User avatar
Gavinmc42
Posts: 3458
Joined: Wed Aug 28, 2013 3:31 am

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 11:41 am

Been thinking about this some more.

We are in effect talking about another Architecture, Aarch64 or ARMv8A, no longer is it just x86, X86_64....
Intel, AMD, ARMv8A and then those older things like PowerPC. SPARC, MIPS....
Raspbian is 32bit at the moment, but Aarch64 will perhaps takeover in the smaller size arenas of IoT and SBC's.
There is lot already in Aarch64, but there is also big holes.
Aarch64 Linux is fairly new, Raspberry Pi's could be the trigger for a major acceptance of Aarch64.

"The Raspberry Pi Desktop OS 64Bit", the title is not "THE RASPBIAN Pi Desktop".

It should be "Raspberry Pi Desktops", plural, and some of these may not even be Linux.
Rust now works on Gentoo Aarch64, that means Redox could be compiled.
Redox is still early stages with many drivers still to be written for x86_64, however a Pi Redox would only need a few drivers.
It has a microkernel, sounds like fun :D

The Pi3B+ is probably as good as it is going to get for the next few years.
32bit Raspbian may get small improvements and small changes will go upstream.

But do we who live in the future need 32bit Linux legacy support.
What is wrong with trying something new, faster, better? Shiny??? :lol:

Retro Pie, 6502 emulation, Acorn OS's.... that's all looking back with Rose coloured glasses.

Kids using Pi's now will be the future coders, do they need to be mired in 32bit land?
Can we not show them a glimpse of the future they will be making, now, this year?

Tertiary Students are going out into the workforce, let them go 64bit ready.

I was going to mention 128bit CPU's but that was for next year or the the year after.
Or we can just use NEON this year, 64bit ARMv8A and128bit NEON ;)
I'm dancing on Rainbows.
Raspberries are not Apples or Oranges

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 23083
Joined: Sat Jul 30, 2011 7:41 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 12:27 pm

Problem with RISCV is that it is early in its life. It's not had the same level of optimisation that x86 has had, or ARM to a lesser extent. Of course they can take on lessons learned from these other architectures, but a lot of the techniques for really making fast chips still need to be applied. ARM are behind x86, but catching, but each architecture improve gains less each time. Arm still get 2,3 times faster over a generation, x86 being far ahead only get a few percentage points. But RISCV is still a baby in comparison to both of them. Not sure of the instructions per clock to get with RISCV compared with x86 or ARMv8 but it's surely way behind at the moment.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
"My grief counseller just died, luckily, he was so good, I didn't care."

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 12:29 pm

Heater wrote:
Mon Apr 23, 2018 11:21 am
Strangely enough the RISC V specification includes a 128 bit version.

They were designing the ISA to be extensible in all kind of ways, including 32 bit and 64 bit operation, so they included the possibility of 128 bit, almost as a joke.

Turned out those guys building gigantic data warehouses are taking the idea seriously...
Its always been available as __int128 in GCC.
Though you probably need a 64-bit platform, so its no harder than doing 64-bit ints on a 32-bit platform.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 23083
Joined: Sat Jul 30, 2011 7:41 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 12:30 pm

Gavinmc42 wrote:
Mon Apr 23, 2018 11:41 am

Tertiary Students are going out into the workforce, let them go 64bit ready.
They already are. Because really, in any high level language, the number of bits on the underlying HW, for most people, is irrelevant. On the whole, the code I write is bit depth independent, unless using HW registers and even them, it's fairly irrelevant whether they are 8,16,32,64 or 128 bits in size.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
"My grief counseller just died, luckily, he was so good, I didn't care."

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 12:34 pm

Apart from being open source, and hardware 128 bit integers, what does RISC-V offer that aarch64 on the Pi doesn't?
Perhaps too broad a question.
I took a quick look and preferred aarch64.

Heater
Posts: 12974
Joined: Tue Jul 17, 2012 3:02 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 2:02 pm

jahboater,
Apart from being open source, and hardware 128 bit integers, what does RISC-V offer that aarch64 on the Pi doesn't?
If the idea of an open standard for the most important interface in computing, that between hardware and software, does not appeal to you than I don't think I can sell you on the idea.
I took a quick look and preferred aarch64.
I'd be interested to hear the reasons for that preference.

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 2:21 pm

Heater wrote:
Mon Apr 23, 2018 2:02 pm
If the idea of an open standard for the most important interface in computing, that between hardware and software, does not appeal to you than I don't think I can sell you on the idea.
It does appeal, don't get me wrong. I was just wondering if there were any good technical reasons as well to get enthusiastic about.
I'd be interested to hear the reasons for that preference.
It was some time ago, but one thing I do remember was the large number of "optional" features - which I really don't like (though I understand the reason why they are there). Unless the programmer/compiler just uses the guaranteed subset only, then its added complication.

ARM is a bit like that, but aarch64 draws a line in the sand with a decent, complete, instruction set, guaranteed presence of NEON floating point and so on.

Heater
Posts: 12974
Joined: Tue Jul 17, 2012 3:02 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 4:38 pm

jahboater,

Yes, all those RISC V options seem like a mess at first sight.

As you probably know having these standardized options is great. When I'm squeezing a RISC V into an FPGA I can make trade offs. Dump multiply and divide, use 16 or 32 registers etc, all the while knowing that my compiler, GCC or Clang, will support the thing.

Heck, I'm writing my own RISC V hardware design, I don't want to have to include anything but the minimum. https://github.com/ZiCog/sodor-spinal. At least not initially.

For sure when Linux running RISC V machines start to appear they will be built on a standard "platform", that combination of standard options that you need for a Linux machine.

The Vector/SIMD stuff might not alway be present, but such is the case in the x86 and ARM world.

I'd say the RISC V options are a lot less chaotic than the ever shifting instruction sets of x86 and ARM.

Heater
Posts: 12974
Joined: Tue Jul 17, 2012 3:02 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 4:46 pm

jamesh,

True enough, it's early days for RISC V.
...they can take on lessons learned from these other architectures,...
I suspect that Intel chips are designed by guys out of Berkeley and such. They use optimizations invented in such places. I see no reason RISC V chips could not use the same.
Not sure of the instructions per clock to get with RISCV compared with x86 or ARMv8 but it's surely way behind at the moment.
RISCV does not have an IPC. It's only an ISA spec. after all.

Actual processors implementing that ISA may well have an IPC. The best of them, e.g. BOOM, are up there with ARM. Not bad going for the work of a bunch of grad students :)

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 5:03 pm

Heater wrote:
Mon Apr 23, 2018 4:38 pm
Yes, all those RISC V options seem like a mess at first sight.

As you probably know having these standardized options is great. When I'm squeezing a RISC V into an FPGA I can make trade offs. Dump multiply and divide, use 16 or 32 registers etc, all the while knowing that my compiler, GCC or Clang, will support the thing.
Yes of course. I agree they are useful for just those reasons. x86 is just not in that market!
I'd say the RISC V options are a lot less chaotic than the ever shifting instruction sets of x86 and ARM.
Yes, ARM is a bit messy until aarch64 came along. But thats not true of x86 that I know of.
They introduce new features all the time of course, but once included they are always available. You wont see processors with odd features missing. Until AVX512 comes along that is. AVX512F provides a reasonable guaranteed foundation, but you can see why some AVX512 stuff is optional. Take this instruction for example:-

GF2P8AFFINEINVQB — Galois Field Affine Transformation Inverse

:)

Heater
Posts: 12974
Joined: Tue Jul 17, 2012 3:02 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 5:29 pm

"GF2P8AFFINEINVQB"

Good grief, what? That's an instruction?!

I'm getting too old for this game.

Sounds like an instruction set designed by somebody from Wales :)

Didn't the AAA instruction go away when AMD came up with the 64 bit version of x86?

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 5:36 pm

Heater wrote:
Mon Apr 23, 2018 5:29 pm
"GF2P8AFFINEINVQB"

Good grief, what? That's an instruction?!

I'm getting too old for this game.

Sounds like an instruction set designed by somebody from Wales :)
You certainly don't need it for twiddling an LED :)
AVX512 is actually a good reason for using HLL's instead of assembler, its quite hard for a human to even understand how to use some of these instructions.
They don't just do floating point stuff by the way, for example even the legacy SSE4.2 set introduced some string handling instructions, such as:-

PCMPISTRI - Packed Compare Implicit Length Strings, Return Index

Hard to understand but very powerful.

