From last years GDC:
http://www.youtube.com/watch?feature=player_detailpage&v=XiQweemn2_A#t=3s
Practical limit to human vision is 8000 X 4000 @ 72FPS with a 90 degree field of view (about 4 feet from a 100 inch screen or 2 feet from a apx 40 inch screen). HDTV viewing 2560 X 1600 30 degree field of view. (Above 1080P but below 4K depending on how far you are from the screen.)
Current PS3 and Xbox is 2nd order approximation, 2 bounces of light off screen images and at 1080P is .25 Teraflops. 3rd order approximation 3 bounces of light needs 2.5 Teraflops or 10 times PS3 power.
2 more generations till we reach human vision limits.
3D stacking mentioned including the same IBM stacking picture I posted.
Video is well worth watching. Given the information in this video, next generation console MUST be at least 10 times more powerful. 3rd order approximation for lighting and true 1080P would need more than what some are speculating for next generation.
Article on the above video by ars technica
The
Global Foundries PDF has some interesting information.
1) 2 year lead time to production
2) 20nm and 3D stacking sometime after 2012 with High K gate first technology.
So we now have Major players both Global Foundries and IBM Forges ready to 3D stack with sub 32nm die by 2013.
And yes they are counting on tech that has not been perfected being available on a timetable 2+ years out as is the industry norm......
Both Microsoft and Sony waiting till at least next year makes sense! The rumors of IBM working with Sony on a Cell 2 several years ago would then make sense if 2+ years is needed from design to forge. So CPU design has been done, waiting on a GPU and processes to make a 10X next generation Game console economically practical. Kepler is somewhere above 2 and less than 3.1 TFLOPS but too hot and too expensive.
How the rumor that Cell is dead got started:
Last month Sony confirmed that it reached a preliminary agreement with Toshiba to sell off its production facilities for advanced chips
(including the Cell in PS3). The sale of Sony's semiconductor business to Toshiba is estimated to be worth about 100 billion yen ($861
million).
Now, however, the word from the Nikkei Business Daily in Japan (via Solid State Technology) is not only is Sony selling its Cell production
facilities, but the electronics company is also planning to withdraw from its R&D project with IBM and Toshiba, which sought to reduce
the Cell chip to 32nm and below. IBM and Toshiba are expected to carry on the project, though.
In addition, Sony intends to cancel all capital investments in production of 45nm and later Cell chips. The current Cell chips being
made for PS3 are using a 65nm process, SCE head Kaz Hira confirmed. Rather than pursue Cell development, Sony is planning to
strengthen its work in CCD and CMOS image sensors. Sony executive deputy president Yutaka Nakagawa explained to the
Nikkei that the move is part of Sony's "asset light" strategy to end chip manufacturing investments after 65nm. The focus will switch to
design over manufacturing.
"We plan to minimize the investment that is required to make packaged IC chips smaller," said Nakagawa. "Manufacturing
cutting-edge packaged IC chips is not considered as important as it once was. The most important thing is what type of chip a company
decides to produce, so we will increase the number of designers depending on the chip's purpose. The fact that we will stop operating
an advanced chip plant does not mean that we are downgrading the importance of the chip business."
He added, "About 100 employees who worked on designing the Cell and other advanced microprocessors have already been reassigned to
divisions where manufacturing technologies for image sensors and analog IC chips are being developed."
Rumor The Cell chip 2 is in development, and goes for PlayStation 4 Oct 2011 (+ 2year plus lead time to production = beginning 2014) Timing adds support to this unconfirmed information.
Mindful of the information comes from a trusted source: apparently, the Cell chip's successor is already being designed. Is not particularly surprise us, but it also makes two details that we have arrived, is being developed by the Barcelona Supercomputing Center, and its most immediate application will be PlayStation 4.
Apparently it is in Barcelona where the Cell chip 2 (according to our source, the official name, and predictable, the Cell that carries in its heart PlayStation 3) is being designed with the help of one of those mostrencos of computing that also used for research and development in many other fields, like medicine. As the original Cell, the most immediate use will chip when its development is completed PlayStation 4, Sony's new console from which, so far, nothing is known.
There are alot of
good discussions in 2008 about Cell-X86-Larabee in a PS4 and even a mini-Cell in a new PSP. What has changed....OpenCL.
