Raspberry Pi is a Low-Power, Credit-Card Sized Computer

[NeoGen]

Upon starting L4T (Linux For Tegra) you are greeted with a screen that the staunch AMD supporter will make thinking "isn't that the logo of the GPUs from that other brand? F#**ing nVidia?", while the more opportunistic user will think "isn't that the logo of the brand that works out of the box when installing Linux? Will I have CUDA running on ARM soon?"

I would think "devil be gone from my machine!" *sprays it with holy water*

[NeoGen]

That's a beautiful family photo

By the way, there' a new VoCore2... still very very small.

[Dirk Broer]

That's a beautiful family photo

By the way, there' a new VoCore2... still very very small.

New family members caused a new family photo:

[Dirk Broer]

I jumped on the Raspberry Pi wagon when I saw the first model that I liked, the Pi 1 Model B+, or as the official name then was: Raspberry Pi Model B+.
Since that date -five years ago- the Raspberry Pi has made much progress

Model/Feature	Pi 1 B+	Pi 2 B 1.1	Pi 2 B 1.2	Pi 3 B	Pi 3 B+	Pi 4 B
Year	July 2014	February 2015	October 2016	February 2016	March 2018	June 2019
Instruction set	ARMv6Z (32-bit)	ARMv7-A (32-bit)	ARMv8-A (64/32-bit)	ARMv8-A (64/32-bit)	ARMv8-A (64/32-bit)	ARMv8-A (64/32-bit)
SOC	BCM2835	BCM2836	BCM2837	BCM2837	BCM2837B0	BCM2711
CPU	700 MHz ARM11	900 MHz ARM Cortex-A7	900 MHz ARM Cortex-A53	1200 MHz ARM Cortex-A53	1400 MHz ARM Cortex-A53	1500 MHz ARM Cortex-A72
FPU	VFPv2	VFPv3 + NEON	VFPv4 + NEON	VFPv4 + NEON	VFPv4 + NEON	VFPv4 + NEON
GPU	VideoCore4, OpenGL ES 2.0, OpenVG 1080p30, H.264 high-profile encode/decode @250 MHz	VideoCore4, OpenGL ES 2.0, OpenVG 1080p30, H.264 high-profile encode/decode @250 MHz	VideoCore4, OpenGL ES 2.0, OpenVG 1080p30, H.264 high-profile encode/decode @250 MHz	VideoCore4, OpenGL ES 2.0, OpenVG 1080p60, H.264 high-profile encode/decode @400 MHz	VideoCore4, OpenGL ES 2.0, OpenVG 1080p60, H.264 high-profile encode/decode @400 MHz	VideoCore6, OpenGL ES 3.x, OpenVG 4000p60, H.265 high-profile encode/decode @500 MHz
GPU GFLOPS	24	24	28.8	28.8	28.8	?
RAM per Core	512 MB	256 MB	256 MB	256 MB	256 MB	256, 512 or 1024 MB
Pipeline depth	8	8	8	8	8	15
OoOE (out-of-order execution)	No	No	No	No	No	Yes, 5-wide dispatch
Execution ports	?	5	2	2	2	8
Super Scalar	?	partial	2-way super scalar	2-way super scalar	2-way super scalar	3-way super scalar
DMIPS/MHz	1.25	1.90	2.24	2.24	2.24	4.72

I'll try to compare this with likewise clocked Athlons

Model/Feature	Athlon K7 "Argon"	Athlon K75 "Orion"	Athlon "Thunderbird"	Athlon XP "Palomino"	Athlon XP "Thoroughbred"	Athlon 64 2600+ "Lima"
Year	October 1997	March 2000	June 2000	Somewhere 2001	Somewhere 2002	Somewhere 2003
Instruction set	x86 (32-bit)	x86 (32-bit)	x86 (32-bit)	x86 (32-bit)	x86 (32-bit)	X86-64 (64-bit)
Speed	700 Mhz	900 Mhz	900 Mhz	1200 MHz	1400 MHz	1600 MHz
FPU	MMX, Enhanced 3DNow!	MMX, Enhanced 3DNow!	MMX, Enhanced 3DNow!	MMX, Enhanced 3DNow! SSE	MMX, Enhanced 3DNow! SSE	MMX, SSE, SSE2, SSE3, Enhanced 3DNow!, NX bit, AMD64, AMD-V

This gets stuck in the integrated graphics. When we also take the production process into account (at how many nanometers the CPU die is processed),
it makes more sense to start with the lesser known AMD G-Series G-T24L, Microarchitecture Bobcat, Platform Brazos and Processor core Ontario.
That starts at 64-bit though, but it is single core, like the ARM11.

