A lot of CPU news is coming out of Computex 2016.
Intel has launched its new Broadwell-E "Extreme Edition" CPUs for "enthusiasts". The top-of-the-line model, the i7-6950X, now includes 10 cores instead of 8, but the price has increased massively to around $1,723. Compare this to a ~$999 launch price for the 8-core i7-5960X or 6-core i7-4960X flagships from previous generations.
Intel has also launched some new Skylake-based Xeons with "Iris Pro" graphics.
AMD revealed more details about the Radeon RX 480, a 14nm "Polaris" GPU that will be priced at $199 and released on June 29th. AMD intends to compete for the budget/mainstream gamer segment falling far short of the $379 launch price of a GTX 1070, while delivering around 70-75% of the performance. It also claims that the RX 480 will perform well enough to allow more gamers to use premium virtual reality headsets like the Oculus Rift or HTC Vive.
While 14nm AMD "Zen" desktop chips should be coming later this year, laptop/2-in-1/tablet users will have to settle for the 7th generation Bristol Ridge and Stoney Ridge APUs. They are still 28nm "Excavator" based chips with "modules" instead of cores.
(Score: 3, Interesting) by opinionated_science on Thursday June 02 2016, @03:57PM
once we get a device with 12 tflops, there'll be a mini-revolution with molecular dynamics....
The APU would be great to achieve this, but I understand the bus between the x86 CPU and onboard GPU is still PCI!!! on the same die!!!
For those reading, the latency of PCI-E is a massive overhead for parallelism, until we got GPU's where the bandwidth became the issue.
My $0.02...
(Score: 2) by LoRdTAW on Thursday June 02 2016, @10:39PM
Correct me if I am wrong, but the idea behind AMD's HSA is to use shared memory between the CPU and GPU. The idea being that since the GPU and CPU share the same memory controller, the data to be processed has its address passed to the GPU and the GPU directly accesses that piece of memory without a pci transfer.
(Score: 2) by opinionated_science on Friday June 03 2016, @03:49PM
that's what I was referring too - but originally (and I dug around the APU I had to hand A10 something) the GPU was on the otherside of a PCI bus, inside the CPU package!!
If I understand correctly Hypertransport is more efficient (allows interleaved transfer) but is only used for CPU-CPU and CPU-MEM.
For GPU's to work, they use PCI as they can be directly addressed from the CPU via PCI, as well as be a DMA master.
If new architectures share GPU/CPU memory with lower latency (1 us), I would be very interested.
As of now, plans are to load as much on the GPU and let it rip - hence my number 12 Tflops. I calculated it a few years ago, though someone actually built a machine that solves the problem out of custom ASICs...
(Google Anton, D.E. Shaw)