The quad GPU in the Vita was designed with OpenCL in mind (known since 2009) and it can be used for codecs, math, AR and more.....mini-Cell is not needed in the Vita. 2 Cell SPUs are about equal to 32 math elements in a GPU and Cell can be easily used with OpenCL (32-64 SPUs). Backward Compatibility requires at least 6 SPUs and if a GPU is 100% being used for graphics it can't be OpenCL used for physics or AI...CPU is still needed and SPUs have no IP cost.
banned source add opa-ages dot com to /forums/topic/12493-detailed-cell2-info-ibm-will-continue-cell-development-for-its-mainframes-but-next-gen-scei-console-wont-use-cell2/
Barcelona Supercomputing Center
http://research.nvidia.com/content/BSC-ccoe-summary said:
NVIDIA and BSC started collaborations back in April 2008 with an initial conference presented by David Kirk on “NVIDIA CUDA Software and GPU Parallel Computing” in Barcelona. In 2010, BSC became the first NVIDIA CUDA Research Center in Spain, with Mateo Valero as PI and Prof. Nacho Navarro as the managing director of the center. In addition to an active research programs and many highly regarded publications, the CUDA Research Center also recognizes BSC’s efforts in CUDA education, highlighted by the second edition of the (PUMPS) Summer School, “Programming and Tuning Massively Parallel Systems”. Prof. Nacho Navarro organized the event, co-sponsored by the University of Illinois, the HiPEAC NOE and NVIDIA, with distinguished faculty members Dr. David B. Kirk of NVIDIA and Prof. Wen-mei Hwu of the University of Illinois.
Now as a CUDA Center of Excellence, BSC will utilize GPU computing equipment and grants provided by NVIDIA to support a growing number of research and academic programs. For the next three years, BSC and UPC plans to build an Education Program on Parallel Programming using CUDA, to provide a cluster-aware programming environment for GPUs, to optimize mutlti-GPU runtime management with GMAC, to support GPU acceleration from the task-based StarSs programming model and its OmpSs implementation and to build a new GPU-based cluster prototype system to explore the potential of low-power GPU clusters as high-performance platforms.
Best guess is there is some truth to the rumor, PS4 OS may in part be designed at BSC but the Cell2 if it's coming is probably a IBM-Sony collaboration. Cluster computing, Massively Parallel Systems, the Cell SPU and OpenCL lend themselves to the
original Cell vision. Oh, and
if BSC is designing the PS4 OS or parts of it then I think Nvidia is again making the PS4 GPU.
The above could support the PS4 having Nvidia Tegra CPUs and Nvidia GPU so we now have rumors supporting CPU types X-86, Power7, Tegra (ARM) and GPUs from AMD and Nvidia.
My best GUESS is Power7 + more than 6 SPUs and a Nvidia GPU. Tuning is needed to assign tasks to and to split off parallel processing elements to perform tasks using OpenCL. This should include multiple SPUs in the CPU.
This is now getting interesting; Next generation is going to use many processes to improve performance and multiple CPUs and GPUs rather than increasing clock speeds is part of this. OpenCL makes using highly parallelized designs practical including multiple SPUs. Since OpenCL code is somewhat cross platform like OpenGL it supports the original Cell vision of Distributed processing. For it's part I think Sony invested in Cell for Coming multi-media like 3-D, 4K and 8K video. These new resolutions are going to need codecs that will require orders of magnitude more CPU power than we had in 2005. Cell may well be dead and replaced by modern GPUs running CUDA or OpenCL or Cell may now be practical because of OpenCL and in multiple new Sony products.
I also find it interesting that BSC, Collabora and Fluendo are based in Barcelona, Spain and
Collabora-Fluendo announced a partnership promoting the GStreamer multimedia framework through the creation of a cross platform software development kit (SDK), targeting desktop and server platforms like Linux, Windows and Mac OS X, and very soon to include leading mobile platforms, such as Android. 1080P, 3-D, 4K, 8K with zooming, up-scaling and down-scaling needs lots of processing power.
In the NeoGAF thread; "PS3 Web Browser Discussion - big upgrade rumoured for long time, but no concrete news" I reported information on Collabora being sent a PS3 Developer Kit to "See what they could do", this was end of 2007. Gstreamer is a Multi-media framework that at that time was being accepted as the multi-media, HTML5 <video>, part of Webkit for Posix platforms like Linux and Unix and part of both FireFox and Opera. There were issues integrating it with Openmax IL and a newer version of OpenMax was delayed from 2008 to 2011 (OpenMax 1.2). Gstreamer 1.0 and OpenMax 1.2 both address issues with a Gstreamer Openmax wrapper and provide for next generation AR support.
