RPC DRAM support in open source DRAM controller | Hacker News Reader

RPC DRAM support in open source DRAM controller

antmicro.com

31 pointspabs311 comments

Sidenote: LiteX is awesome. Knowing next to nothing about it, I was able to create a customized system-on-chip with RiscV CPU, several SPI/UART/I2C buses, load it on a devboard and blink some LEDs in an evening or two. The resulting Socs have been shown to boot Linux (albeit, slowly).

That's an unfortunately heavily overloaded name, the FPGA builder is way down the search results behind the football club and the swimwear vendor.

That's impressive, and could lower the cost of entry for a lot of hardware projects.

I don't see it as lowering the cost of entry much (i.e. FPGA chip with reasonable number of gates for Linux is on the order of 50EUR/pc, and general purpose CPU performance is not even remotely close to, let's say Allwinner or Rockchip ARM SoC), but as something that either opens up new possibilities as a whole (for example, a trusted computing platform, where you can actually verify that no backdoors exist- see https://www.bunniestudios.com/blog/?p=5971 ) or something that provides unprecedented flexibility in the future proofness of embedded designs: for example when you realize that you need another UART in a product that's been shipping for 5 years, and the MCU family you have been using just does not have the part you need.

I am particularly excited about the hardware flexibility- it seems a very good platform for scientific/engineering instrumentation (LiteX actually seems to be conceived because of this niche).

> I am particularly excited about the hardware flexibility- it seems a very good platform for scientific/engineering instrumentation (LiteX actually seems to be conceived because of this niche).

That's doubly exciting! Yes it's great to see new possibilities opening up. The cost reduction I mean is the lowering the development costs for custom platforms. If you need a team of expert hardware designers to design this stuff then most labs/small companies will be "priced out" of that arena.

Oh yes, it can bring the cost of design and development down, agreed.

I love RPC; it’s much better utilization of pins - same bandwidth over 22 as over ~ 45, IOW about twice the bandwidth per pin. Downside is a slightly higher latency because address and command have to travel over them too, but this is amortized over the burst transfer.

The real problem with RPC is that devices has tiny capacity; 256 Mb today and max spec’ed to 8 Gb (yes, I am aware that they consider it a feature, but I don’t). Also, it’s actually pretty expensive today compared to regular DRAM.

Guessing address pins have been replaced with SPI or something similar?

That would have been terrible. No, they have effectively routed the address/mask information over the data bus and eliminated those signals. Yes it takes multiple beats, but it's at least not bit serial.

Size comparison:

https://twitter.com/GregDavill/status/1300562224474279936?s=...

I want to love the litex components but I have only ever had problems with them. Never gotten liteeth or litedram to work.

Reply on news.ycombinator.com