Hacker News Reader

Comment by mikepurvis | Hacker News Reader

From the perspective of PC building, I've always thought it would be neat if the CPU/storage/RAM could go on a card with a PCIe edge connector, and then that could be plugged into a "motherboard" that's basically just a PCIe multiplexer out to however many peripheral cards you have.

Maybe it's gimmicky, but I feel like you could get some interesting form factors with the CPU and GPU cards sitting back-to-back or side-by-side, and there would be more flexibility for how to make space for a large air cooler, or take it up again if you've got an AIO.

I know some of this already happens with SFF builds that use a Mini-ITX motherboard + ribbon cable to the GPU, but it's always been a little awkward with Mini-ITX being a 170mm square, and high end GPUs being only 137mm wide but up to 300mm in length.

View on HN

Oh, going back to a backplane computer design? That could be cool, though I assumed we moved away from that model for electrical/signaling reasons? If we could make it work, it would be really cool to have a system that let you put in arbitrary processors, eg. a box with 1 GPU and 2 CPU cards plugged in

I believe PCIe is a leader/follower system, so there'd probably be some issues with that unless the CPUs specifically knew they were sharing, or there was a way for the non-leader units to know they they shouldn't try to control the bus.

But if we're dreaming, we can have the backplane be actually multiple (Nx thunderbird 5 cables connected each slot to all other slots directly).

Then each device can be a host, a client, at the same time and at full bandwidth.

If every device is directly connected to every other one of n devices with Thunderbolt cables, each with its own dedicated set of PCIe lanes, you'd be limited to 1/n of the theoretical maximum bandwidth between any two devices.

What you really want is for every device to be connected through a massive PCIe switch that allows PCIe lanes to be connected arbitrarily, so, e.g., a pair of EPYCs could communicate over 96 lanes with 32 lanes free to connect to peripheral devices.

That's basically what S-100 systems were, isn't it (on a much slower bus)?

Yes, but also in many other scenarios. The last backplane systems I saw were 90s industrial 486s.

There were also PC compatible systems based around ISA backplanes. This was especially common for industrial computers but Zenith/Heathkit made ISA backplane based systems for the business and consumer markets. I own a Zenith Z-160 luggable computer from 1984 which uses an 8 slot 8-bit ISA backplane. 1 slot is occupied by a CPU card which also has the keyboard connector. My system has 2 memory cards which each provide up to 320k along with a serial and parallel port. Zenith sold a desktop version of this as the Z-150. They later released models based upon 16-bit ISA backplanes. I think but am not sure of the top of my head that the last CPU they produced a 16-bit card for was the 486.

This was (is?) done - some strange industrial computers for sure and I think others, where the "motherboard" was just the first board on the backplane.

The transputer b008 series was also somewhat similar.

That would crush latency on RAM.

The RAM and CPU would still be on the same card together, and for the typical case of a single GPU it would just be 16x lanes direct from one to the other.

For cases where there are other cards, yes there would more contention, but few expansion cards are able to saturate more than a lane or two. One lane of PCIe Gen5 is a whopping 4 GB/s in each direction, so that theoretically handles a dual 10gige NIC on its own.

Reply on news.ycombinator.com