-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
On 04/04/2016 11:00 AM, Paul Boddie wrote:
On Monday 4. April 2016 17.06.23 Timothy Pearson wrote:
General Overview of Alternate Architectures
=== ARM ===
While the ARM architecture may be more wildly known for locked-down
"widely"
Yes, that was a typo.
What about all the single-board computers available plus initiatives to make open hardware [*] laptops and netbooks? Maybe those initiatives need our support in preference to various products from the usual corporate players who have readily introduced those control/surveillance technologies into their other products.
Some quick links:
http://rhombus-tech.net/community_ideas/laptop_15in/
https://olimex.wordpress.com/2015/11/24/a64-olinuxino-oshw-linux-laptop-idea... becomes-more-real/
[*] Remind me again what the "correct" term is for fully-documented and freely-modifiable/distributable hardware is, as it clearly isn't that memorable.
ARM is very nice for low end / remote work and embedded systems, there's no question about that. However, at least in my experience, it is woefully inadequate for non-trivial content creation tasks (editing video, engineering design work, etc.).
I've noticed a lot of IBM promotional activity around POWER8 of late: there was a Fellowship blog post that suddenly appeared promoting the POWER architecture, I've seen targeted adverts featuring POWER, and now there's this message. Indeed, the blog post used very similar language to "POWER is the only architecture currently competitive with Intel in terms of raw performance", which may or may not be true, but I can't help feeling that a bunch of people have been asked to let us all know.
That statement was based on actual in-house benchmarks: https://raptorengineeringinc.com/TALOS/power_advantages.php
We're not paid by IBM to promote their processors by any means. The simple fact is Intel is the dominant processor manufacturer, and they lead hands down in price, performance, and power consumption. They also refuse to give machine owners full access to their hardware, putting people in a rather sticky situation.
If you know of another architecture that can compete with Haswell in terms of raw performance please let us know about it!
Unlike various other architectures, POWER risks sitting in the same position as it did at the start of its life: at the top end and largely unavailable to most of us. It possibly needs another big vendor to give it a boost - just as Apple managed to do - but that doesn't necessarily change the availability situation for open hardware.
Yes, it's a risk. However, everything seems to be risky right now. x86 cannot be trusted, POWER itself has somewhat poor power efficiency, and no other architecture we are aware of comes even close to the performance of Intel and AMD's processors from a decade ago. You're going to have a hard time convincing me that the future is holding on to obsolete hardware from 10 years ago, purchasing overpriced systems that perform worse than said hardware, or licensing the right to run limited FOSS software from a proprietary vendor. From our perspective, based solely on internal evaluation and comparison of various machines available right now, POWER offers the best chance to provide reasonably modern libre systems over the next 5 - 10 years.
There's also Ingenic, Atheros (now part of Qualcomm) and an assortment of other MIPS vendors. Performance might not be great amongst some of their products - multi-gigahertz, multi-core monsters are probably not the focus here for router manufacturers - but these CPUs are still interesting. My impression is that MIPS (Imagination) are targeting IoT applications alongside trying to get their proprietary graphics technologies into various products (unfortunately).
More quick links:
http://rhombus-tech.net/ingenic/jz4775/
https://www.kickstarter.com/projects/imgtec/creator-ci40-the-ultimate-iot-in... a-box-dev-kit
Whether MIPS-based products are recommended or not, in contrast to other architectures, perhaps depends on whether you can actually buy the products concerned. If POWER is now available from a range of sources like ARM and MIPS devices are, then maybe it is more interesting. Otherwise, it mostly isn't, apart from in some kind of paper exercise about what is "best".
Yeah, MIPS seems mostly targeted at embedded systems, and honestly this is an area where RISCV could shine while it tries to ramp up to ARM-level performance. Regarding POWER, bear in mind the architecture was only opened up for licensing / third party manufacture a couple of years ago, and China is already manufacturing custom CPUs using POWER8. If people are willing to take a risk on POWER it has the chance to compete directly with Intel. If not, the sad reality is it will be many, many years (possibly never) before FOSS systems can reasonably be used to design new computing products. RISCV is just too far down the line to be a viable option yet; nothing has been taped out and we're still only talking single core CPUs in the pipeline for eventual production.
=== RISCV ===
While this architecture is extremely limited in performance, price, and performance per watt compared to x86, ARM, or POWER, it is also one of the only fully open source CPU architectures available outside of an FPGA. and may eventually be competitive with MIPS in terms of raw performance. Currently there are no RISCV SoCs in production, however projects such as lowRISC aim to change that: http://www.lowrisc.org/
This is definitely one to watch, given the backing from various companies that probably don't care about Intel's agenda. I'd like to hear some more of the detail behind that performance assessment. Having played around a bit with MIPS recently, there are some fairly obvious areas where something like RISC-V could easily achieve greater performance than MIPS. Indeed, I think that this is why some people treated OpenRISC - based on the first version of the MIPS architecture - with some disdain.
(And thinking about OpenRISC reminds me about LM32 and various other architectures that have probably only remained within the FPGA space - at least with regard to general availability - because of lack of funding.)
It's difficult to evaluate performance on a product that isn't even taped out yet, however all indications are a single core variant is going to be the first production model. Immediately that puts it far behind the 8 Loongson MIPS cores on the Loonson 3B.
- -- Timothy Pearson Raptor Engineering +1 (415) 727-8645 (direct line) +1 (512) 690-0200 (switchboard) http://www.raptorengineeringinc.com