Thats CISC for you, and with 10nm you have the space.
Didn't the AAA instruction go away when AMD came up with the 64 bit version of x86?
Yes.

jamesh
Raspberry Pi Engineer & Forum Moderator
Raspberry Pi Engineer & Forum Moderator
Posts: 23083
Joined: Sat Jul 30, 2011 7:41 pm

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 8:22 pm

Heater wrote:
Mon Apr 23, 2018 4:46 pm
jamesh,

True enough, it's early days for RISC V.
...they can take on lessons learned from these other architectures,...
I suspect that Intel chips are designed by guys out of Berkeley and such. They use optimizations invented in such places. I see no reason RISC V chips could not use the same.
Not sure of the instructions per clock to get with RISCV compared with x86 or ARMv8 but it's surely way behind at the moment.
RISCV does not have an IPC. It's only an ISA spec. after all.

Actual processors implementing that ISA may well have an IPC. The best of them, e.g. BOOM, are up there with ARM. Not bad going for the work of a bunch of grad students :)
Early days indeed. WhIch means the silicon implementations are not yet fast enough, and won't be until, probably, Pi6. And by that time, where will ARM be?

Imagine, the Pi5 being slower than the Pi4. That'll sell. Not.
Principal Software Engineer at Raspberry Pi (Trading) Ltd.
Contrary to popular belief, humorous signatures are allowed. Here's an example...
"My grief counseller just died, luckily, he was so good, I didn't care."

User avatar
Gavinmc42
Posts: 3458
Joined: Wed Aug 28, 2013 3:31 am

Re: The Raspberry Pi Desktop OS 64 bit

Mon Apr 23, 2018 10:01 pm

Imagine, the Pi5 being slower than the Pi4. That'll sell. Not.
That is a hint Pi4 will be ARM based :lol:

2 years before a Pi4 and 2 more years for a RISC-V Pi5?
I think 4 years is too short to get RISC-V faster than ARM A72/75

I am interested in if it gets any Machine Learning hardware, a bunch of general purpose ML/CV/NN CPU's.
ARM is up to 2nd generation of Object Detection ;)
And is the ARM ML machine learning processor real silicon yet?

I will not be going back to a 32bit OS just for those.
I'm dancing on Rainbows.
Raspberries are not Apples or Oranges

User avatar
bensimmo
Posts: 4128
Joined: Sun Dec 28, 2014 3:02 pm
Location: East Yorkshire

Re: The Raspberry Pi Desktop OS 64 bit

Tue Apr 24, 2018 3:57 pm

Once Broadcom buys Qualcomm, they (RPi) will have nice fast ARM chips to plug in.
At which point Windows will use their 64bit OS on it and also 64bit AndroidOS will appear officially. Google also release in MagPi freebee HAT based on a TPU.
All will be happy as PC sales crumble and become a niche for *coin crunchers (since gamers cannot afford the discrete GPU anymore) and small powerful deskop take over with less powerful IoT boards running the cutdown versions.
All still for $10 and $35 as things scale up, even with rainbow coloured boards. (64bit colour banding of course)


Personally I couldn't careless for RISC untill they have a usable processor competeing in either the Intel/AMD Desktop market or the ARM based market.
ARM, x86 seem to be doing well at what they do for a price I'm prepared to pay.
Works for me.


Yeah, time I should be going home ...

deepakdeshp
Posts: 94
Joined: Thu Oct 26, 2017 7:23 pm

Re: The Raspberry Pi Desktop OS 64 bit

Thu May 10, 2018 4:33 pm

jamesh wrote:
Wed Sep 06, 2017 8:40 am
feelslikeautumn wrote:
Wed Sep 06, 2017 6:40 am
I don't think it necessarily works like that. From a ff bug report:
Of the 15m sold, a large number of raspberries will be sitting unused in drawer, a large number will be doing something embedded, a large number are too slow to do anything reasonable on a desktop. Pi usage will be a tiny percentage of the total number of Firefox users.
Only disadvantage, no Windows.
No Windows and no viruses is the biggest advantage.

ejolson
Posts: 3251
Joined: Tue Mar 18, 2014 11:47 am

Re: The Raspberry Pi Desktop OS 64 bit

Thu May 10, 2018 5:00 pm

bensimmo wrote:
Tue Apr 24, 2018 3:57 pm
Personally I couldn't care less for RISC untill they have a usable processor competeing in either the Intel/AMD Desktop market or the ARM based market.
You may be interested to know that ARM originally meant Acorn RISC Machine. While complicated instructions have been added to recent models as well as Spectre-capable speculative execution (not the Pi), I think many still consider ARM in the tradition of a RISC design--at least it has a reasonably large set of registers.

