That’s one smart phone right there. Can it run Android?
Heads up: Fujitsu tips its hand to reveal exascale Arm supercomputer processor – the A64FX
Fujitsu has unfurled the blueprints for its homegrown high-performance Arm-based processors dubbed A64FX, the brains of its Post-K supercomputer. The designs were shown on Tuesday at a gathering of semiconductor engineers in Silicon Valley. The Post-K is a 1,000 peta-FLOPS monster – an exascale machine – that will supersede …
COMMENTS
-
-
-
Wednesday 22nd August 2018 08:12 GMT Mage
Re: ARM dreams of being in a laptop?
I'm amazed I didn't know of the A4. Wikipedia doesn't seem to mention it, or not obviously.
http://chrisacorns.computinghistory.org.uk/Computers/A4.html
http://www.computinghistory.org.uk/det/22807/Acorn-A4-Laptop/
Has no-one made a Raspberry Pi into a "laptop"? I found that the barrier to using an ARM tablet with a decent USB keyboard, USB hub, mouse etc as a "laptop" is the abysmally unfinished nature of Android and fact it's optimised for a small screen. Sometimes you can have two windows, some applications can print (the Brother print driver seems to work without the cloud), some applications support external storage. You need a third party file manager. A decent keyboard with AltGr, \ beside Z and non-USA English support and mapping needs a third party program.
Even my ancient Sony Ericsson phone had HDMI, USB2Go etc and with Android 4.x sort of worked with an HD screen, USB HDD, mouse and keyboard.
The cheap Tablet hardware is out there to make an ARM Laptop, but now LESS useful as makers drop separate charger port, 3.5mm jack, HDMI connector, SD card slot etc. Also unless you can root it an put on Linux, the Android OS is too consumption & phone orientated to use as a laptop. Hence Chrome, but if I got a free Chrome thing, It would be given away if I couldn't put Linux on instead. Chrome purely exists as Google Services client, though some offline use possible.
The problem is that even Linux Laptop usage is still a minority sport helped by WINE. Sadly for people creating content, both x86 and Windows have dominated for too long. Apple can easiest do an ARM laptop, because they have a history of dropping CPUs (68000 family, Power PC, 32bit x86), they seem to be going 64bit only on the x86-64 cpu with blocking even 32bit applications.
IBM PC and Windows 9x (rather than multiplatform NT from 3.1 to 5.x --Win2K,XP Server 2003) have really held back the industry.
-
-
Wednesday 22nd August 2018 09:05 GMT Charlie Clark
Re: ARM dreams of being in a laptop?
Has no-one made a Raspberry Pi into a "laptop"?
Yes, and thie PiTop is a very interesting device. But the Pi is not the way to go due to Broadcom effectively canning development. The current chips are are now quite a way behind those being put into phones.
-
-
-
Thursday 23rd August 2018 16:38 GMT Neil Spellings
Re: ARM dreams of being in a laptop?
Shame the A4 never saw the light..I designed a very nice dual-mode Parallel port/floppy disc interface for it (long before USB came along plus Acorn insisted on propitiatory floppy drives, and the was no room for both ports on the case). Ahhhh those were the days....
-
-
-
Wednesday 22nd August 2018 09:00 GMT Charlie Clark
Re: Why no ARM servers?
There are some ARM servers but the biggest barrier has been the lack of a common boot framework which makes them less suitable as drop-in replacements: x86 boxes can be swapped in and out but this is not true for ARM.
This leads to the chicken and egg situation of no one manufacturing in volume because the lack of demand and no demand because of the lack of supply. For most data centres the CPU is not the most important cost. This means that ARMs are used where things like power draw are key.
Things will change as the eco-system changes and the workloads become more suitable to ARM's hardware optimisations.
-
-
Wednesday 22nd August 2018 17:12 GMT Charles 9
Re: Why no ARM servers?
Right. Most ARM systems weren't built with modular, replaceable components in mind which is how servers normally operate (hot swapping, automatic failover, etc.). Most ARM systems today are SoCs full of trade secret sauce (because Qualcomm, Mediatrk, etc. are at each other's throats). Portable ARM systems are a whole different kettle of fish from server ARM systems. What you describe is relatively new and will need time to reach a broad-enough consensus.
-
-
-
-
Wednesday 22nd August 2018 09:39 GMT TiddlyPom
Re: Why no ARM servers?
This is where open source applications have the benefit. If you have the source code (assuming it is pretty much architecture independent) then shifting platform becomes little more than a recompile. As a trivial example, I have developed Java applications on x86 then run them on ARM (on a Raspberry Pi). I have also written Qt C/C++ applications on x86 and recompiled them for ARM. KVM/QEMU virtualization works fine on ARM (but obviously depends on the features of the particular ARM chip).
-
Wednesday 22nd August 2018 11:53 GMT Charlie Clark
Re: Why no ARM servers?
This is where open source applications have the benefit. If you have the source code (assuming it is pretty much architecture independent)
The open source claim is a red herring here; open source has its own benefits but architecture independence isn't one. More important is whether the source code contains optimisations for the architecture and how good the compiler is at optimising. For example, I've seen benchmarks where Intel's compiler produces significant improvements on x86 over gcc.
Still, LLVM and other initiatives have brought general improvements for all and we're reaching a point where compilers can always produce better results than people: optimisations can generally be formalised which makes them a good subject for automation.
