Hacker News new | past | comments | ask | show | jobs | submit login

Very interesting and useful.

However, does anyone have experience with obtaining RISC-V server hardware on a commercially useful scale, i.e. more than 1pcs? Back when ARM servers where all the rage, there were lots of announcements, but almost never products one could order. Is RISC-V any better in that regard?




I don't think RISC-V servers are anywhere near that level of development.


Competitive RISC-V general purpose servers are probably about 10 years off. Special applications will happen first.

I'd note that ARM servers are finally becoming competitive in specific use-cases. The AWS Graviton2 CPU is useful for many apps.


From what I got, the future should look like : -You choose the RISC-V design first. -From that design, you choose the fab that will produce the design. Each fab has its own price and trustfulness.


The 3 big fabs that you'd like to use are fully booked until 2023 or so. So you kind of forgot the "wait 4 years" part.


Maybe, but that would mean building our own hardware. I don't think anyone here would be comfortable with that, we are not big enough for such an endavour to make any sense at all. We'd rather buy off-the-shelf hardware, preferably from a selection of vendors...


Sure, we will fab your design how many units do you need? We can manufacture 2000000 for you. Oh, you don't need that many? Is 250000 low enough?


I do not think there is anything to obtain yet. The only two designs that I am aware of and that are out-of-order are Berkley's BOOM and Alibaba's Xuantie 910. And then you also need a complete and standardized platform, i.e. chipsets, firmware interfaces, etc. There are also a few standard extensions, including "Packed-SIMD", "Vector" and "Transactional Memory" that are still in a draft state.


> Berkeley


Buying handful units at list price isn’t at business scale


Yes. I just meant that when we tried to obtain a few ARM servers, the only quantity we ever got was "1 sample machine". Thus we couldn't even really try it out ourselves, let alone have customers do meaningful tests on a handful of machines.

Of course in "a few datacenters full of"-like quantities there was just no chance of getting anything. And of course, for making our own hardware (you can get ARM chips, just not full systems) we don't have the expertise, size and will.


Sure, but you need a handful of units to evaluate them properly. Availability of units is a huge concern and (imo) the largest barrier to boader adoption of non-x86 servers; ie, no sane person would hedge their operations on a single server vendor that can't reliably supply compatible units when needed. At least, not willingly.


Your product is worthless if nobody is using it.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: