Motherboard SFF Threadripper motherboard petition thread
- Thread starter jeshikat
- Start date
you know......this kinda explains why the AM4 screws are so far apart. its kinda like they want the same coolers to work for both.
haha, I'm not that interesting. A search for my username is pretty telling what my interests are, haha.I'm stalking you on reddit now
Looks like Mini-ITX on EPYC may be pretty doable since it's a SOC: http://b2b.gigabyte.com/Server-Motherboard/MZ31-AR0-rev-10#ov
That chipset heatsink is for the 10G NIC so it could be eliminated.
Edit: The board is pretty bare, so really there just needs to be room for the RAM (SODIMM?), VRMs, and internal connectors:
That chipset heatsink is for the 10G NIC so it could be eliminated.
Edit: The board is pretty bare, so really there just needs to be room for the RAM (SODIMM?), VRMs, and internal connectors:
Last edited:
So there's no PCH or Southbridge to speak of? that's amazing o_o
Seriously looking forward to the mATX version of this. If that never gets developed I wonder how many m.2 or PCIe signal-carrying sockets people would be able to squeeze into an ITX board (or if it's possible to riser out a full x16 connection).
(the sketch has two CMOS batteries, LOL. the GbE and USB ICs on the M.2 riser board could be moved to the battery socket)
Seriously looking forward to the mATX version of this. If that never gets developed I wonder how many m.2 or PCIe signal-carrying sockets people would be able to squeeze into an ITX board (or if it's possible to riser out a full x16 connection).
(the sketch has two CMOS batteries, LOL. the GbE and USB ICs on the M.2 riser board could be moved to the battery socket)
Last edited:
I just realise full DDR4 sockets on motherboards tend to be soldered through the PCB. Are there versions of the (vertical) sockets that just sit on the surface? (although I imagine the socket would be easily broken when the ram is forcibly tilted sideways when mounted).
The X299 ITX could get away with M.2 sockets under the SODIMM slots because the contact pins are not soldered through, so I'm thinking...
The X299 ITX could get away with M.2 sockets under the SODIMM slots because the contact pins are not soldered through, so I'm thinking...
Why not SODIMMs so the power delivery can be moved onto the board underneath the SODIMM slots? Or, have all 4 on one side, and the socket shifted slightly to the left or right, giving some more space for more VRMs and such. Could have the VRM's high and low side split between the top and the bottom of the board, giving more space on top for maybe a second vertical M.2 slot (though that'd present some issues when it comes to heatsink clearance.Looks like Mini-ITX on EPYC may be pretty doable since it's a SOC: http://b2b.gigabyte.com/Server-Motherboard/MZ31-AR0-rev-10#ov
That chipset heatsink is for the 10G NIC so it could be eliminated.
Edit: The board is pretty bare, so really there just needs to be room for the RAM (SODIMM?), VRMs, and internal connectors:
SODIMMs are really expensive and have a high CAS latency. I personally feel that it's best to avoid them if possible.
I think at this point, we're half-jokingly testing out various SFF designs with the massive TR4 socket to say, "Hey, look, it could happen!"
We haven't even bothered to illustrate heavy use of daughter boards like what ASRock pulled off with the impressive x299 ITX board. Hell, even the VRM risercards that ASUS is famous for would be pretty slick.
Due to the TR4 socket allowing quad slots of ram out both the left and the right side, in the video I opted to support 4 slots on the right side to support quad-channel ram. The benefit to placing all 4 slots on one side also means that you could potentially align 2x m.2 slots on the left side of the socket on the backside of the motherboard.
Dreamy.
I think at this point, we're half-jokingly testing out various SFF designs with the massive TR4 socket to say, "Hey, look, it could happen!"
We haven't even bothered to illustrate heavy use of daughter boards like what ASRock pulled off with the impressive x299 ITX board. Hell, even the VRM risercards that ASUS is famous for would be pretty slick.
Due to the TR4 socket allowing quad slots of ram out both the left and the right side, in the video I opted to support 4 slots on the right side to support quad-channel ram. The benefit to placing all 4 slots on one side also means that you could potentially align 2x m.2 slots on the left side of the socket on the backside of the motherboard.
Dreamy.
Yup, everything is on the EPYC CPU itself, though I suppose a Super IO chip may still be needed for fans, temp monitoring, etc.
I think Mini-DTX would make way more sense than that.
Uh, it's a server board, so it's a redundant backup battery! Yup
Yes, SODIMM would be better since it would free up space on the PCB for other goodies. But I wanted to see if everything would fit even with DIMMs since it's a more conservative route.
Having all the RAM on one side is more difficult though, especially with higher channels, since the traces all need to be equal length.
Wait, a VROC header?...tsk tsk, do not encourage AMD to have that shit
There are three ways to handle soldering of pins onto a PCB: through-hole, through-hole with back-drilling, or blind vias. Through-hole is the typical mounting method, where you see the soldered pins sticking out the back of the board; it is also the cheapest. Those pins can be back-drilled, reducing signal reflections, improving signal integrity, removing the pins sticking out (so things like M.2 can be mounted over them), but adding substantial cost. Blind vias are holes that only go part way through the board, giving you the same sort of benefits that back-drilled pins do, but shift the extra manufacturing costs to earlier stage in the production process. Asus' Auto Extreme technology uses blind vias.
This. It requires fewer PCB layers to be used to get all of the all of the traces laid out. PCB layers = cost and mITX boards already require more to manage all of the traces going everywhere (EVGA advertises that the Stinger boards have 10 layers, whereas the Classified boards only need 6, despite the addition I/O).
That, plus USB controllers, as I'm not sure Epyc (or Threadripper) has it integrated or not. NICs are evidently not included, either. But I did see this:
Note the: "Links are Infinity Fabric, PCIe, and SATA capable." This is super interesting from a Vega standpoint...I'm curious if Vega correspondingly has a data link that is both PCIe and Infinity Fabric capable, then when connected to one of these processors, can it switch to the low latency Infinity Fabric (mind you memory speeds affect the operating speed of the Infinity Fabric to maintain coherency)?
I wonder if it would be viable to fund X399 ITX development for AsRock via kickstarter? We are probably talking an obscene amount of money. But it would be interesting, and then we could also have some input on layout.
I don't know. From our stand point no problem what so ever, but I don't know if Asrock is in a position as a company to be flexible enough to use crowdfunding.
They absolutely didn't see a market for it when I asked. I imagine it'd take an upfront order of 3000-5000 boards to get their attention.
That dislike bar though :|
Yo, that would be so awesome!
Well, what are we waiting for? Let's do it!
EDIT: Broke formatting
Last edited:
Yea, I intentionally excluded the use of music and transitions timed to a beat since everyone's tastes are different. I even kept it at 30 seconds to ensure that it wasn't too long and boring.
Can't please everyone though ¯\_(ツ)_/¯
Perhaps they are just bummed that it isn't real.
Can't please everyone though ¯\_(ツ)_/¯
Perhaps they are just bummed that it isn't real.
As much fun as it is to dream about, does a crowdfunding campaign for 3000+ boards actually seem realistic to anyone? Because we're probably talking $1100+ for the CPU and mobo alone.
The board would be very expensive for an AMD board (granted, of course), but I suspect that only the high-end 16-core processor binned for high clock speeds will command around $1000-1100. The lower end Threadripper processors with not many more cores than an R7 and slower clock speeds than an 1800X may very well be around or less than $500. At that point you're gaining only memory bandwidth and PCI-e connectivity, but at slower clocks and higher TDP; the value wouldn't be as great.
Similar threads
