Another player emerges in the CPU landscape: Qualcomm is introducing its first socketed processor for the enterprise market.
Image credit: PC World
A 24-core design based on 64-bit ARM architecture has reached the prototype phase, in a large LGA package resembling an Intel Xeon CPU.
From the report published by PC World:
"Qualcomm demonstrated a pre-production chip in San Francisco on Thursday. It's a purpose-built system-on-chip, different from its Snapdragon processor, that integrates PCIe, storage and other features. The initial version has 24 cores, though the final part will have more, said Anand Chandrasekher, Qualcomm senior vice president."
Image credit: PC World
Qualcomm built servers as proof-of-concept with this new processor, "running a version of Linux, with the KVM hypervisor, streaming HD video to a PC. The chip was running the LAMP stack – Linux, the Apache Web server, MySQL, and PHP – and OpenStack cloud software," according to PC World. The functionality of this design demonstrate the chip's potential to power highly energy-efficient servers, making an obvious statement about the potential cost savings for large data companies such as Google and Facebook.
HOLY SHIT.
I bet your ass
HOLY SHIT.
I bet your ass that Intel shat some major violet bricks right about now. Holy crap, Qualcomm…and absolutely out of nowhere, too…AND it’s LGA…WTF.
Doesnt Intel have 18 core (36
Doesnt Intel have 18 core (36 thread) xeon already (not prototypes)
based on their old 22nm process.
From what I recall all benchmark show intel core to be more powerful, and actually not worse in term of power efficiency in the real world.
I’m sure Intel takes ARM in the server space seriously, but 24 ARM core is not impressive in any ways, shape or form.
Now, if they have ~128MB of cache, 512bit memory bus, run at 4ghz, etc.. then yea, Intel would be scrambling. But this seem more like a chip with lots of weak cores.
Also depends on the power
Also depends on the power consumption; Intel does have 18 core chips but they are using a lot more power than this one
Intel 18 core chips are like
Intel 18 core chips are like $7000 dollars. These are probably much smaller cores, so they may be much cheaper for the core count, but each core will be weaker. They may be a great solution for throughput computing problems that can take advantage of a large number of cores. We will see if that have been able to design a cache architecture to support such core counts.
“The initial version has 24
“The initial version has 24 cores, though the final part will have more, said Anand Chandrasekher, Qualcomm senior vice president.”
So this 24 core variant is probably for a development platform, with the higher core count products to follow. What Qualcomm needs to do is make some custom ARM cores with SMT, and ARM cores being smaller(Less transistors to implement a RISC ISA) Qualcomm could probably easily have as many cores as a Xeon Phi. Maybe when AMD’s custom K12 ARM cores arrive we will see lots of SMT enabled ARM cores paired with a Greenland Graphics accelerators and HBM on an interposer systems enter the market.
Xeon Phi is a totally
Xeon Phi is a totally different market and not comparable. The Xeon Phi core is significantly simpler than these ARM cores, but it has a 512-bit vector unit. I don’t know how the size of the cores compare, but these chips would have completely different “un-core” designs also. This processor would be designed to run server workloads, so it would require high hit rate, low latency caches. This is a different problem compared to the Xeon Phi, which requires very high bandwidth for streaming code, but would be significantly less latency sensitive.
Really and still be using the
Really and still be using the x86 CISC ISA, the Phi’s core’s still take more space/transistors to implement than the ARM RISC based ISA does, and the same space savings goes for the Power/Power8 RISC ISA! Why do you think that the power8 processor can support 8 processor threads per core and a 12 core Power8 can have 96 threads total. The RISC designs can support more cores, and more instruction decoders per core in the case of the power8, along with the power8’s 14 execution units per core to run those 8 processor threads. The Xeon phi is a large chip and because it has to support the x86 CISC ISA it can not support as many cores/processor threads per unit area and those extra transistors needed to implement the x86 ISA generate more heat.
