No Raspberry Pi 4 in 2019.
-
Someone on the raspberry pi forums asked Jamesh what "cool" meant to him. His response is below :D
"Cool is not an attitude - It's a state of mind. Imagine an eagle flying through the Rocky Mountains, suddenly, it spies its prey, and falls, falls, falls, it's wings guiding it, but to what?
Think of the primitive thoughts cascading through the eagles mind as it plummets towards the rocks below. That is the Pi4."
-
What if we have a little guessing game on what the Raspberry Pi 4 will be and then we can see when it launches who was closest?
It's easy to come up with ideas for improvements, but it's much harder to determine what's realistic given the obvious cost constraints. Also, the Raspberry Pi Foundation has stated multiple times that they will only consider a feature if it is of high importance to a very large part of its user base.
Here's my optimistic but hopefully not entirely unrealistic guess:
SoC:
- 12/14/16 nm class SoC
- Cortex-A73 (4 cores, ~1.6 GHz) -> 100-150 % performance improvement compared to Pi 3 B+
- VideoCore V -> More than 100 % performance improvement compared to Pi 3 B+
RAM:
- 2GB LPDDR3
Interfaces:
- HDMI 2.X (4K, 60 Hz)
- 2xUSB3
- 2xUSB2
- Each USB port is a host port (i.e. no integrated hub)
- Gigabit Ethernet (capable of 900 Mbps+) on its own interface (i.e. not attached to USB)
- SD card: UHS-I 208 MHz (up to 104 MB/s)
- Micro-USB power
Price:
- 35 USD
Form factor:
- Same physical size. Similar or identical connector placement. Might even be compatible with most old cases.
Power consumption:
- Approximately 500 mW per core. Similar to Pi 3B and ~30 % lower than Pi 3 B+.
-
@Brunnis
Here is my prediction I made in January. At least we agree on RAM, GPU, USB and HDMI. I forgot the Gigabit Ethernet. About "Each USB port is a host port (i.e. no integrated hub)", why in the first place did they use an integrated hub? For cost saving, does it make any much difference for "most" users or is it just a very specific case? The cpu of my choice is much more powerful and probably isn't realistic, but I also took a price tag of $ 50 in perspective. The Raspberry Pi foundation already said they will stay at $ 35 price point, so this is very unlikely. I wish there would be a powerful pro version of the regular Pi and I am sure that I am not the only person.Warning: The below link to Cortex-A75 Wikipedia article is pointing to wrong A73. Here the right link: https://en.wikipedia.org/wiki/ARM_Cortex-A75
@thelostsoul said in Raspberry Pi4 Speculation:
I wouldn't mind if the starting price would be higher than the previous entries. The previous cpu was "Broadcom BCM2837B0 VideoCore IV" and "Cortex-A53 (ARMv8) 64-bit SoC @ 1.4GHz". Wikipedia page about VideoCore: https://en.wikipedia.org/wiki/VideoCore
Spec:
- CPU: Cortex-A75 (ARMv8-A) 64-bit SoC @ 2.8GHz
- SOC: BCM7251
- GPU: VideoCore V (500Mhz? compared to 400Mhz on Pi3)
- 16GB fast internal space
- DDR3 2GB RAM (compared to DDR2 1GB)
Connections:
- 4K/60Hz HDMI output (or dual 1080p/60Hz output)
- 2 x USB 2.0 Type A (our current for legacy)
- 2 x USB 3.2 Type C (the smaller new type)
- 1 x SATA revision 3.0 (6 Gbit/s) (probably not)
- onboard power and reset switch (probably not)
- dedicated RCA Composite Video output (probably not)
50 $/€ initial price tag (instead 35 of the previous). Ok, this is more like a wishlist. I even don't know if these
cpu and gpu and soc would work together and if this is realistic in that price point. But hey, I expect groundbreaking changes without compatible with older version. -
@thelostsoul
Nice! Regarding the CPU, Cortex-A75 is probably hard to justify due to (probably) higher licensing cost, larger die area and higher power consumption. A73 is just more economic, especially since we can't expect the Pi 4 to use a more modern process than 16/14/12 nm.BTW, the reason I suggest rather low frequencies is that I expect them to optimize the Pi 4 for sustained performance during passive cooling. A core frequency exceeding 2 GHz will almost certainly result in quick down-clocking after the core reaches high temperatures. The side effect of low default frequencies would likely be a pretty high overclocking potential, given extra cooling.
