ryanvade wrote:If I were a game dev, PhysX would be at the top of the list. It provides some of the BEST graphics..Most of the top latest games have some form of PhysX. I hear Battlefield 4 and Battlefront III will have it. As well as Star Wars 1313, Tom Clancy's The Division and more. I think PhysX gives the best quality to games. But yes, if it were Open Source many more games would have it.
I do have to ask. If many game engines are not hardware dependent, why can't I play Battlefield 3 at full quality on my GT 630M ?
Modern gaming GPU's are like 486 dx2 Intel chip. Its performance is only so much. It does not address compute power. A scenario would be the single core 4Ghz pentium 4 flagship chip that Intel scraped in 2002-2003. It served no purpose other than for marketing.
A serious clause would be to compare cpu optimized code versus gpu compute. Many a times a cpu beats a gpu on a 'single' system with small amounts of data. A simple reason; CPU technology and multicore efficiency has taken a quantum leap. A video rendered in 10 hours, with new tech gets reduced to 2 hours, newer stuff will pull that to 1 minute. A 4770k is a testament to that. It makes the 12 threaded 980x whimper at nearly quarter the power consumption.
Same applies to GPU. Today OpenCL and CUDA are infants who are fresh out of the womb. CUDA is almost a year old compared to OpenCL in terms advancement. AMD recently scraped OpenCL because they need to optimize and release it for hUMA APU chips that share the cache. If it needs more memory, pointers in the code say go to ram. Maxwell too is a revolution.
Another thing is regression factor in GPU's that Opencl exposed. A small amount of data that requires a lot of compute power runs faster on CPU than GPU. The scenario changes when an exponential amount of compute is required. OpenCL is way faster.
Even a simple upgrade path to GDDR6 is flaccid technology. Unless hardware takes a stand in making coding simple, it would be like rubber never got invented and everybody uses carbon steel for tires.
No more CPU's from AMD, only APU's. Servers will be using it, which concludes a high level of Opencl and Opengl performance. Data shuttling between CPU and GPU at high speeds sharing the same cache. Serial processing will be done on CPU and when high level of number crunching is required, a programmer will be oblivious of the GPU because of transparency that Opencl will provide...automatically data will go to the GPU. Imagine a day when no drivers will be required for GPU's!! All inbuilt into windows and linux kernels. They will end up as interconnects. Newer hardware will work faster when installed because opencl API will scale up automatically to newer pipelines.
Most people aren't aware but Intel is making real advancements that can beat AMD APU's. They are silent about it. Iris pro would be laughable 2 years from now but at the same time its shocking evidence of things to come. HD 4600 itself is super advancement.
ANother major component amonst CPU's and GPU's are DSP's ( Digital signal processors). Real time processing is the job DSP's. GPU's are more like pseudo-DSP's. You would notice everything in a computer has a 'wait' time, although most of it is because of slow hard disks. DSP's take in data that requires immediate processing ...like an audio amplifier that takes in sound immediately and outputs it in high fidelity with minimal latency, though most amplifiers dont use DSP except the expensive ones..... One reason why octa core mobile chips exist. They dont call them DSP's anymore because of small dies manufactured on the latest nodes that cut power consumption drastically just as a cpu. Most stock mobile ARM chips are rated at 2 to 3Ghz and are extremely power efficient. So it makes sense to pull down the clocks and lower voltage and amperage even more..