System76 Shares With Us More Details On Thelio Open Hardware, Pricing Starts At $1,100 USD

[phuclv]

Anyone know how good support is in POWER9 for day to day stuff? If it were capable of running all the typical open-source desktop software in Ubuntu, for example, I would just use it. I don't really game anymore.

From the benchmarks on Phoronix it can be seen that multimedia support was not very impressive due to the lack of optimized software for the Altivec ISA
- POWER9 Benchmarks vs. Intel Xeon vs. AMD EPYC Performance On Debian Linux
- 13-Way IBM POWER9 Talos II vs. Intel Xeon vs. AMD Linux Benchmarks On Debian
- Test Driving A 64-Thread POWER9 Workstation, Initial Performance Against A 96-Core ARM

Web-browsing experience is far from what being called "usable". Many people are still working on bringing up Chrome and Firefox on PPC. At the moment they can run with JIT disabled so Javascript performance would be very terrible. And due to the above problem, Chrome also struggled playing VP9 on POWER9.

IMHO right now it's only useful if you do extensive scientific programming due to the high performance and large number of threads. For day-to-day stuff an ARM laptop would be even better since you can surf the web, write documents, take notes, watch movies... with much smaller power requirements like on PPC

[freespirit]

this price for a complete system is not that bad, if is it arm is it possible to be completly open as talos system or is still not possible to have a complete open firmware stack?

[GruenSein]

So... Am I the only one only seeing a SATA connection board? It’s nice that it appears to be open and all but it hardly helps anyone to get rid of blobs and closed firmware...

[madscientist159]

From the benchmarks on Phoronix it can be seen that multimedia support was not very impressive due to the lack of optimized software for the Altivec ISA
- POWER9 Benchmarks vs. Intel Xeon vs. AMD EPYC Performance On Debian Linux
- 13-Way IBM POWER9 Talos II vs. Intel Xeon vs. AMD Linux Benchmarks On Debian
- Test Driving A 64-Thread POWER9 Workstation, Initial Performance Against A 96-Core ARM

Web-browsing experience is far from what being called "usable". Many people are still working on bringing up Chrome and Firefox on PPC. At the moment they can run with JIT disabled so Javascript performance would be very terrible. And due to the above problem, Chrome also struggled playing VP9 on POWER9.

IMHO right now it's only useful if you do extensive scientific programming due to the high performance and large number of threads. For day-to-day stuff an ARM laptop would be even better since you can surf the web, write documents, take notes, watch movies... with much smaller power requirements like on PPC

I don't think that's quite fair. Since the benchmarks were published, a lot of fixes went out to optimize the software. LAME for instance has seen a massive speed boost. Web browsing is also quite fast and pleasant with the Chromium browser, though Firefox probably did leave a bad taste in peoples mouth due to its lack of JIT.

Regarding VP9, there is VSX support in tree. Chromium uses it on e.g. Debian, where the system libraries are used instead of the hacked / out of date internal Chromium version. Also there are issues with the open amdgpu drivers simply not supporting video acceleration on older cards vs. the proprietary driver stack. If you have access to an OpenPOWER box, preferably with a newer GPU like Vega, give it a try -- you might like what you see.

In general, if you have no real processing requirements, get an ARM Chromebook for $150 and install coreboot on it. That's a lot cheaper than the System76 machine, and likely gets you a much more open firmware stack if you choose a model with the right Chinese SoC. If you want to actually create open software, or use a 4k screen with a fast disk, etc. POWER is a quiet, low-power choice especially with the 4-core CPUs.

[madscientist159]

this price for a complete system is not that bad, if is it arm is it possible to be completly open as talos system or is still not possible to have a complete open firmware stack?

It depends on exactly what ARM SoC they choose. However, given the claims of high performance, either they are likely using x86 (in which case no freeing will ever be possible) or a closed high-end ARM device like TX2 (same problem). The price point (to me) strongly hints toward an Intel desktop processor, though an AMD low end processor would be possible as well.

If for some reason they are calling a Chinese cell phone / tablet SoC "high performance", that would be freeable. I highly doubt that is the case here, especially considering the RAM support table they have published.

EDIT: OK, there's a picture of part of the mainboard at https://betanews.com/2018/10/26/syst...-thelio-linux/ Here's my educated guess putting all the pieces together:

Mini-ITX Intel board, 2 DDR4 RAM slots, Core series LGA1151 processor, one PCIe slot, based on similar reference design to the Gigabyte mini-ITX board here:
https://www.gigabyte.com/us/Motherbo...WIFI-rev-10#ov

It's possible they have a daugherboard ("riser") that provides additional PCIe slots as well.

Probably the claim to fame bits are in the implementation of the design in KiCAD (hopefully anyway, though I find this a bit dubious given KiCAD's current limitations vs. Allegro or similar) and the "open hardware" backplane bits. This would be a fully closed and locked firmware stack machine if I'm right, which kinda renders the "open hardware" aspects worthless from a practical perspective.

