Mantle Multi GPU's Memory

Why not this been talked about?
Mantle enabled games will allow Multi GPU's to work in meny different ways to before.

Now we have Crossfire each GPU have 3GB = 3GB total, With Mantle this all change's Two GPU's will have there own Memory so two x7970 = 6GB Vram

See this video here
Watch
http://youtu.be/N_6CAneoW-0?t=23m43s [youtu.be]

Key Wording

4 GPU's Each with Different MEMORY

Next
Mantle will also allow each GPU to work interdependently of each other, for example One can be Rendering a scene and the other can be doing a more demanding task
Physics, AO, GI or AA like Super sampling.

 

i'd first like to see mantle actually work in a game open to the public instead of a slide or presentation.

then some nvidia support would be neat
but seeing how AMD is marketing this as using there architecture to the fullest i doubt that until some master hackers try to get it working : /

 

Sounds great, but will we ever see it, im sick to bloody death of hearing about it now, Mantle this, Mantle that, they've been spouting the same things, and showing the same slides for months now, just get the bugger out!

 

Battlefield 4 is the issue here. Dont expect wonder's from that game
Mantle starts with Thief in my book.

 

It does not take a "Master Hacker" to write driver for graphical api. All it takes is decision to hop in made by nV.

 

drivers are done, thats not the problem.

Dice has to implement the API calls in their engine.

THATS the problem.

 

not really

dice already has that.

they just need to give it to us in a patch

and they didnt

thats the problem

 

Too much talk and no show.

 

the patch IS the engine update.

 

Mantle video update from starswarm.. The difference in this " demo " is xtreme due to the extremely high AI, physic, objects counts ( 4000 to 5000 units ). but worth a look for see the difference in fps between dx to mantle.. ( basically you have ~ 10fps in DX, when you hover between 40 to 85fps in mantle ) ( change is made at ~ 0:45 )

http://www.engadget.com/2014/01/14/oxide-star-swarm-real-time-strategy-mantle-demo/

 

This is great news. If your using multi gpus then theres no real reason to worry about memory limitations now. This is good for people on 3GB cards that are near maxing them out. This should dramatically increase framerates in memory hogging games.

 

Not really big benefit for games. Can you imagine how long it will take to give GPU1 2GB chunk of data stored in VRAM2 vie PCIe 3.0 x16?

Such feature is just part of AMD's HUMA and will be very valuable to us in future, like 2017/18 with different board architecture or much higher bandwidths.

 

About an eighth of a second. PCIe 3.0 delivers almost 1GB/s per lane. But even with a new PCI bus it will always be faster to feed the gpu with dedicated memory. The advantages offered by HUMA are more geared at the software side and reducing the overhead of moving data between gpu and cpu memory space - now a pointer can be passed and the underlying memory controller can handle the moving rather than relying on a separate software driver. AMD is doing serious work to make the GPU no longer second-class to the CPU in terms of accessibility. The benefits this offers to games are hard to fathom now because of how fundamentally different this will allow future rendering engines to be programmed.

 

All this is talk is only marketing untill we can test it on our PCs.

 

Thats not how it works. Current API's treat 2 GPU's as 1 and its down to the drivers to distribute the workload. Mantle will be able to see two separate GPU's and spread the workload to each gpu accordingly. It will probably need a lot of optimising at game engine level, but its certainly an interesting development.

 

BF4 is confirmed to use AFR in mantle, but with better scaling than previously seen. According to Johan Andersson, AFR still offers higher levels of scaling than we are likely to see if entire GPUs are dedicated to only specific parts of the rendering pipeline, as it is difficult to ensure that both GPUs have equal workloads that way.

https://twitter.com/repi/status/424221080010510337

https://twitter.com/repi/status/420299644954894337

 

In OP's theory 1 + 1 = 2 and not 1 as we have it now. Such feature is very possible.
And my reply points towards weakest point of such feature, and is written in way to make people think for themselves, take calculator and divide few numbers to find how bad impact would such method have on performance.

I know that my english is at level of 4 years old. And due to my unwillingness to use g+trans not even native speaker can get a glimpse of what I wrote.
Maybe OP wrote it in way you may understand to what my reply was
Or you can read Haz_mat's post beneath mine as he managed to "decrypt" post of mine.

 

So I end up with DDR 4 ram one socket = 1 ram channel and GPU will be the same 1 pcie3 = 1 whatever that is named for 4 channel of pcie3 whatever ? Mmif this is accurate we might need more CPU . no wonder AMD is getting ready a 16 core CPU . sadly again AMD probably ignored ati view on this .how many time will AMD screw up . and at I has to glue the broken piece.

 

That 16core thing is just opteron iteration. It will have improved ipc and will be one silicon (saving of transistors/power draw).
Yes, people who will want to get AMD system will likely soon go for their server parts.
As I would even now if I wanted pure multithreaded CPU performance as it's very affordable compared to intel. But single threaded perf. still matters for games and will continue to for next 4~5 years.

 

i have 3 x HD 7970 = in Trifire but only 3GB of Vram is used.it will be good if AMD make something that all 3 GB Vram on all 3 crads = 9GB is used so Vram Limitation is no Problem anymore...i dont think Mantle can enable is ?