Openmax IL provides low level APIs and Codec support for multi-media and is provided by the hardware vender. It's used by Android and Sony to support Multi-media and if Gstreamer is used as the upper level player a Gstreamer to openMax wrapper is needed. With that, OpenMax IL serves the same function as a Gstreamer Core and Codec plugins. It appears to this point that Sony has been using AVM+ Flash open source for DASH players while companies like Netflix have been using Gstreamer-openmax. AVM+ can't be used for commercial IPTV (without paying Adobe) and to this point Sony has not had a video streaming IPTV DASH player service like Netflix. Best guess is they were waiting for OpenMax IL 1.2 and Gstreamer 1.0 and will now use Gstreamer for multiple multi-media features like Streaming AR, DASH player for IPTV, DLNA, Video conferencing, video editing and more (my opinion).
Gstreamer is now production ready (with Gstreamer 1.0 and OpenMax 1.2) and can now support multiple platforms using Open Standards.
Two multicoreplatforms are concentrating an enormous
attention due to their tremendous potential in terms of
sustained performance: Cell BE and NVIDIA GPGPUs
Other libraries/environments that support Cell BE programming
• Data Communications and Synchronization Library (DACS) and Accelerated
Library Framework (ALF) are included in the latest releases of the IBM SDK
• Cell Superscalar (CellS) developed at Barcelona Supercomputing Center
• Multicore Plus SDK Software developed by Mercury Computing Systems Inc.
Meeting on Parallel Routine Optimization and Applications – May 26-27, 2008
18
• Multicore Plus SDK Software developed by Mercury Computing Systems Inc.
• RapidMind Multicore Development Platform for AMD and Intel multicore x86
CPUs, ATI/AMD and NVIDIA GPUs and Cell BE
Conclusion in 2008 is Nvidia GPU is affordable and easier to program (CUDA) than Cell. OpenCL changes the latter but affordability I can't answer. The reasons for posting the cite is it has a bearing on Cell viability and Barcelona Super Computing Center is mentioned again supporting the above Rumor for at least a OS based on Cell/Nvidia Highly parallelized SPU/GPU in a PS4.
http://www.training.prace-ri.eu/uploads/tx_pracetmo/gpuvideo1.pdf
Mapping Iterative Medical Imaging Algorithm on Cell Accelerator
the algorithm running on one SPE is over 5 times faster than on one core of the AMD Opteron processor. For 360 subsets, the Cell BE is 2.7 times faster than AMD Opteron processor. Note that for larger subsets, the number of DMA transfers between the local store and main memory increases on the Cell BE, increasing execution time. However, compared to AMD Opteron processor, the Cell BE still performs better.
If Sony is going to use the custom CPU/GPU they develop for the PS4 for medical imaging (NMR or CT to video and display) a X86 is not practical except to manage Cell SPUs or GPU. Power7 is 2.7 times faster than X86 and can be used to manage SPUs and there is Sofware owned by Sony to do this.
Medical Imaging
Computed Tomography (CT) reconstruction is a computationally and data-intensive process applied across many fields of scientific endeavor, including medical and materials science, as a non-invasive imaging technique. A typical CT dataset obtained with a CCD-based X-ray detector, such as that at the Australian Synchrotron with 4K×4K pixels captured over multiple-view angles, is in the order of 128GB. The reconstructed output volume is in the order 256GB. CT data sizes increase at 1.5 times the number of pixels in the detector, while the data-processing load generally increases as the square of the number of pixels, hence data storage, management and throughput capabilities become paramount. From a
computational perspective, CT reconstruction is particularly well suited to mass parallelisation whereby the problem can be decomposed into many smaller independent parts. We have achieved significant performance gains by adapting our XLI software algorithms to a two-level parallelisation scheme, utilising multiple CPU cores and multiple GPUs on a single machine. In turn, where data sizes become prohibitively large to be processed on a single machine, we have developed an integrated CT reconstruction software system that is able to scale up and be deployed onto large GPU-enabled HPC clusters.
Medical Imaging is more that I first thought if Sony is going to be working with the raw data out of a NMR. Upgrades to the old computers and imaging system that are now almost 10 years old can massively speed image processing, I would guess this is an open market for Older NMR and CT equipment.