Regarding USB host ports: Since they're building an SoC from scratch and die area for a USB host port probably isn't very large, I'm guessing it makes sense to forgo using a hub. The big plus with not using a hub is that each port has its own bandwidth and doesn't have to share it with the others.
-
@Brunnis I agree on the cpu, the A73 with lower clocks sounds much more realistic. About the USB host ports, if there are USB 3 and 2, then at least those two need its own host, right?
-
@thelostsoul said in No Raspberry Pi 4 in 2019.:
About the USB host ports, if there are USB 3 and 2, then at least those two need its own host, right?
EDITED: Actually, it seems it's possible to have a combined USB2/USB3 hub hanging off a USB3 host port. So, a single USB3 host port combined with a hub could be a real alternative. The design will probably be dictated by cost.
-
Probably shooting my .... guesses early here, but I don't suppose we think the Pi folks will be able to put some kind of AV1 decoder chip / acceleration on the v4 in 2020?
Be nice to be able to do that, I know it's coming along rapidly. (AV1, not the Pi)
-
You can get an Intel NUC generation 6 for not much more than a pi + case + power/sd/etc and they will run up to PS1 level pretty easily.
-
OUT NOW!
3 versions:
1GB ram, 2GB ram, 4GB ram -
@Brunnis said in No Raspberry Pi 4 in 2019.:
Here's my optimistic but hopefully not entirely unrealistic guess:
SoC:
- 12/14/16 nm class SoC
- Cortex-A73 (4 cores, ~1.6 GHz) -> 100-150 % performance improvement compared to Pi 3 B+
- VideoCore V -> More than 100 % performance improvement compared to Pi 3 B+
RAM:
- 2GB LPDDR3
Interfaces:
- HDMI 2.X (4K, 60 Hz)
- 2xUSB3
- 2xUSB2
- Each USB port is a host port (i.e. no integrated hub)
- Gigabit Ethernet (capable of 900 Mbps+) on its own interface (i.e. not attached to USB)
- SD card: UHS-I 208 MHz (up to 104 MB/s)
- Micro-USB power
Price:
- 35 USD
Form factor:
- Same physical size. Similar or identical connector placement. Might even be compatible with most old cases.
Power consumption:
- Approximately 500 mW per core. Similar to Pi 3B and ~30 % lower than Pi 3 B+.
So, it turns out I wasn't completely off with my guesswork! The CPU performance is in line with what I expected, but they chose to go with the A72 instead of the A73. The A72 and A73 perform very similarly, but the A73 is slightly more efficient.
We don't know much about the GPU right now, but I kind of doubt it's 2x the old one, like I had hoped.
Overall though, this looks like a truly great upgrade. The biggest "issue" I see with it is that they weren't able to go with a 16/14/12 nm class manufacturing process for the new chip. The first tests indicate even higher power consumption than the Pi 3B+. A more recent process node would have helped tremendously with that, but I'm guessing it would haven increased the price of the board.
-
@Brunnis What do you think of the different RAM options? Which one will you be getting? (if you are getting one...)
-
I'll be getting the 2GB version. I just don't see the point in getting the 4GB version unless you're going to run desktop type workloads or know, specifically, that your workload is memory intensive. I have a hard time seeing how the emulators that you can run on this board will benefit from going with the 4GB model. But I could be wrong, of course. :D
Oh, and I just found this:
"The VideoCore 6 is about 4 times faster than the Video Core 4."
It's on page 22 in the PDF linked here: https://www.raspberrypi.org/forums/viewtopic.php?f=63&t=243372#p1483956
Sounds very promising, but we'll see when more benchmarks start showing up.
-
@Brunnis said in No Raspberry Pi 4 in 2019.:
I'll be getting the 2GB version. I just don't see the point in getting the 4GB version unless you're going to run desktop type workloads or know, specifically, that your workload is memory intensive.
More RAM is also useful for speeding up repeated hdd/ssd/sd access, since Linux uses unused RAM for buffering the file system. You can see this via the command
free -h
("h" means "human readable" for more human-friendly numbers). Example from my Laptop:total used free shared buff/cache available Mem: 7,7G 2,2G 2,5G 405M 2,9G 4,8G
Although the "free" memory seems to be only 2.5 GB, temporary buffers take 2.9 GB. The real amount of free RAM is 4.8 GB, since the system clears the buffered data as soon as the memory is needed elsewhere. (Don't ask me why the numbers don't seem to add up perfectly, I don't know the actual math behind it.)
Contributions to the project are always appreciated, so if you would like to support us with a donation you can do so here.
Hosting provided by Mythic-Beasts. See the Hosting Information page for more information.