I'm pretty sure DICE was referring to the D3D driver, and not the KM driver. As you can see by that graph, the D3D driver is run across all threads in both DX11 and DX12.
As to why the KM driver disappears in DX12, I have no idea..
I wonder if this isn't in fact what DICE calls the driver thread, which isn't present in Mantle either? If not, what is it exactly, and why does it go away completely with DX12?
I'm pretty sure DICE was referring to the D3D driver, and not the KM driver. As you can see by that graph, the D3D driver is run across all threads in both DX11 and DX12.
As to why the KM driver disappears in DX12, I have no idea..
Pretty much confirms it, if Johan is saying he had a direct hand in DX12 design then its basically multi-vendor Mantle. Kudos to MS for pushing it, all gamers will benefit with alleviating CPU bottlenecks, few things are worse than fps chokes during intense scenes in games.
As to why the KM driver disappears in DX12, I have no idea..
Sorry, but that's absolute nonsense. DX12 has been in design for four years according to NVidia, and it's much more similar to DX11 than it is to Mantle..
DX12 uses a more advanced and efficient form of command lists (just like DX11) to enable multithreading, and also relies on the D3D driver to create threads for resource management and draw calls... DX12 also runs on various architectures..
Mantle has none of those attributes.. I'm sure some similarities exist, particularly when it comes to exploiting hardware at a lower level, but overall, DX12 and Mantle aren't very similar..
Is it me or are we pretending that DX11 doesn't work just fine? :hmm:
If DX12 has been in development for 4 years then why do we have to wait another 2?
6 years? how long does it take to develop an API, Mantle took 2 years, not 4, certainly not 6.
Ah,
Microsoft chose nVidia to showcase DX12. A fix to their problem would be a new console. D:
We've already seen the gains that AMD can get with an efficient new API- getting NVidia on stage for the demo lets Microsoft demonstrate that the same is also true for NVidia. NVidia get the same gains that they would have gotten from supporting Mantle, without the shame of supporting an AMD API. We all get more efficient games thanks to AMD giving Microsoft a kick up the backside, and everybody wins.
So, putting aside all the arguing about whether DX 12 is mantle or not, or who was first or whatever, the bottom line comes down to what place does mantle have when DX12 comes out. Intel and nVidia throwing their support behind DX12 makes it pretty clear they are not going to support mantle. Obviously all developers will have to code for DX. So if DX12 can give similar performance to mantle, why would a developer want to code for a separate path for AMD only hardware?
The Mantle name is probably dead. AMD will hopefully support DirectX 12, and transition their Mantle games to DirectX 12. AMD liked to talk about supporting other vendors with Mantle, and they got their wish- they just need to accept the DirectX name.
(Of course, "transition" may consist of find-and-replace "Mantle" with "DirectX", but that's another matter.)
The only casualty in this is the possibility of Mantle on Linux. Which was, of course, Microsoft's goal.
So, putting aside all the arguing about whether DX 12 is mantle or not, or who was first or whatever, the bottom line comes down to what place does mantle have when DX12 comes out. Intel and nVidia throwing their support behind DX12 makes it pretty clear they are not going to support mantle. Obviously all developers will have to code for DX. So if DX12 can give similar performance to mantle, why would a developer want to code for a separate path for AMD only hardware?
The king is dead, long live the king?The Mantle name is probably dead.
And I wonder what the DX12 announcement means for NVidia's plans to build a co-processor into their GPUs?
It won't disappear, but the new display driver model will use a very-very thin KM driver. This will decrease the number of driver threads near to zero. We don't need them, these are the enemies in the current APIs, because they conflicts with the app threads, and it will decrease the performance. This is why Mantle is so efficient, it completely eliminates the driver threads.
SourceDirect3D 12 introduces a new model for work submission based on command lists that contain the entirety of information needed to execute a particular workload on the GPU. Each new command list contains information such as which PSO to use, what texture and buffer resources are needed, and the arguments to all draw calls. Because each command list is self-contained and inherits no state, the driver can pre-compute all necessary GPU commands up-front and in a free-threaded manner. The only serial process necessary is the final submission of command lists to the GPU via the command queue, which is a highly efficient process.
You dont even know if command lists still exists in DX12, but keep repeating BS like always.
IMO as a developer, command lists existed to extract even the smallest idle time of the kernel mode driver.
This is how I see it: the new model is straightforward, just like mantle the gpu can execute tasks directly without the need of a KM driver. The code just submit tasks to the gpu and continues running submitting new tasks directly to the GPU, which means the GPU keeps busy all the time.
The opposite is having a single thread doing that (KM driver) from a list of tasks (which in turn needs to be filled, using command lists or not).
SourceDirect3D 12 introduces a new model for work submission based on command lists that contain the entirety of information needed to execute a particular workload on the GPU. Each new command list contains information such as which PSO to use, what texture and buffer resources are needed, and the arguments to all draw calls. Because each command list is self-contained and inherits no state, the driver can pre-compute all necessary GPU commands up-front and in a free-threaded manner. The only serial process necessary is the final submission of command lists to the GPU via the command queue, which is a highly efficient process.
How many times in this thread have I posted a direct quote from Microsoft affirming that DX12 uses command lists. I'll post it again:
Source
Mantle does not use command lists, or driver threads, so the model is completely different from Mantle in that respect.
That's what I'm said. The KM driver is still exist, but it won't run much server threads, and even the workloads will be very light.According to Microsoft themselves, DX12 will use command lists, and command lists rely on driver threads:
Source
So DX12 won't get rid of driver threads, but they will be made thread safe, and consume less CPU time, both of which were big problems with DX11 MT.
The only way they can probably get rid of driver threads completely would be to go the Mantle route, but that's unfeasible for a multi-architectural solution imo.