SoylentNews is people
SoylentNews
Early to embed and early to rise? Western Digital drops veil on SweRVy RISC-V based designs
Western Digital today finally flashed the results of its vow to move a billion controller cores to RISC-V designs. WD said last year it needed an open and extensible CPU architecture for its purpose-built drive controllers and other devices. As we explained then, no one knew for sure what processors WD has used for its disk and SSD controllers, though they was likely Arm-compatible chips – such as Arm9 and Cortex-M3 parts. It is known that the firm uses Intel CPUs with its ActiveScale archive systems and Tegile all-flash and hybrid arrays.
Last year, the disk and solid-state drive manufacturer vowed that RISC-V was its future, and today it announced the SweRV core, a networked cache coherency scheme, and a SweRV instruction set simulator.
[...] The SweRV core has a two-way superscalar design and is a 32-bit, nine-stage pipeline core, meaning several instructions can be loaded at once and execute simultaneously to save time. It is also an in-order core, whose relative single core performance (a simulated 4.9 CoreMark/Mhz) is expected to exceed that of many out-of-order cores, such as the Arm Cortex A15 (actual 4.72CoreMark/Mhz). Clock speeds go up to 1.8Ghz and it will be built on a
28mm[28nm] CMOS process technology.WD said it hopes open-sourcing the core will drive development of data-centric applications such as Internet of Things (IoT), secure processing, industrial controls and more. We understand WD's ambitions for using RISC-V CPUs go beyond disk and flash drive controllers.
Previously: Western Digital to Transition Consumption of Over One Billion Cores Per Year to RISC-V
Related: WD Announces Client NVMe SSDs with In-House Controllers
(Score: 1, Insightful) by Anonymous Coward on Wednesday December 05 2018, @06:12PM (7 children)
Did Musk get an electric go-cart into the hands of campus-weary students or mobile hobbyists? No. He built a goddamn, mother-fucking Hot Rod for the wealthy players.
We need a FOSH GPU as a RISC-V extension—something with some serious computing ability; something which is expensive.
Otherwise, ain't nobody gonna be interested.
(Score: 4, Insightful) by bob_super on Wednesday December 05 2018, @06:24PM (6 children)
To get a processor to be popular, it needs code.
Remember Alpha, Itanium, and so many others ? Lottsa power, but very expensive and therefore only doing a few specific things.
If you're gonna go incompatible and ask for a lot of money, the people willing to shell out that money need instant amazing benefits, or you don't get funding for OS and program development. That's getting harder and harder to do against the x64 juggernaut.
ARM started small, and took over the computing world, supported by companies looking to save power, and happy to get developing for cheap.
(Score: 0) by Anonymous Coward on Wednesday December 05 2018, @06:38PM (3 children)
It's in the name: General-purpose computing is, well, a large and difficult arena.
That's why you concentrate on a niche market first. You computer folks keep trying to go for the low-end niche market (e.g., IoT crap), which the market keeps telling you isn't real; I'm suggesting you instead go for a high-end niche market, like GPUs.
(Score: 2) by bob_super on Wednesday December 05 2018, @07:01PM (2 children)
Might be because the established players are throwing billions at the high-end, hiring very smart people and patenting all the math and Si tricks those can think of, so kicking their ass is far from cheap and trivial.
It's a bit like my startup: When you provide equipment above a certain grade, customers expect (demand) a lot of legacy features to be implemented and be somewhat familiar. That takes a lot of resources, even will all the open code available these days.
(Score: 0) by Anonymous Coward on Wednesday December 05 2018, @07:30PM (1 child)
Kicking their ass on the high end isn't even near what's required - you just have to be competetive with the run-of-the-mill stuff, while ATTRACTING the high-end money people. Consumers are fickle.
(Score: 2) by bob_super on Wednesday December 05 2018, @08:04PM
If you're new and just "competitive", you need to be cheaper. If you're better, you can command the high-end margins.
You need those margins to get anywhere, because even being "competitive" on the "high-end" against dominant established players requires a lot of resources, and gaining market share is even harder.
(Score: 0) by Anonymous Coward on Thursday December 06 2018, @03:51AM
Progress is being made. Thousands of packages in Debian run on riscV:
https://people.debian.org/~mafm/posts/2018/20180402_debian-gnulinux-port-for-risc-v-64-bit-riscv64-in-debian-infrastructure-debian-ports/ [debian.org]
(Score: 0) by Anonymous Coward on Thursday December 06 2018, @04:21PM
There were even relatively cheap (1.5-2x the cost of x86) alpha motherboards/processors in the late 1990s.
What killed Alpha was the Compaq->HP merger and the decision to collaborate with Intel on a PA-RISC successor rather than getting Alpha onto better processes. The Alpha cores at the time better than Intel, but HP/Compaq's later treatment of them priced them out of mindshare and market, before killing them to focus on Itanium, even when it became obvious itanium was a boondoggle that would have better resulted in backpedalling to either PA-RISC or Alpha (which had core designs which exceeded the performance of Itanium if they had been allowed to go into mass production.)
If you have any doubt about this, go look into the Chinese supercomputers from a few years back that used Alpha derived architecture for the floating point processing. I forget if it was the same or a different system, but they also used sparc cores as the i/o backbone for filling the FLOP cores on one of their supercomputers as well.
Consolidation and Intel's behavior damaged the tech industry in ways a lot of people still don't realize. The current spate of Spectre vulnerabilities is a prime example, since the standardization of little endian and speculative execution for processors has left it much easier to craft exploit code that works similarly everywhere, rather than a diverse assortment of hardware with low lever differences that defy devious optimizations.