the 8GB?

With the main stream AMD cards having 8GB does that mean developers will make my 6GB obsolete soon?

1st gen HBM is limited to 4 gigs, BUT the 4 gigs on the Fury X are FAST AS HELL (in memory bandwidth)

[FairladyZ]

HBM - High Bandwidth Memory

 

The memory is basically stacked into one chip instead of many chips all around the PCB. This is part of the reason the Fury and Fury X will be so short, because there won't be a need for extra length due to lots of memory chips. And HBM has a limit of 4GB as of right now.

 

EDIT: Also, HBM is incredibly fast even though it has a low clock speed of 1000MHz, the extra bandwidth makes up for the lack of clockspeed.

[-rascal-]

They implemented Data & Colour compression on GCN 1.2 -- which was available on the "Tonga" R9-285.

I don't see why they won't have that available with HBM on the Fury(X) and the Nano...

[79 Potatoes]

1-I just got a GTX 980Ti. I was wondering why AMD decided to with 8GB on the 390x and only 4GB on their flag ship Fury X.

2-Seems the 980Ti rests in the sweet spot between the two.  Does the 390x really have the horsepower to make good use of the 8GB?

3-With the main stream AMD cards having 8GB does that mean developers will make my 6GB obsolete soon?

1-The threw some HBM memory at the Fury X. 0.5 freaking Terabytes per second.

2-Not sure what you're trying to say but if you mean if it can allocate the 8 GB then yeah

3-Uh... not related in any way? PS: No.