Introducing the Milk-V Oasis with SG2380: A Revolutionary RISC-V Desktop Experience

Account terminated? Video unavailable.

1 Like

you may need to try again, the video is live for me, maybe it could be re-upped to another hosting service?

Yep, the milk-v YouTube account returned after 1 or 2 days.

1 Like


so far I think the best you can do is order “pre-purchase” coupons. and it seems like you can only order 2 coupons. I read somewhere that they are limited quantity. I think we will be lucky if it ships NOV/DEC…I got my wait and see googles on. hopefully we will get meles boards between now and when it ships.

1 Like

Somehow I am just now noticing that the P600 series of cores from SiFive support sv57 vm layout.

From the P600 series Data Sheet:

virtual memory support, with up to 57-bit addressing, with precise exceptions

What is the SG2380 using? I assume it is probably sv48 but would like to know because of my curiosity.

1 Like

Does SG2380 support PCI-E bifurcation?

I need 10Gbe Ethernet, but don’t want to waste whole 8x slot for that. I would be great to plug 2xM2 card into 8x slot, use one slot for NVMe, and 10G SFP+ card in second slot. And use M2 slot on the board for SATA adapter. 10GBe NAS use-case will require alot bifurcation and SATA splitting to support flexible configuration…

Guess what? You don’t need to scarify the PCIe to have 10Gb network, SG2380 supports 25G network natively.


The latest SG280 spec. It should be freeze if there is no further change…



Doubled NPU TOPs asss compared to back then! Now I’m really hyped!

We responded to your inquiry about the Milk-V Oasis on April 10th.Please check your inbox and spam folder as emails can sometimes be redirected.

If you still can’t find our response, feel free to reach out directly at

According to the spec there will be 4 x1 high speed interfaces that can be configured to work as a PCIe gen 3 or 2x 25GBe + 2x 5GBe. How that would physically look like on the mobo, like a 4 PCIe x1 slots? Could you use only one of those slots as an ethernet and other ones as a PCIe?

Lookup “PCIe expander” or “PCIe bifurcation adapter” , bitcoin miners used to use them to plug multiple GPU’s into a single PCIe slot on a motherboard. If you lookup the pinout of a PCIe slot on wikipedia you will see at least in theory how a single x16 slot could be expanded into 2 x8 slots or 4 x4 or 8 x2 slots or 16 x1 slots or 1 x8 + 2 x4 or …

It all depends on what is delivered in the motherboard revision that finally ships. On paper there are a lot of ways to expand a 1 x16 gen 4 PCIe slot. e. g.
1 x16 gen 4
2 x8 gen 4
4 x4 gen 4
8 x2 gen 4
16 x1 gen4

2 x16 gen 3
4 x8 gen 3
8 x4 gen 3
16 x2 gen 3
32 x1 gen 3

(only some configurations are available from the PCIe controller in the SoC, but others are possible with additional external chips)

Just remember that the info in the image above is for the SG2380 SoC and not the motherboard. A lot of PCIe lanes will be pre-allocated on the motherbord to provide USB, NVMe, SATA,…

Me personally I’m waiting until the final tape out of the SG2380 SoC is confirmed before asking questions about the final board configuration. And I expect the ITX motherboard will go through a few revisions before it ships.

P. S. A standard Mini-ITX form factor motherboard only provides a single PCIe slot (see diagram on Wikipedia) is the reason that I mentioned PCIe bifurcation expanders. We have no idea what the Milk-V Oasis board will provide in terms of PCIe slots (yet).

Does the, now I am guessing frozen, IP used in the SG2380 supports JESD318 Version 1.02 (Nov 2023) standard (CAMM2: Compression Attached Memory Module Common Standard).
I have not read the standard, because I do not have access, but I have read that ECC is supported, both in Single & Dual Channel, with JESD318 Version 1.02.

So I’m guessing that “if” LPCAMM2 modules are available by Q3 and “if” the IP used in the SG2380 supports the JESD318 Version 1.02 (Nov 2023) standard and “if” LPCAAM2 memory is available with ECC, then there is at least a chance that ECC Memory “might” be possible on the Milk-V Oasis ?

LPCAMM2 doesn’t require new IP in the SoC, just the standard LPDDR5/5x controller. So far we are not sure if we can go with LPCAMM2 since Micron is just shipping and the price is high and not widely available.

