Comment by cesarb

Comment by cesarb a day ago

> Well, patience as a consumer might pay off in the next year or so when the music stops and hyperscalers are forced to dump their inventories.

Their inventories are not what consumers use.

Consumer DDR5 motherboards normally take UDIMMs. Server DDR5 motherboards normally take RDIMMs. They're mechanically incompatible, and the voltages are different. And the memory for GPUs is normally soldered directly to the board (and of the GDDRn family, instead of the DDRn or LPDDRn families used by most CPUs).

As for GPUs, they're also different. Most consumer GPUs are PCIe x16 cards with DP and HDMI ports; most hyperscaler GPUs are going to have more exotic form factors like OAM, and not have any DP or HDMI ports (since they have no need for graphics output).

So no, unfortunately hyperscalers dumping their inventories would be of little use to consumers. We'll have to wait for the factories to switch their production to consumer-targeted products.

Edit: even their NVMe drives are going to have different form factors like E1.S and different connectors like U.2, making them hard for normal consumers to use.

nine_k a day ago

I bet that friendly Chinese entrepreneurs will sell inexpensive E1.S to m.2 adapters, and maybe even PCIe riser cards for putting an OAM and a bunch of fans, and maybe even an HDMI output. Good hardware won't be wasted, given some demand.

Reply View 3 replies

bilegeek 20 hours ago

Inexpensive, probably not. E1.S isn't just a different connector, it's a completely different protocol than PCIe.

Reply View | 2 replies
- snuxoll 20 hours ago
  
  > it's a completely different protocol than PCIe.
  Wrong. It is still just NVMe over PCIe like every other modern SSD form factor.
  
  Reply View | 1 reply
  
  bilegeek 18 hours ago
  
  You're right, I was confusing E.1S and CXL.
  
  Reply View | 0 replies

mycall 14 hours ago

> Consumer DDR5 motherboards normally take UDIMMs. Server DDR5 motherboards normally take RDIMMs. They're mechanically incompatible, and the voltages are different.

All you need is a fixed-latency, dumb translator bridge where the adapter forces everything into a simplified JEDEC-compliant mode.

CA/CK Line Translator with a Fixed Retimer as the biggest mismatch between RDIMM/UDIMM is the command/address path.

RDIMMs route CA/CK to RCD to DRAM, and the UDIMMs route CA/CK to DRAM directly, take the UDIMM CA/CK, delay + buffer + level shift it, feed it into a "RCD" like input using a delay locked loops (DLL).

Throw in a SPD translator, PMIC and voltage correction, DQ line conditioning and some other stuff into a 10–12-layer PCB, retimer chips, vrm, and level shifters.

It would cost about $40 million to fab and about $100 per adapter but would make bank with all the spare UDIMMs when the bubble bursts.

Reply View 0 replies

arjie a day ago

I imagine the cost is primarily in the actual DRAM chips on the DIMM. So availability of RDIMMs on the market will affect DRAM prices anyway. These days lots of motherboards come with Oculink, etc. and you can get a U.2 PCIe card for rather cheap.

I put together a small server with mostly commodity parts.

Reply View 0 replies