I love their spirit, effort, and products very much.
I’m just thinking, if they can release a laptop with something like OpenPower or OpenSparc, it would be a true unique opensource device.
And if with RAID+ECC support, rugged design… A monster is born.
Purism is eyeing RISC-V for future devices, but they want it to mature more first.
In any case, a computer is more than a CPU. It needs the necessary chipset for interface to peripherals. Laptop customers are going to expect SATA or NVMe, some kind of graphics, USB, if you’re lucky ethernet, maybe PCIe, other sundry peripherals such as HD audio, maybe SD card reader, monitoring buses (e.g. SMBus or I2C) - and drivers for all of that.
Customers are going to expect multi-core implementations and at least reasonably comparable performance (in the laptop market). You get that kind of performance through sophisticated CPU implementations that are completely or largely invisible in the ISA itself - multi-issue (superscalar), pipelining, out-of-order execution, branch prediction, horrible buggy things like speculative execution
Laptop customers are probably also interested in bang per Watt, which again requires maturity of implementation that goes beyond mere ISA.
General purpose computers probably need good crypto (particularly your average Purism customer) and I don’t think RISC V has any standardised extension for that.
Customers may expect greater modularity than a typical embedded SoC gives you.
I think if you built a RISC V laptop today, it would be quite low end. I definitely wouldn’t be a potential customer. My current x86 laptop is a bit long in the tooth but I’m sure it would easily outperform a RISC V laptop today.
You need economies of scale to get an acceptable price, or acceptable to all but the most zealous.
Support in the Linux kernel and in the Debian family of distros needs to be stable enough for customer use (as opposed to use by developers).
i’d buy a RISC-V laptop at (almost) any cost but yes, it will be a while probably
The main issue is the GPU, there is no open hardware gpu out there at moment, risc-v is still too young to be seriously used, there is not enough horse power to even have an i3 like, and the software is just missing.
For have an open laptop the best option will be ibm power 10 that is not out yet, probably will be on 2020 or 2021 because power 9 is still too much power hungry while power 10 should be good enough, there is still the gpu problem, i don’t know if something like mali gpu or vivante gpu could be “extracted” from the whole chipset to be used as mobile gpu with the open firmware, i hope someone could dig on this, otherwise there is the asmedia ast2500, the ast2600 should be out in the near future, but it’s just a 2d framebuffer so no 3d accelleration and even web browsing should be processed by cpu, consuming a lot of energy and that’s not good from a mobile standpoint.
There is also another interesting project for a real open 3d gpu https://libre-riscv.org/3d_gpu/ the man behind this project wroted in this forum, because purism gave to him a sponsorship, but the first project will be not enough for consumers.
We need a miracle i think, or someone to trow a lot of money on this project to speed up
I’ve just bought a new laptop. Judging by my buying history, I’ll buy the next one in like 12-15 years. I hope it will be a RISC-V one.
Interestingly though Nvidia plans to use RISC V as the processor on their GeForce graphics cards (source Wikipedia). I don’t know whether that plan has come to fruition or been abandoned or what. That of course wouldn’t make it “open” because the Nvidia card would be running a blob anyway.
There isn’t a free lunch for power consumption for graphics processing. Whether there is no graphics chip / card, and it is done in the CPU, or it is done elsewhere, it still consumes some power doing it. For sure a graphics chip / card may be better optimised to do it. It wouldn’t surprise me if a dedicated graphics card increased total power consumption when used for light graphics processing, because those who bother to get a dedicated graphics card, will usually get something fairly grunty.
For the topic as a whole, I would guess: Watch this space, but it won’t be any time soon.
it depends. modern dedicated GPUs have better power consumption (at idle) than CPUs … but yeah compared to iGPUs they draw more … but the thing is you are not limited in intensive graphics accelerated workloads (like high-end-gaming [will it run Doom-Eternal ? ], GPU compute, raytracing, CAD/CAE modeling etc).
for simple coding, media-consumption, light-gaming, trolling the Purism forums, paying bills, hacking Mr. Robot - the iGPU is enough. the AMD APU has a little more power but it’s not as “freed” as the intel one that Purism uses in their Librem 13/15 laptops.
The gpu we need is not a gaming one, but a gpu able to run at least fullhd with DE, video player and browser accelleration, i mean every desktop uses case.
Use the cpu for these task are really power and task consuming, it’s not ideal even on desktop where lifetime battery is not a thing. Just run a VM and try to play video inside it with VLC or youtube video, and watch your cpu usage on host, it will mostly spkike 100% usage on all core
there is the powerpc notebook project, using a Quorl NXP processor, unfortunately it is PowerISA v2.08B and going EOL.
we (libresoc) are talking with a couple of clients, one of whom recommended using ECC DDR4 and with Linus’ recent rant about Intel and Rowhammer it’s pretty clear why.
as a 450-600 pin SoC which will be somewhere around the USD $7m development NREs in 28nm we are kinda reluctant to throw in the kitchen sink however USB3 OTG HDMI PCIe Host and Client plus eMMC RGBTTL GbE and standard QSPI I2C UART etc all the things you see in mid range SoCs, these are all achievable and if you know the performance of equivalent Rockchip RK3399 you will know these eat even mid range Intel budget processors for breakfast.
it would result in a laptop that would need SATA and NVMe done via PCIe (then yes you could have RAID), would not need an external GPU, could get away with 1920x1080p easily, 4k would be a bit more expensive but also doable. multiple screens might be a bit much. 16 GB of DDR4 ECC RAM would be achievable.
how does that sound?
except the one we’re developing
actually this was just to keep within the power budget of a potential customer at the time (2+ years ago now) who wanted no more than 2W absolute max, entire product power consumption.
our current design is, quad core 1.5 ghz, somewhere around the 120 GFLOPs FP32 mark. total overshoot by a factor of… errr… 20? now, whether that gives anything like a sane power budget for a mobile class CPU remains to be seen.
there’s a reason why we are focussing on video playback as part of the LibreSOC core, it’s to be able to do these kinds of mundane non-exciting tasks.
by extending OpenPOWER with a Cray Vector style augmentation we can save power by reducing executable size as well as number actually executed.
the sigarch “SIMD considered harmful” article is very informative, here. it takes several re-reads to spot the significant part, which is that a Cray style Vector loop can take only 100 instructions to process thousands of pieces of data but SIMD requires 10x more to be executed, and actual binary size is 5 to 20x larger.