SG2380 doesn’t support DDR5 so we can not use DIMM or SODIMM, so there is not much choice for us…


Regarding the memory option, what about at least including the LPCAMM2 connector on the motherboard but can leave it empty. Continue with the plan to release the SoC with the different memory configurations. Then for the base model, offer a 2GB SG2380 SoC as the introductory price in lieu of the extra expense to include the connector. Shift the cost of the newer memory to users, let each person decide if they have money to buy the more expensive LPCAMM2 memory currently. For the rest of us (poorer) people, wait for Toshiba, Samsung and others to produce competing LPCAMM2 modules in the future and hopefully bring the prices down to upgrade the board down the line.

Including new upgradeable memory configuration would be innovative and future-proof the board. That would make a decision to purchase the product irresistible if the board not only offers expandability via the already announced PCIe, M.2, USB, etc, but also being able to update individual processor AND memory components as well. Add to that, if there is a readily available mainstream and consumer-friendly OS image like Rocky Linux mentioned above available for easy install on an OpenSBI/U-boot enabled board at launch would likely make it very popular.

Uh… sadly, I have to pull that tooth.
Right now, there is no such thing as ACPI for RISC-V. There is a Spec that leans towards what ARM would call SystemReady, but right now, you would need a device tree first and foremost to even think about booting a Linux kernel.
Further, RISC-V support is, at best, “experimental”. Alpine doesn’t have it aside from their edge builds and Debian treats it like that funny little thing that exists.

Here is the Debian “thing” I run on my VisionFive2 for comparison:

root@riscboi /etc# cat os-release
PRETTY_NAME="Debian GNU/Linux trixie/sid"
NAME="Debian GNU/Linux"
root@riscboi /etc# cat apt/sources.list
# Suggested via wiki:
deb sid main contrib non-free non-free-firmware
#deb unreleased main
deb-src sid main

I have been trying to get in touch with at least Alpine to sort something out there but haven’t had a lot of time lately to follow up on that.

The next problem is the kernel itself and the required drivers. They have to be freshly built - and as the SG is just about comming out, kernel support is going to live in a downstream fork for a while untill it is merged - if it is merged. o.o

1 Like

Uh… sadly, I have to pull that tooth.

Not sure what about DTBs detracts from “mainstream and consumer-friendly OS”, it is possible to have all three.

Alpine doesn’t have [RISC-V] aside from their edge builds

Alpine 3.20.0 released | Alpine Linux :wink:

EDIT: That being said though, I do agree that it is definitely the early days for RISC-V, so don’t expect a smooth ride out of the gate by any means.

1 Like

Exactly. What ever the situation with Alpine or other distros and the VisionFive board, that is different hardware with another manufacturer. This thread here and now is about Milk-V and the new Oasis looking ahead to its future possibilities. Which is there seems to be active interest for this platform by the Rocky Linux team on here with actual funds and intention to procure hardware, and it seems like Milk-V is willing to respond having exchanged buddy info and all. Hopefully that portends to some collaboration with fruitful results from those first steps, such as a working device tree out of the gate (why else would Rocky Linux buy this hardware and not bother to go the extra mile to get Linux working well on it)… Mayhaps even, once Milk-V has the first engineering samples physically on hand they ship a few ahead of time to Rocky Linux members (hint, hint) who seem more than willing to lend a hand on porting and bring up of the distro.

It is understood that RISC-V is a new hardware platform with its ecosystem being somewhat on shaky footing right now. However being able to simply download an image from a known Open source group and write it to an SD card to bring up a Linux environment immediately, is much more consumer-friendly than having to figure out on your own that a board that you just bought brand new requires somehow updating firmware BEFORE EVEN booting to an OS followed by having to search various pages to finally find a download link to some personal Google drive that contains a highly customized Linux system.

In any case, given the knowledge that RISC-V is a fairly new entrant on the scene there would be no problem if any initial image is labeled as “experimental” being out of band from mainline at the onset, as long as the system is “usable” and all patches, drivers, device trees and what not are on public repos with the corresponding source code being submitted to the relevant parties to be merged. Which if this work was being completed by Rocky Linux there would be great trust that would be done since they are currently doing it and have the expertise and infrastructure in place already as part of their workflow.

This is not to say or prevent Milk-V from putting together their own distro if they have such plans, but just having an OS image also available directly from a major team like Rocky Linux and maintained by them on said physical hardware, backed by Milk-V with the goal to eventually have an “official” release alongside x64 and the Pi under their downloads page would lend some major credence to the Milk-V Oasis as an OPEN RISC-V platform. That would draw a lot of attention and support from the community to the board and Milk-V. At the very least any work done could be leveraged by all other distros such as Alpine to compile their own ports if they so desire.

1 Like

So guys, it’s now pretty much June, we are about 9 months in from the SG2380 promise
What’s the time to tapeout and boards?