Heater
Posts: 12974
Joined: Tue Jul 17, 2012 3:02 pm

Re: The Raspberry Pi Desktop OS 64 bit

Thu May 10, 2018 5:29 pm

bensimmo,
Personally I couldn't care less for RISC until they have a usable processor competing in either the Intel/AMD Desktop market or the ARM based market.
This makes no sense in so many ways:

The ARM is a RISC processor. In part and perhaps indirectly inspired by the work of Patterson and co. at UCB with their RISC architecture work back in the 1980's.

Intel and AMD essentially convert incoming instructions on the fly for their RISC engines to deal with efficiently.

If you mean the "RISC V" guys, well, they don't make a processor or any kind of chip. They just specify an instruction set that is open for anyone to use.

If you actually meant "RISC V", well yeah, I agree. Perhaps this whole idea takes off and we see people implementing RISC V machines with the performance you desire.

Or perhaps not...

In my dreams the Raspberry Pi Foundation is in just the right place to take on a RISC V design. For education's sake.

Either way, I'm having fun running a RISC V core on my FPGA boards, and now I'm trying to create my own RISC V core design.

jahboater
Posts: 4595
Joined: Wed Feb 04, 2015 6:38 pm

Re: The Raspberry Pi Desktop OS 64 bit

Thu May 10, 2018 7:26 pm

ejolson wrote:
Thu May 10, 2018 5:00 pm
While complicated instructions have been added to recent models as well as Spectre-capable speculative execution (not the Pi), I think many still consider ARM in the tradition of a RISC design--at least it has a reasonably large set of registers.
For interest/amusement, Intel x86 Skylake has 180 physical integer registers and 168 vector registers.
(The user see's only 16, same as the 32-bit ARM though).
If the size of the register file makes a cpu RISCy then x86 wins :)

User avatar
bensimmo
Posts: 4128
Joined: Sun Dec 28, 2014 3:02 pm
Location: East Yorkshire

Re: The Raspberry Pi Desktop OS 64 bit

Thu May 10, 2018 8:34 pm

Heater wrote:
Thu May 10, 2018 5:29 pm
bensimmo,
Personally I couldn't care less for RISC until they have a usable processor competing in either the Intel/AMD Desktop market or the ARM based market.
This makes no sense in so many ways:

The ARM is a RISC processor. In part and perhaps indirectly inspired by the work of Patterson and co. at UCB with their RISC architecture work back in the 1980's.

Intel and AMD essentially convert incoming instructions on the fly for their RISC engines to deal with efficiently.

If you mean the "RISC V" guys, well, they don't make a processor or any kind of chip. They just specify an instruction set that is open for anyone to use.

If you actually meant "RISC V", well yeah, I agree. Perhaps this whole idea takes off and we see people implementing RISC V machines with the performance you desire.

Or perhaps not...

In my dreams the Raspberry Pi Foundation is in just the right place to take on a RISC V design. For education's sake.

Either way, I'm having fun running a RISC V core on my FPGA boards, and now I'm trying to create my own RISC V core design.
My bad, I mean a RISC-V processor, not the architecture. :oops: I probably assumed, given the topic of talk (or a typo, it was a while ago now) the V would be there as I type.
So, I'm not interested until I can buy it for a similar price, for a similar performance as my desktop processors.
OR it does something where it is cost effective.

I don't see what Raspberry Pi has to do with it and what it has to do with education's sake. Unless that education is teaching and showing how to take the blueprint, to semiconductor processes (or Organic), deposition, lithography, manufacture etc, troubleshooting and redesign and then to interface that to make a working processor and board. So the physical engineering processes, it is no different to and x86 board or an ARM board.

What the OS actually runs on is of little consequence to education, as long it it works, is affordable.

I can see ideology, but that is a different thing.

Now you are doing the education thing, which is ace :-) Not my sort of thing ... yet.
Though if you can just develop it and replace these Nano/Uno boards or get it to the equivalent 'Nintendo Switch' speeds (which are not exuberant with Octa-core, 4×ARM Cortex-A57 & 4×ARM Cortex-A53 @ 1.020 GHz*) at a similar cost today...

Return to “General discussion”