AMD Explains DX12 Multi GPU Abilities

WYP

News Guru
AMD Explains the ins and outs of DX12's Explicit Multi GPU Abilities. SLI and Crossfire is going to get much more viable moving forward.

28050336223l.jpg


Read more on the benefits of DirectX 12 on multi GPU setups here.
 
Last edited:
Some of these elements were in mantle as well! AMD should know how they will optimise the drivers for DX12 (hopefully!)
 
I hope the multiple vram pools get used more than i think they will,anyway DX12 has me a little excited
 
Biggest word to note in these slides is "potential". You'll only see these benefits if the developers feel that it's financially feasible.
 
I hope the multiple vram pools get used more than i think they will,anyway DX12 has me a little excited

Never going to happen, PC hardware is no where near up to it.

There is no way the PCI-E 3.0 system is anywhere near quick enough to handle the amount of data required.

Biggest word to note in these slides is "potential". You'll only see these benefits if the developers feel that it's financially feasible.

+1

Remember you can err stack memory in Mantle and look what happened there, nothing.:D
 
Very nice write up except for the following
One of the first features AMD is looking to show off is the ability to use an APU and a GPU together, allowing the APU's iGPU to do some of the smaller tasks,

why are you putting this as a new feature when amd had this technology for a while now, it's called "Dual Graphics". All am saying is seems a little misleading but besides that is all good
 
Very nice write up except for the following


why are you putting this as a new feature when amd had this technology for a while now, it's called "Dual Graphics". All am saying is seems a little misleading but besides that is all good

AMD dual graphics is the APU and the GPU trying to do AFR, it doesn't work that well.

This is different, assigning specific things to the APU to help the GPU out. This will work with both lo end and high end stuff, not just a puny GPU for AFR.
 
Very nice write up except for the following


why are you putting this as a new feature when amd had this technology for a while now, it's called "Dual Graphics". All am saying is seems a little misleading but besides that is all good

Correct me if i'm wrong, but dual graphics isnt the same as offloading lighter workloads onto a less powerful GPU/IGPU. Dual graphics was just a poor mans crossfire.
 
AMD dual graphics is the APU and the GPU trying to do AFR, it doesn't work that well.

This is different, assigning specific things to the APU to help the GPU out. This will work with both lo end and high end stuff, not just a puny GPU for AFR.

Ok thank you for the clarification, very well written article looking forward dx 12
 
I hope the multiple vram pools get used more than i think they will,anyway DX12 has me a little excited

It is very nice to sit here and think about and dream of what might happen with the vram pools. While i am with you on hoping it is taken advantage of a bit more than i am thinking. I do think it will be awhile till the developers actually do utilize it more. While we might not be where the BUS is able to manage the data to fully take advantage of memory pools, I still think it will be a noticeable improvement with what we have now.

This gets me more excited to buy a 300 series card (and project cars) but makes it more difficult when I start thinking of how many to buy and which ones.
 
I think currently with the cards we have now all that data could easily be able to go through a single PCIe 3 x16 slot. That is why AMD have Xfire going through the bus, it's faster and there's room available. The newer cards with HBM when they get huge increases in total memory bandwidth it may become more of an upcoming issue, I think the only problem would arise is when you have two AMD GPUs with HBM memory on one PCB that it may become and issue.
Just remember though guys, DX12 is almost entirely benefiting the CPU. This new GPU stuff just expands and improves their capabilities rather than solely increasing their performance. individually. Pretty much removing the CPU as a limiting factor with this new API it just allows it to feed the GPU faster, many with draw calls, so in theory it will become faster but more often than not you should become GPU limited rather than CPU. Which is how it should be. I'm just trying to explain it how i've seen SPS talk about, he's the know it all;)
 
Back
Top