[freespirit]

It depends on exactly what ARM SoC they choose. However, given the claims of high performance, either they are likely using x86 (in which case no freeing will ever be possible) or a closed high-end ARM device like TX2 (same problem). The price point (to me) strongly hints toward an Intel desktop processor, though an AMD low end processor would be possible as well.

If for some reason they are calling a Chinese cell phone / tablet SoC "high performance", that would be freeable. I highly doubt that is the case here, especially considering the RAM support table they have published.

we'll see what they are going for, but if they write open hardware and they sell a x86 no words at all
about prices, i still do not understand what make it so different from x86, i've read the blackbird motherboard could cost around 700$ looking the specs is comparable to a low end x86 mobo that could cost under 100$ even the heatsink cost 3x more than a noctua, that is itself not cheaper, why this kind of hardware is so much expensive? i think what should be done is to make the price comparable to x86, or at least to be double, let's say 200$ for the blackbird mobo

[madscientist159]

we'll see what they are going for, but if they write open hardware and they sell a x86 no words at all
about prices, i still do not understand what make it so different from x86, i've read the blackbird motherboard could cost around 700$ looking the specs is comparable to a low end x86 mobo that could cost under 100$ even the heatsink cost 3x more than a noctua, that is itself not cheaper, why this kind of hardware is so much expensive? i think what should be done is to make the price comparable to x86, or at least to be double, let's say 200$ for the blackbird mobo

Some of the additional expense is because volumes are so much lower than x86 commodity hardware, and the x86 commodity ecosystem is partly funded though sale of personal data, closed-source advertising and malware (think Lenovo), etc. In contrast, we do all of our development in the open, so we have to fund it all up front and hope products take off in the market, with no other revenue streams possible. This is largely the opposite of the x86 closed development model.

Another source of cost is simply quality -- we don't want to put out a mainboard that will fail in a year because the regulators were massively under-speced for instance. We do QA on our hardware and firmware to make sure it is stable and reliable, and have had to bring POWER out of the server space into the desktop realm -- this means getting things like graphics drivers fully operational and comparable in quality to those same drivers on x86. Finally, POWER has a lower integration level -- basically, there's more chips on the board vs. the Intel SoC-in-a-socket model. This is part of why we can even offer open systems, but it does add cost as each device placed, soldered, and tested on a PCB adds some amount of cost to the final product.

EDIT: One last item...pricing isn't out yet for Blackbird due to a few factors (including predicting what will happen with the trade war and figuring out ways to avoid cost spikes including from knock on effects like sudden surges in demand from non-PRC suppliers, I guess this could be called "supply chain engineering"). Let me just unofficially state that if the final price is $700 I will not be happy at all.

[WolfpackN64]

I'm a bit biased but for the most part things just work. We use P9 desktops as daily drivers on this side, and the last missing piece (Chromium) is falling into place nicely: https://mobile.twitter.com/RaptorEng...01835230408711

The only hassle is if you have proprietary applications with no source code. But that's what the Windows box in the corner or the set top box by the TV is for, right?

Does Firefox work on POWER9 systems?

[madscientist159]

Does Firefox work on POWER9 systems?

It works properly starting with version 64 AFAIK, but it's dog slow because of the lack of JIT. Mozilla doesn't really care about owner controlled platforms, they are more interested in x86/Windows and ARM/Android at this point, which is both concerning and disappointing. I don't know if community requests can change that attitude or not, sadly -- Firefox was a great browser back in the day but lately it seems to have gone places that are the opposite of what it used to be.

[freespirit]

Some of the additional expense is because volumes are so much lower than x86 commodity hardware, and the x86 commodity ecosystem is partly funded though sale of personal data, closed-source advertising and malware (think Lenovo), etc. In contrast, we do all of our development in the open, so we have to fund it all up front and hope products take off in the market, with no other revenue streams possible. This is largely the opposite of the x86 closed development model.

Another source of cost is simply quality -- we don't want to put out a mainboard that will fail in a year because the regulators were massively under-speced for instance. We do QA on our hardware and firmware to make sure it is stable and reliable, and have had to bring POWER out of the server space into the desktop realm -- this means getting things like graphics drivers fully operational and comparable in quality to those same drivers on x86. Finally, POWER has a lower integration level -- basically, there's more chips on the board vs. the Intel SoC-in-a-socket model. This is part of why we can even offer open systems, but it does add cost as each device placed, soldered, and tested on a PCB adds some amount of cost to the final product.

EDIT: One last item...pricing isn't out yet for Blackbird due to a few factors (including predicting what will happen with the trade war and figuring out ways to avoid cost spikes including from knock on effects like sudden surges in demand from non-PRC suppliers, I guess this could be called "supply chain engineering"). Let me just unofficially state that if the final price is $700 I will not be happy at all.

thank you for your reply, as always is very appreciated