Your Vote is Needed: Should Milk-V Oasis Come with LPCAMM2 or LPDDR5?

Your vote is important to us.

FYI

SG2380 Product Brief

How to vote

1 Like

Choose LPCAMM2

Advantages: Users can replace the appropriate size memory sticks (LPCAMM2) according to the usage scenario, and can obtain the Milk-V Oasis at a lower price. There is no need to pay for memory. (Oasis supports 2x LPCAMM2)
Disadvantages: As of July 3, 2024, the global LPCAMM2 production capacity is low, not yet fully popularized, with few purchasing channels and high prices. (Crucial 64GB LPCAMM2 LPDDR5X-7500 memory: $329; Crucial 32GB LPCAMM2 LPDDR5X-7500 memory: $179)

Choose LPDDR5

Advantages: Compared to LPCAMM2, LPDDR5 is relatively cheaper.
Disadvantages: Memory cannot be upgraded. Users need to pay for memory at the time of purchase.

I do partially agree that LPDDR5 is a sensible choice for this product.

But personally I am interested in the Oasis to play with some experimental hardware. And LPCAMM2 definitely falls in the category of experimental hardware. I would personally really enjoy messing with LPCAMM2.

When the Oasis is shipped, the prices of LPCAMM2 have most likely decreased at least somewhat already.

The LPCAMM2 can also be seen as a future-proof investment. More boards will likely support it in the future and then you’ll already have some.

So my vote goes to shipping with no memory. Though the decision would be easier if we could get an estimate of how expensive the LPDDR5 memory would be.

LOL, if only it was that simple. It is not really the cost of the new memory itself that is at issue here. The sticking point about the price, now that the situation is understood, is that there are no existing LPCAMM2 memory modules in smaller sizes like 2GB, 4GB to be able to bundle with the motherboard even remotely near the already advertised $120 discount price. The only way at the moment to reach that low price point is to not include any memory. Anyone who has built their own PC is familiar with this situation since that is how it has always been when buying an AMD or Intel motherboard.

No matter what, everyone is going to have to pay for some ram in one way or another in order to having a booting device. It really boils down to whether people want the ease of having soldered-on memory with the board or NOT (and it sounds like the feasibility to have one onboard memory and one LPCAMM2 connector is off the table).

Therefore the key bullet points are:

  1. Go with LPCAMM2, the board will be offered with no memory installed so users must source it themselves (however the price of the motherboard may even be cheaper. Quote @hoka: “can obtain the Milk-V Oasis at a lower price”)

  2. Go with soldered memory, the board can be sold with LPDDR5 included and still start at the lower price tier. It is limited to specific memory sizes to select but is more convenient (though one is stuck with the decision - and possibly not have the option to max out the 96GB capability of the SoC)

To Milk-V the cost of the new memory is not that much a concern because they will pass on the cost in all cases, but more of time and effort.

On the other hand buyers ARE sensitive to the price tag and Milk-V understand, rightly so, that low cost is a key factor to attract users right now and achieve a higher demand than other dev boards.

The plan was likely to offer configurations starting with the usual 4GB, 8GB, 16GB options for that reason. To bundle LPCAMM2 now would mean the price has to start out at $300 for a base model with memory. That would be over the $150 threshold for many people to consider purchasing something that is essentially a hobbyist pursuit at the moment, even if you are getting what you paid for. So it would be low volume sales and not worth the trouble to set up the supply chain currently just for the convenience of a few users, especially this late in the game to be able to get a product out by the end of the year. Let’s be realistic, that $120 headline for such a full-featured mini-itx (the PCIe slot and M.2 and SATA ports alone are worth it for their usefulness in development work) is what got thousands of interest. If Milk-V can actually come through on producing and selling the board at this price, the Oasis is going to be in great demand for huge quantities beyond other RISC-V products sold so far (hope they can continue keeping up with production after launch). Oasis would be the first board with a chance to launch RISC-V to the masses because of its affordability and the provided hardware which would help overcome problems that current RISC-V SBC have.

Seems that opinion is pretty split between the two options.
It might be worth considering splitting the product line to accommodate both, but of course I don’t know the finances behind that.

Now, I personally a fan of LPCAMM2. Although, sure, it probably won’t be breaking any speed records compared to current x86 workstations, the SG2380 (on paper) has more than enough compute to function as a lower end workstation or server, roles which, in my opinion, greatly benefit from flexible RAM options, especially given the large up front cost of having to commit to both the Oasis and the required memory for their usecase, rather than allowing them to commit to smaller capacities for testing. I’m also certain that more desktop oriented users would at least appreciate the modularity and would take the platform more seriously if LPCAMM2 was used.