But, as others have noted, it's not just about optimised code: drivers for storage and networking have to work just as well and this is easier on "industry standard" x86 than for any particular custom ARM SoC. Fortunately, things are improving here, too.
-
Thursday 18th April 2019 13:55 GMT Alan Brown
Re: Why no ARM servers?
"For example, I've seen benchmarks where Intel's compiler produces significant improvements on x86 over gcc."
Almost all of this is down to the Intel compiler using undocumented cheats vs the GCC one playing strictly by the rules. Remember when MS used to pull that with undocumented APIs?
-
Thursday 18th April 2019 14:00 GMT Alan Brown
Re: Why no ARM servers?
"drivers for storage and networking have to work just as well "
For some values of "work".
The vast majority of proprietary drivers are barely functional pieces of utter crap only built and tested on one or two versions of windows.
Companies won't publish APIs in case they get sued for patent infringements and they won't release source code because they usually obtained it from somewhere else without getting ownership of the copyright.
-
-
Wednesday 22nd August 2018 12:30 GMT Justin Clift
Re: Why no ARM servers?
> As a trivial example, I have developed Java applications on x86 then run them on ARM (on a Raspberry Pi). I have also written Qt C/C++ applications on x86 and recompiled them for ARM.
Go (the language) will have ARMv8 support in the next release (1.11). Been testing it already in production deployment with the recent betas, and it works well.
-
-
-
Wednesday 22nd August 2018 09:48 GMT GrumpenKraut
Re: Why no ARM servers?
In recent news: World's largest ARM supercomputer is headed to a nuclear security lab.
-
Wednesday 22nd August 2018 09:54 GMT Herring`
Re: Why no ARM servers?
I remember talking to an HPC chap a couple of years back. I asked the question why people weren't interested in the better FLOPS/Watt that you could get from ARM. His response was that the (proprietary) software that he used was licensed per core and that that cost vastly outweighed the savings in power/cooling/hardware.
-
Wednesday 22nd August 2018 12:28 GMT Anonymous Coward
Re: Why no ARM servers?
> I still don't understand why no-one is making ARM-based servers.
ARM servers do exist, though they seem to be just now starting to hit the market.
→ https://www.gigabyte.com/Press/News/1622
→ https://www.servethehome.com/cavium-thunderx2-review-benchmarks-real-arm-server-option
You can rent ARM based cloud servers online already:
With those Scaleway servers, the ARM64 (ARMv8) ones are very good, and stable in operation (from personal experience).
The ARMv7 ones though are generally underpowered, and have a reputation for being a bit unstable. Ok for playing around with though.
Workstations look to be arriving as well:
-
Wednesday 22nd August 2018 12:40 GMT Anonymous Coward
Re: Why no ARM servers?
Also:
→ https://www.phoenicselectronics.com/gigabyte.html
No pricing though. :(
-
-
Wednesday 22nd August 2018 20:16 GMT chasil
Re: Why no ARM servers?
ARM 64-bit support only emerged in 2011, and it's vastly different from the 32-bit ISA (I understand it's much more like MIPS).
This also came late to x86 with the Opteron in 2003.
MIPS owned supercomputing in the 90s starting with the 1991 release of the 64-bit R4000.
The ARM 32-bit ISA had design decisions that limited performance. I would say that Sophie's ISA was perfect for an '80s Acorn, but not so much for a Cray.
https://www.jwhitham.org//2016/02/risc-instruction-sets-i-have-known-and.html
-
-
-
Wednesday 22nd August 2018 09:40 GMT Steve Todd
Re: So the A64FX is officially at 7nm engraving
To be fair, Intel's 10nm process is about the same size & density that other manufacturers are claiming for their 7nm processes. But yes, they have lost their technological lead in fab processes (there was a time that they were about 1.5 nodes ahead of everyone else).
-
Wednesday 22nd August 2018 11:48 GMT Anonymous Coward
Re: So the A64FX is officially at 7nm engraving
It's interesting that the package only appears to need 594 pins.
I'm guessing that it would need to provide a similar number of data lines to comparable amd64 chips to really be a viable alternative, so I wonder if the reduced pin count reflects lower or more easily managed power requirements.
-
-
-
Wednesday 22nd August 2018 21:29 GMT Anonymous Coward
Re: Bah!
Well, at a mere 40MW you could run the backup generator off one of the smaller* Wartsila Diesel engines. You could even use the waste cooling water from the computer to keep the engine ready for a warm start.
The UPS must be a bit big though. Perhaps a new venture for Tesla; really big UPSes.
*For some value of smaller, the biggest one is over 100MW IIRC.
-
-
Friday 24th August 2018 05:02 GMT Anonymous Coward
ARM64 on Fujitsu enterprise systems?
Looking under the hood of A4FX, it seems as it was built by the same shop that is responsible for the Fujitsu SPARC64 processors, this indicates a lot in common in their specifications. The K-Computer predecessor uses SPARC64-VIIIfx chips which helped to kick-off SPARC64-X processors on Fujitsu enterprise field. Two years ago, Reg told a story about the transition of SPARC64 to ARM in Fujitsu https://www.theregister.co.uk/2016/09/07/fujitsu_arm_supercomputer_delay/, it was almost a year before Oracle finally killed Solaris and SPARC, and now, it's clear that A4FX side show is the last nail in the coffin of SPARC. The only question is what is the future of ARM64 on Fujitsu enterprise systems and whether ORACLE knows about it.
Anonymous coward.