Today a bit more information about Intel's upcoming Knights Landing platform appeared at The Register. The 60 core and 240 thread figure is quoted once again though now we know there is over 8 billion transistors on the chip, which does not include the 16 GB of near memory also present on the package. The processor will support six memory channel, three each in two memory controllers on the die, with a total of 384 GB of far memory. The terms near and far are new, representing onboard and external memory respectively. There is a lot more information you can dig into by following the link on The Register to this long article posted at The Platform.
"Intel has set some rumours to rest, giving a media and analyst briefing outlining details of its coming 60-plus core Knights Landing Xeon Phi chip."
Here is some more Tech News from around the web:
- Silent server monitoring: A neat little cure that doesn't kill the patient @ The Register
- Qualcomm, MediaTek shifting some 28nm chip orders away from TSMC @ DigiTimes
- Quebec Plans To Require Website Blocking, Studies New Internet Access Tax @ Slashdot
- Intel and Micron team up on 3D NAND flash memory for 10TB SSDs @ The Inquirer
- Toshiba and Sandisk announce BiCS as the first 48-layer 3D flash chip @ The Inquirer
But can it run Crysis?
But can it run Crysis?
This is the future.
Remind
This is the future.
Remind me of the time when our FPU where a separate chip from the CPU, we are in the same transitional era.
We got the GPU to cozy up with the CPU on the same die, but we have yet to unify the computation.
Another analogy is where the vertex compute pipeline was separate from the pixel compute.
Fusion was, and is the future.
Now, I’m have no clue how nvidia can retain their HPC market long term.
HSA and dedicated Ray Tracing
HSA and dedicated Ray Tracing circuitry on discrete GPU is also the future. Intel gets too much press, and the Adapteva and the multicore ARM companies get too little. And no one is paying the licensed Power8 market any attention, at least until Google or others make some definite moves to using them in their server rooms. For sure the Chinese are all over Power8 with China’s Suzhou PowerCore Technology’s CP1 will be based on Power8.
Both AMD and Nvidia should get into Offering dedicated Ray Tracing IP on their discrete GPU, and save their customers from having to purchasing expensive Xeon/other server SKUs for workstation graphics work. Imagination Technologies’ PowerVR wizard would make a fine Addition to a Professional grade graphics tablet, where its dedicated Ray Tracing hardware could be put to good use.
With all the multicore goodness supposedly baked into both DX12 and Vulkan it’s about time for the discrete GPUs to get a few CPU cores on those interposers along side the HBM, or even on the discrete GPU’s DIE itself, to accelerate gaming too the next level. Imagination Technologies(IT) needs to get out from under Apples shadow, and put together one of its advanced 64 bit MIPS CPU core SOC with a PowerVR wizard GPU for inexpensive graphics tablets(Tablets that can run Full Linux), But IT is caught up under an Apple sphere of influence that will cause them much harm in the long run.
Intel’s Knight’s landing is definitely a DP powerhouse, but lots of DP units are not what makes for a good GPU, as that requires thousands of cores and Tessellation/other graphics specific hardware. For sure the Knight’s landing would make a nice Ray Tracing accelerator, but so could a chip with many ARMv8A cores, that Cavium’s ThunderX with 48 ARMv8-A (64-bit) cores is one, and users could probably get more than one of these ThunderX SKUs for the price of a Knight’s landing.
do you ever shut up
do you ever shut up
Do you ever stop wearing that
Do you ever stop wearing that large yellow prophylactic, and get over your irrational x86 worship pathology!
Looks like there will be lots of Power8, up to 8 processor threads per core, competition for x86 in the server room.
Can’t wait for Apple to replace the Xeon in their Mac Pros with a Power8 SKU. And it looks like AMD will be adding SMT(more than 2 processor threads per core) to their custom ARMv8 ISA based K12 design. Those RISC core’s execution pipelines can sure be doubled and tripled, per core, to help keep that greater than 2 processor threads per core SMT working. The Power8 with 10 instruction issue per cycle, and its transactional memory instructions actually work, unlike those in certain Haswell SKUs. So AMD can take its K12 RISC custom ARMv8 ISA based design and add plenty of SMT resources per core, with plenty of execution pipelines, more so than on any CISC design, as CISC designs require much more on die space to implement the needed SMT execution resources. AMD will have its ZEN, and K12 ARM based systems using SMT, so it looks like in the server room at least things will get interesting. Everybody ARM included needs to keep an eye out for those licensed power8 RISC designs, but since the Power8 is licensable through OpenPower AMD could add a third ISA for their server SKUs should AMD need to. Don’t discount Imagination technologies MIPS based RISC SOC either, epically the newer MIPS cores with SMT options, and the newer PowerVR graphics.
The Power8 includes two floating point
units (FPUs) and two vector(VMX) units that perform
SIMD operations on 128-bit data using IBM’s VSX
extensions. When not used for vector operations, the VMX
units can also execute standard FP instructions. Each FPU/
VMX unit can execute one double-precision FP multiplyadd
instruction per cycle, so the peak throughput at 4.0GHz
is eight double-precision operations per cycle, which works
out to 32Gflops per core or 384Gflops per Chip. Not too bad for an 8 core CPU. Please, Apple get this for your Mac Pro!
Edit: Not too bad for a 8
Edit: Not too bad for a 8 core CPU.
To: Not too bad for a 12 core CPU.
12 cores 8 threads per core, 14 execution units, FP, VMX/FP, and others. Quite a RISC processor that power8, I wonder what the Power9 will do! The Hot Chips symposium for this year, or next year should have some Power9 presentations, maybe some HBM, or other goodies, who knows but it will be interesting to see.