Model/Feature	AMD G-Series G-T24L	AMD Z-Series Z-01	AMD C-Series C-50	AMD G-Series G-T44R	AMD E1-1200	AMD A4-5000
Year	March 2011	June 2011	January 2011	January 2011	June 2012	May 2013
Instruction set	X86-64 (64-bit)	X86-64 (64-bit)	X86-64 (64-bit)	X86-64 (64-bit)	X86-64 (64-bit)	X86-64 (64-bit)
Speed	800 Mhz	1000 Mhz	1000 Mhz	1200 Mhz	1400 Mhz	1500 Mhz
FPU	MMX, SSE, SSE2, SSE3, SSE4a, ABM, NX bit, AMD64, AMD-V	MMX, SSE, SSE2, SSE3, SSE4a, ABM, NX bit, AMD64, AMD-V	MMX, SSE, SSE2, SSE3, SSE4a, ABM, NX bit, AMD64, AMD-V	MMX, SSE, SSE2, SSE3, SSE4a, ABM, NX bit, AMD64, AMD-V	MMX, SSE, SSE2, SSE3, SSE4a, ABM, NX bit, AMD64, AMD-V	MMX, SSE, SSE2, SSE3, SSE4a, SSE4.1, SSE4.2, ABM, AES, AVX, BMI1, F16C, NX bit, AMD64, AMD-V
GPU	Radeon HD 5000 "Evergreen"	Radeon HD 6250 "Wrestler"	Radeon HD 6250 "Wrestler"	Radeon HD 6250 "Wrestler"	Radeon HD 7310 "Ontario"	Radeon HD 8330 "Kalindi"
GPU FLOPS	n/a	64	64	64	80	127.2

[NeoGen]

And the family keeps growing... can it be arranged in a sort of family tree to see each generation and its descendants?

[Dirk Broer]

And the family keeps growing... can it be arranged in a sort of family tree to see each generation and its descendants?

Though the oldest records of the Raspberry Pi family claim that the arch version was round and actually tasted like raspberry, I'll confine myself to those models that were designed by the Raspberry Pi Foundation.
In the beginning that was but one Pi and it's model was..Model B. Not a bad beginning when you're constructed in the same country that gave us Terry Pratchett. Mind you: it was still 2012 back then.
Model B (aka 0002) later -2013- got a brother named Model A (aka 0008), which was cheaper and simpler by virtue of having no network capability, just one USB port and having no more than 256MB of RAM.
Introduced in June 2014 there was an illegitimate kind of offspring introduced that needed a Compute Module development board to be able to function, the so-called Compute Module(aka 000d), known for its hideous prices at the resellers, especially the combo of development board and compute module, charging hundreds of euro's for them. They can stick them where the sun doesn't shine! Actually a very painful experience, viewed by the GPIO pins...
In 2015 the youngest family member of the first generation (though some claim it was a child of Model A) saw the light: the Raspberry Pi Zero (aka 900092), that for the sake of cheapness had lost all that was not strictly needed for basic operation, including a large part of PCB surface.

That first generation has in common that they are all based upon the Broadcom BCM2835 SOC, featuring an ARMv6Z architecture ARM11 CPU and a VideoCore IV GPU. There are some minor variants/descendants worth mentioning of the first generation:
Variants:
Where there is profit there will be people or companies that will help themselves to a share of it, sometimes even without knowledge of the original manufacturer. In a klatch-as-klatchian-can the models 0003 and 0004 can be said to be minor variants of 0002, but China-made 000d -that came out in both red and blue instead of the familiar green- defies family resemblance. Revision 2.1 of the Model B (aka 00oe) was again UK-made, while 000f was again China-manufactured, but green this time.

Descendants:
Model B sprouted Model B+ (aka 0010) in July 2014. B+ gave us the basic form of the modern Raspberrie Pi's, including the four USB 2.0 ports. The model B+ was also manufactured in China, using a fitting red PCB and also reporting as 0010, just like the original.

