Nvidia's Facebook Page Teases That Something is Coming

Page 5 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

Gundam288

Distinguished
Sep 23, 2011
281
0
18,790
[citation][nom]xdeathghostx[/nom]the first Direct X 12 videocard![/citation]
Double posting does NOT make a false statement true, just saying....

As for a real guess it's ether a new 6xx card or the new Tegra(i think it's 3, don't really keep up to date on tegra.), im leaning more on the card side than Tegra since it seems more logical.
 
G

Guest

Guest
It's another way to say please people don't buy AMD/ATI cards. Wait,wait... even though the new product will not being something revolutionary but that way can pull some time...
 
G

Guest

Guest
So after all the waiting it's the gtx 690, what a waste. Maybe that's why there is no gtx 680 availability.
 

dragonsqrrl

Distinguished
Nov 19, 2009
1,280
0
19,290
G

Guest

Guest
Now that HD4000 has arrived, can OpenCL be used to enhance the performance of the Ivy Bridge processor while simultaneously using a discrete graphics processor? If OpenCL can utilize GPU cycles for general purpose compute tasks then It should be able to utilize the Intel integrated GPU for more general purpose processing power in addition to the Ivy bridge's other CPU cores, while the discrete GPU uses its resources for the graphics. OpenCL should see all the hardware on the computer as an available resource and It should be able to do this? If not then what is described as Heterogeneous computing has not completely arrived yet! Or is it just a matter of waiting for the software to catch up?
 

bit_user

Polypheme
Ambassador
[citation][nom]Heterogeneous[/nom]Now that HD4000 has arrived, can OpenCL be used to enhance the performance of the Ivy Bridge processor while simultaneously using a discrete graphics processor?[/citation]yes, an OpenCL program can see all resources in the system. Currently, a program would have to link in runtimes from each vendor, however.

IBM has made a library that takes care of this, allowing you to link in their middle layer and see all devices through one runtime library. They call in the OpenCL Common Runtime, if any devs out there are curious.

OpenCL does expose the divisions between devices, so it would be extra burden on a developer to split their computations across multiple devices. Even with the middle layer from IBM.

BTW, you don't need 2 GPUs for this. A program can use a single GPU for both graphics and compute, though it must be done with care (existing APIs are still somewhat primitive around resource contention and preemption).
 
Status
Not open for further replies.

TRENDING THREADS