25 comments

[ 5.9 ms ] story [ 66.4 ms ] thread
The triangle in this is the worst bit of communication of quantitative information that I've seen lately. No one show it to Tufte.
No joke. How was this better than a bar chart like how every performance metric comparison ever uses.

And that 2 points are “more is better” while one is “less is better” is inexcusable.

I'll like to know why there's a Koala behind the triangle!
It's basically a spider chart, but setting the new version (rather than the old) to 100% makes it especially confusing.
Using a radar chart for just three pairs of data points is pretty inexcusable, too. It's not a great style of chart in general, but at least it makes some sense when there's more like ten or fifteen comparisons involved.
Also, the axes have opposite "good" directions, so the size of the triangle means nothing.
It's particularly confusing because of the mixed scales; two negative characteristics mixed with a positive characteristic. I was reading it initially as “while we regressed on power and area, we got a lot of performance out of that!”.

I think that ultimately it achieved its goal, by being extra suspicious and wrong, and inviting you to look at it for long periods of time. Cute koala clip art too, I can sorta guess where in the office that would have come from, if they were at the office.

I like how one of the metrics is Area causing me to try to compare the area of the two triangles.
Using CoreSight and SVD -- both ARM-specific technologies -- for debug is a bold move. I wonder how ARM is going to feel about that...?
Good for bootstrapping the embedded ecosystem in Rust though, their tooling is based on auto-generating a low-level HAL based on SVD. Can a file format even be protected?
Anyone buying their risc-v IP and socs, other than hobbyists? The company has been around for a few years now.
They've had some design wins. For instance, the new PolarFire SoC FPGA-plus-CPU systems from Microsemi (recently bought by Microchip) use two SiFive designs, a simple microcontroller-like core and a quad-core application processor: https://www.microsemi.com/document-portal/doc_download/12445....
There is also a dev kit for this [1], could use more RAM though.

[1] https://www.crowdsupply.com/microchip/polarfire-soc-icicle-k...

Yeah, I was looking at that. At $500 USD, that's a little outside of what I'd want to spend on something like that. I'm still thinking about it though.
I really wish they just used a (SO)DIMM slot but there we go.

Still tempted to get one, the dual NICs and FPGA fabric could make an interesting network gateway. I’d wonder what throughput you could get if you used the FPGA for routing and maybe even some inspection.

I work for a defense contractor and we’ve been using these in R&D for a little over a year. They seem promising and SiFive has been working to fix every bug we find. I’m encouraged about the future of RISC-V in aerospace.
This is one of the more encouraging things I've heard about this space recently. Obviously no worries if not, but is there any more detail you can provide about the (I'm assuming) niches you're using it in?
Are you looking to use seL4 or Chisel with your RISC-V projects?
We’ve been using chisel, but SiFive’s creation tools mean you don’t have to write anything. Before we went the SiFive route, we were generating CPUs with rocket and getting frustrated when there was no documentation.
Are any companies producing Linux capable SBCs from SiFive SoCs?
SiFive were, if you had a rather large amount of money. You won't see SBCs for RISC-V in general for a while, but recently AllWinner announced a run of 50 million RV64 chips (for smart home / consumer electronics applications) which should mean that there will be a surplus next year of cheap chips and there's hope someone will build an SBC around them.

https://pandaily.com/alibaba-t-head-works-with-chinas-leadin... https://kr-asia.com/china-brief-alibaba-teams-up-with-allwin...

(comment deleted)