Model A sprouted Model A+ (aka 0012) -we observe a pattern here- in November 2014. A+ suffered from dwarf-growth, albeit not in such a severe form as their youngest sibling(?) Pi Zero. It even managed in a later form (aka 0016) to get more memory, 512 MB.

Model Zero has, besides it very basic 1.2 version a 1.3 version (aka 900093, with MIPI interface), a W version (aka 9000C1, Wireless plus Bluetooth) and a sort of illegal WH version (same as previous but also with onbard GPIO headers).

November 2014 family picture of the first generation, taken before the advent of the Raspberry Pi Zero.

The 2nd generation was kicked off by the introduction in February 2015 of the Raspberry Pi Model 2B (aka a01041), a model that would have made Hamlet proud -if he had lived that long. The Model 2B was a jump foreward, as it used a BCM2836 SOC, featuring a Quad-core ARMv7-A architecture ARM Cortex-A7 CPU and.... a VideoCore IV GPU. RAM was increased to 1GB and the speed was increased from the original 700 to 900 MHz.
After the introduction in 2016 of the 3rd generation Model 2B was produced in a variant 2B 1.2 (aka a22042) which used a BCM2837 SOC, featuring a Quad-core ARMv8-A architecture ARM Cortex-A53 CPU and.... a VideoCore IV GPU. This quickly introduced hybrid 2/3 version may be the reason that there are no further descendants and/or subversion of the Model 2. The (un)availabiilty of the BCM2836 SOC may play another role here.

The 3rd generation first saw the light with the introduction in February 2016 of the Raspberry Pi Model 3B (aka a02082), a model that got much lamented for not being named Model 3.14. The Model 3B was another jump foreward, as it used a BCM2837 SOC, featuring a Quad-core ARMv8-A architecture ARM Cortex-A53 CPU and.... a VideoCore IV GPU. Speed was increased to 1200 MHz, and WiFi and Bluetooth were standard onboard features.
Model 3B has variants produced in Japan (aka a32082) and in Brazil (aka a02082), the later made with a blue-green PCB.
A descendant of Model 3B is the Raspberry Pi Model 3B+ (aka a020d3) that differs in a still higher clockspeed (1400 MHz), faster ethernet, two-band WiFi, later bluetooth version and a metal-covered SOC.
The third generation also features a new Compute Module development board with a choice of two Compute Modules: Compute Module 3 (aka a020a0, with 4GB eMMC) and a Compute Module 3 lite (with provision for a SD-card)

[Dirk Broer]

Let's say you bought yourself the latest Raspberry Pi, the Raspberry Pi 4B with 8GB of RAM, what can you do with it? And what couldn't be done with earlier Pi's?

Let me first stress that the 8GB, together with the ARM Cortex-A72 allow for a 64-bit OS -e.g. Raspberry Pi OS to be used to the fullest extent.
From the Raspberry Pi 3B on (and even the 2nd edition of the Raspberry Pi 2B) the board had a 64-bit ARM Cortex-A53, but with only 1 GB of RAM. The Pi 4B was originally introduced with 1, 2 and 4 GB, the latter just having enough RAM to be able to run a 64-bit OS (and more than enough to run a 32-bit OS). The 1 GB model has been withdrawn since a few months and a 8 GB model has now been added to the line. That is as much RAM as is there in the PC of my missus, and twice the RAM as in the PC of my daughter! (she uses a hand-me-down Socket 775 Dell that won't accept more than 4GB and is due for a Ryzen 5 upgrade, though with that new B550 chipset a Ryzen 3 from the 3000 series might do too...)

What projects have apps for ARM/Linux? And do they cater 64-bit?
*Does anyone have an overview for WCG? COVID-19 and MIP can be done 32-bit natively and 64-bit with a hack at least*

If you have installed a 64-bit OS and want/need to run 32-bit apps, you might need to install 32-bit libraries to do so.
In case of wanting to run 32-bit apps under a 64-bit OS you get a message like "This project doesn't support computers of type aarch64-unknown-linux-gnu" from that project.
You then need to add:

<options>
  <alt_platform>arm-unknown-linux-gnueabihf</alt_platform>
</options>

to the cc_config.xml file

[Dirk Broer]

Nvidia Jetsons for the future.