Intel has been making its cores smaller with process node shrinks, but others are catching up to the 14nm node and now those ARM chips are beginning to come at 16nm and 14nm, so expect the ARM cores to use even less power. Intel has a hard time just attempting to compete with 28nm ARM ISA based chips on their 28nm node with the power usage/space savings metrics, and Intel has had the 14nm process up and running, and still those 28nm ARM chips are more power efficient. Add to that HSA and GPGPU compute that the entire ARM based industry is embracing and Intel has to compete with the GPU accelerators that have more FP performance that even the Xeon Phi can support, and the GPUs can do so while using less power per flop/s than the Xeon Phi!
That’s not ARM, though. This
That’s not ARM, though. This here Qualcomm’s stone is 100% ARM.
It’s hard to tell because
It’s hard to tell because some of the technical press is not indicating if this Qualcomm is a custom design or an ARM holdings reference design, they just say 64 bit ARM.
The Platform states in their article: “Qualcomm is not disclosing the core design or other attributes of its 64-bit ARM server chip except to say that it hews to the ARMv8-A specification from ARM Holdings and that it has 24 cores. Qualcomm is not revealing what process technology it is using to make the prototype chip or any details on clock speeds, cache memory, main memory, on-chip controllers, and other features, and it is not talking about its code-name, either. What the company is saying is that it will deliver a production-grade ARM server chip using an advanced FinFET 3D transistor process that will have more cores than the pre-production chip showed off today.”(1)
So the Qualcomm Chip Runs the ARMv8A ISA, but Qualcomm fails to disclose the code-name of their core, so if it is a custom or reference ARM design is still unclear.
Apple’s Custom A7 chip is twice as wide as the ARM Holdings A72, with the custom A7 Cyclone having 6 instruction decoders per core to Arm Holdings’ A72 reference cores 3 instruction decoders per core. The Apple A7 also has more execution pipelines to feed from those instruction decoder’s output, than the Arm Holding’s A72 reference design. There is no telling the resources of Apples A8/A8X, or A9/A9X as no one to date has listed a complete ad thorough description of ether of Apple’s 2 latest cores. Qualcomm is goint to have to be more forthcoming that Apple about this Chips CPU core design because All the 64 bit ARM chips custom and Arm Holdings reference can run the ARMv8a ISA!
(1)
http://www.theplatform.net/2015/10/08/qualcomm-shows-off-prototype-arm-server-chip/
P.S. I can see Apple in the
P.S. I can see Apple in the Future maybe taking its Custom ARMv8A running cores and designing an in house server chip for its internal server needs! There will be lots of competition for ARM server SKUs in the next 5 years, including Jim Keller’s designed K12 custom ARM chips from AMD that may even have SMT capabilities, and an extra wide order superscalar design like Apple’s A7s/A8s/A9s! Those reference Arm Holdings designs(A72,A57,A53) are going to have to compete with the custom ARM chips that companies like Qualcomm/Apple/AMD/Samsung/Others only license the ARMv8A ISA and have their own custom micro-architecture to run the ARMv8A ISA!
64 bit?
64 bit?
Fixed the typo, thanks!
Fixed the typo, thanks!
we’ll need to wait until the
we’ll need to wait until the price, scaling and power draw numbers become available.
any competition for intel from ANY front is welcome news.
It’s ARM, man. ARM is already
It’s ARM, man. ARM is already cheap as hell (sheer price-per-yield wise).
The thing is huge.
The thing is huge.
Not really. Judging by that
Not really. Judging by that picture alone, it’s about the size of a top Sandy/Ivy Xeon. It’s LGA, after all, you can’t really make very big stones for that, even if it’s a socket like that of 2011 or similar.
Unlocked? ( ͡° ͜ʖ ͡°)
Unlocked? ( ͡° ͜ʖ ͡°)
Server chip, so no.
Server chip, so no.
Current Intel 2011-v3
Current Intel 2011-v3 E5-1600v3 Xeons are unlocked. The X5xxx series were unlocked too.
Weird
Weird
These would make a nice
These would make a nice storage server lol freenas box or nice Linux distro that’ll support them.