It is also important to remember that, although prices for LPCAMM2 modules aren’t great right now, it likely won’t remain that way for long. Whether or not the price drop will come before the Oasis ships, I don’t know.

now dimm can already support 6400mhz.

lpcamm2 is so much more expensive than dimm, but it’s only 1100mhz higher.

is it really worth it?

While I do usually prefer to be able to upgrade ram, I think for this particular board, LPDDR5 is makes more sense and I’d be happy to pay for the more ram at time of purchase.
I put my vote on the LPDDR5.
Keen for more updates on the Oasis!

The max is now 128GB, with 256 bit memory bus width.

I Choose LPCAMM2

MY VOTE GOES WITH THE LPCAMM2 WITH NO MEMORY INSTALLED.
However the price of the motherboard may even be cheaper
Quote @hoka: “can obtain the Milk-V Oasis at a lower price”

IF WE GET ENOUGHT LPCAMM2 MEMORY WE CAN GET A WHOLESALE PRICES ON THE MEMORY AT A BETTER PRICE.

I’m sorry if this is a stupid question, but is it possible to have both? Like route 8 of the 16 channels to soldered on LPDDR5, and route the other 8 to a LPCAMM2 socket? That way cost conscious buyers could just leave the LPCAMM2 socket unpopulated, but one could buy a 32 GB version of the board and at a later point in time slot another 32 GB LPCAMM2 module in to upgrade to 64 GB total?

I think LPCAMM2 is the better choice, as it will be the first cheaper RISC-V device with removable memory and the base board is going to be cheaper as well. This is perfect for a “Desktop experience”.

For me the volume of RAM is more important than its speed or price, so any solution that would provide 128 GB is good. If you would ship soldered 128GB LPDDR5 then it would be OK, but if the maximum soldered amount is 64GB then I’ll vouch for LPCAMM2.
(but for now I’ve voted for LPCAMM2)

The poll on Twitter/X has now ended:
116 votes ; Final results - 45.7% LPDDR5 (53 votes) ; 54.3% LPCAMM2 (63 votes)
When added to the above (ignoring any double votes - where people voted on both)
LPCAMM2: 28 votes
LPDDR5: 26 votes
Would give a grand total (so far) of:
LPCAMM2: 91 votes
LPDDR5: 79 votes

Just use the cheaper one for considering of poor people

LPDDR5 soldered on is fine for me - it’s what was originally suggested and the larger factor is that the ecosystem is still very much a-change, but we won’t be able to switch CPUs to a newer generation. In that situation, I think it’s wiser to buy (right at the start) the configuration that the system should be, and leave it like that. So if you expect to have need for 32GB, just get it like that, and don’t plan on upgrading it 2 years later when the need arises.
I think it would not make economic sense, and the same it would not be super important to be able to take out the RAM and use it elsewhere, since the prices will come down over the years anyway, so you could just get RAM for whatever other system we’re talking about, and keep the OASIS going.

Make sense? Wrong?

Another same board with soldered RAM? Anyone who prefers that can just buy the JUPITER now, right?
Makes no sense to have all boards identical. How about evolving the hardware here and add something new to RISCV like being first with LPCAMM2. Next step is to switch CPUs too.

There sure are lots of assumptions about boards with all these higher RAM to buy then. Officially it does not seem anything in writing making firm commitments. The only thing for certain is the coupon selling for the cheapest board. Everything else have been comments or assuming it true. Most SBC that can actually be ordered now only go to 16 GB like Jupiter. Could be that would be the highest choice to buy for a while at the start or at all. Those planning larger sizes are taking a chance if voting soldered RAM.

I voted for LPDDR5. I will definitely not buy this board if I have to invest a few hundred USD just for the memory. Don’t get me wrong, the Oasis will be quite a powerful RISC-V machine, but it is still in no way comparable to modern workstations. Do not get too hyped-up. The software ecosystem will also be limited, at least in the beginning. It is something to play with, to develop for. It is not something I will use for any serious business. Therefore, if the overall costs are too high, it isn’t worth it for me, sorry.

Well, then I would expect ECC, at least four LPCAMM2 slots, remote management and so on. All the things (or: half of the things) I have in my current Xeon workstation and in my home micro server. For productive use for, e.g., Internet-facing servers, I would wait a bit longer until a lot of teething problems have been ironed out. But this is just my opinion.