What should Nvidia do to make their ARM products more competitive? There are several strategies, but let's start with a sore point on my part: the price.
The 2GB Nano comes at a price halfway the 4GB Pi 4 and the 8GB Pi4 here, while the 4GB Nano is more expensive than the 8GB Pi 4 and even more expensive than the 4GB Odroid-N2+!
The next model, the Xavier NX comes at a whopping 500+ Euro's.....

My proposals for Nvidia:

• Lower the price of the 2GB Nano to $35, the 4GB Nano to $55, so they can compete with the Raspberries. The USP (Unique Selling Point) might be the better GPU. CPU-wise the Raspberry Pi 4's are one generation ahead already.
• Bring out a 8GB Nano. This to prevent messages in BOINC like this for my Jetson-Nano2GB:
  Rosetta@home 4/9/2021 4:00:11 PM Message from server: Rosetta needs 6675.72 MB RAM but only 1879.68 MB is available for use.
  Amicable Numbers 4/11/2021 12:25:21 AM Message from server: Amicable Numbers up to 10^21 needs 2861.02 MB RAM but only 1879.68 MB is available for use.
  Even the 4GB would get this from Ralph and Rosetta, so a 8GB would be very nice, not only to keep up with the Raspberry Pi's, but to do meaningful BOINC work.
• Bring out more versions of the Xavier NX, and lower the price. At this moment we only have the far-too-expensive 8GB model, make a 6GB and a 12 GB (16 GB?) model too, and make it a bit more affordable.
• Bring OpenCL support for your GPUs to the Linux/ARM platform. CUDA works at least but for BOINC (and other applications) to really get things into gear, OpenCL support for the GPU is needed.

[Dirk Broer]

Nvidia Jetsons for the future.

What should Nvidia do to make their ARM products more competitive? There are several strategies, but let's start with a sore point on my part: the price.
The 2GB Nano comes at a price halfway the 4GB Pi 4 and the 8GB Pi4 here, while the 4GB Nano is more expensive than the 8GB Pi 4 and even more expensive than the 4GB Odroid-N2+!
The next model, the Xavier NX comes at a whopping 500+ Euro's.....

My proposals for Nvidia:

• Lower the price of the 2GB Nano to $35, the 4GB Nano to $55, so they can compete with the Raspberries. The USP (Unique Selling Point) might be the better GPU. CPU-wise the Raspberry Pi 4's are one generation ahead already.
• Bring out a 8GB Nano. This to prevent messages in BOINC like this for my Jetson-Nano2GB:
  Rosetta@home 4/9/2021 4:00:11 PM Message from server: Rosetta needs 6675.72 MB RAM but only 1879.68 MB is available for use.
  Amicable Numbers 4/11/2021 12:25:21 AM Message from server: Amicable Numbers up to 10^21 needs 2861.02 MB RAM but only 1879.68 MB is available for use.
  Even the 4GB would get this from Ralph and Rosetta, so a 8GB would be very nice, not only to keep up with the Raspberry Pi's, but to do meaningful BOINC work.
• Bring out more versions of the Xavier NX, and lower the price. At this moment we only have the far-too-expensive 8GB model, make a 6GB and a 12 GB (16 GB?) model too, and make it a bit more affordable.
• Bring OpenCL support for your GPUs to the Linux/ARM platform. CUDA works at least but for BOINC (and other applications) to really get things into gear, OpenCL support for the GPU is needed.

nVidia brought out a 16GB Xavier NX module! And for those with a smaller budget who want to upgrade their 2GB or 4GB Jetson Nano developer kit there is also a TX2-NX module. Those of us who have enough money might also want to upgrade their Jetson AGX Xavier with a 64GB module, or wait for the AGX Orin.

[Dirk Broer]

Recents SBC's compared:

If I were working at Sinovoip (the makers of Banana Pi's), I'd start with the Banana Pi M5 -yesterday.
I want the next Raspberry Pi to have eMMC and/or M.2, just like its main contenders for the title 'best consumer SBC'....
Those 32-bit boards all come at around $80-90, so roughly twice the money of a 2GB Raspberry Pi 4. Of the 64-bit boards only the Banana Pi M4 can compete with the Raspberry Pi 4, price-wise, but loses on specs. Most interesting board is the BeagleBone AI: would it be possible to engage all CPU, FPU and GPU cores for BOINC?