What future Intel cpu's will look like

[deadrats]

i'm sure some of you have seen the HP "Machine" story i linked to and probably thought "not any time this century".

well, Intel seems hell bent on staying the dominant player in many markets and today announced some pretty bad assed Xeons:

http://hothardware.com/News/ardware-Boost-Intel-To-Offer-Custom-Xeons-With-Embedded-FPGAs/#!0HamN

The FPGA provides our customers a programmable, high performance coherent acceleration capability to turbo-charge their critical algorithms. And with down-the-wire reprogramability, the algorithms can be changed as new workloads emerge and compute demands fluctuate. Based on industry benchmarks FPGA-based accelerators can deliver >10X performance gains. By integrating the FPGA with the Xeon processor, we estimate that customers will see an additional 2X in performance thanks to the low latency, coherent interface.

it will be a while before we see consumer grade cpu's with this technology but considering that all technology introduced in Xeon class processors eventually ends up on the desktop versions as well, i wouldn't be surprised if in a couple of years there's an "i7 E" that has this type of accelerator.

[jagabo]

I've used PGAs in a few projects in the past -- back when a few thousand gates constituted a large device. They're pretty cool but you won't be programming things like a state of the art 3d graphics device with this.

[nic2k4]

That's not likely to become a consumer product. FPGA are great if you want to convert a software function to hardware, but it's not like you could make a Virtualdub chip. It wouldn't even be good for a hardware H264 encoder, that takes a lot of parallel data processing something a GPU is better suited at. FPGA are good at serial data processing, one good consumer application would be on the fly data encryption.

[deadrats]

That's not likely to become a consumer product. FPGA are great if you want to convert a software function to hardware, but it's not like you could make a Virtualdub chip. It wouldn't even be good for a hardware H264 encoder, that takes a lot of parallel data processing something a GPU is better suited at. FPGA are good at serial data processing, one good consumer application would be on the fly data encryption.

this is just too funny, whenever i mention gpu + video encoding someone invariably shows up to say that gpu's are good for highly parallel jobs but that video encoding is inherently a serial process. a debate then ensues as to whether or not video encoding is a serial job or a parallel one.

now i post a link about new tech and someone shows up to say that these chips are well suited for serial data processing but a gpu would better for video encoding because h264 takes lots of parallel data processing.

i just can't win.

here's the reality, video encoding is both a serial and highly parallel data processing job depending on how you structure the code and how you thread the application; for instance f265 features 4 different threading models.

the thing with a chip like this that it would simplify the threading model used; if these chips are great for serialized work then you simply don't thread your app and handle all the processing in a serial model, this makes the code simpler, more manageable, less memory intensive and less I/O bound, so personally i'm all for Intel adding these chips to their processors.

[nic2k4]

Hey, thanks for the link! Just pointing out that the applications for that chip are limited, even more so as a consumer product. For this chip to be released as a consumer product it would have to have functionality that serves everyone. On the fly data encryption is an obvious consumer level use that comes to mind, video encoding is too specialized to be considered consumer level.

Intel is only marketing a "service" they have offered to select customers in the past (Google, Facebook, Yahoo...), they're targeting anyone who runs a data center. Now, the application you suggest would be awesome, I just don't think it's possible. Modern CPU's have had a thread scheduler for years (HT P4), what you suggest is a thread generator. If someone can come up with the algorithm and convert it to Boolean functions, we could have such a chip. I just wonder how many gates it would take.