Many years ago, before the economies of the world went to shit, the enthusiast computing market was strong and each new motherboard was continually out-classing the motherboard du jour before it. If the Asus fuck wagon had 48 lanes of PCIe then the ASRock fuck mobile had to have 54 lanes which didn't matter because a month later the Gigabyte fuck train was going to have 58 followed by the MSI fuck cruiser with an unheard of 64 lanes.

The natural progression was more lanes, faster buses and bigger slots. I expected to see seven PCIe 16x slots on a single motherboard with full-width buses; none of this bullshit where you had to pick between two 16x slots or three slots at 16x 8x 8x or bullshit like that.

I don't know what the fuck happened but there is currently no motherboard on the market that supports 112 PCIe lanes at any level and I find that unacceptable. There is no northbridge that supports that many lanes natively and there is no motherboard that has that many lanes accessible.

This thread is about the solution to that. One of the many projects I never finished thanks to my debilitating and untreated schizophrenia is a motherboard the likes of which the world has never seen. I'm trained as an electronics engineer and have decent experience in the design of PCBs. I've never before tackled many of the problems of motherboard design—such has high frequency LVDS, high-current power planes, proximity/latency mitigation and synchronization in component placement and many other minor things like capacitance in the fibers that constitute the PCB adding resistance to the LVDS paths for RAM causing sync issues and memory corruption—but I took on the challenge and have conquered those design considerations.

I'm designing an AMD socket G34 motherboard with the following basic features:

CPU sockets	4
RAM slots	48
PCIE lanes	168
SATA ports	6+

I'm going off AMD reference designs and will add GbE NICs, SATA/SAS RAID controllers, USB 3.0 controllers and high quality audio. I have companion projects utilizing ePCIe that will provide various features that are useful to gamers, server administrators, hackers and power users. I've designed an optional inline encryption system utilizing jumper configuration matrices that would require knowledge based authentication at the hardware level; this would allow all drives to be encrypted without CPU overhead, be protected by the fifth amendment and be easily disabled in the event of police raid (simply pulling off the jumpers which could be attached to strings or such would render the entire drive cluster unreadable as encrypted by proven uncrackable encryption).

The price point of this motherboard would be outrageous and even conservative estimates based on some of the basic parts such as blank PCBs that need to have the traces routed, sockets/slots for CPUs/RAM and the chipset ICs put a most basic model at well over $1,000 for just the parts at the one-at-a-time prices buying each component for just one motherboard. With all the planned additions to the reference designs making one of these from scratch would cost a few thousand if just one was being made.

The reason I've written all this up is that obviously it makes more sense to make bulk orders of the parts and crank out several of these at a time. The motherboard alone would be the single most expensive part of a computer that has no cheap configuration requiring not less then four CPUs and sticks of RAM. There is no case that would support it and it would necessitate an entirely custom mounting solution; shipping would also be outrageous on this motherboard that would range from ten to twenty pounds and be nearly a meter square. Power usage be damned; this would need multiple PSUs or a four-pronged, three-phase power connecter only used in electric clothes dryers.

All of that being said, there would be no computer that begins to approach the level of power a rig built on one of these would have. It would be the last motherboard you ever buy. With a total of sixteen channels of DDR3 RAM for up to 512 GB (registered, ECC), eight PCIe x16 slots at full width (none of that x16 physical but only x8 electrical bullshit) and as many hard drives as your heart desires this motherboard has an unparallelled upgrade path.

I just wish AMD had a northbridge out for the Opteron platform that supported PCIe 3.

Regardless, I'm hoping to gauge or drum up interest in this pointlessly power motherboard in hopes of convincing myself to start a Kickstarter campaign to get these into production. These would jump start the currently dead enthusiast computing market and I honestly just can't afford to make something worth more then my car. I would need to raise about $1,000 each plus the cost of the equipment to make these which would be about $5,000 so those are the numbers I'd use to configure the campaign. Let me know what you think. If you have specific ideas that you'd like to see just put them here and I'll list them in this post.

I recommend working covertly on it. I could use that as the core of my new host.

_________________

The reason why you should work covertly at stage 1 is for two reasons.

1) Evil things from between the darkness of galaxies may try to kill you to silence the tech.

2) It helps to have some research in progress when you set up a kickstarter fund later.

_________________

I second Yom, already having something to show when you set up the Kickstarter will make it much more appealing!

The development code name for this motherboard was "The God Killer" and I started it in July of 2010. I am a registered AMD developer and have conducted the following research.

• Integrated Circuit sourcing and pricing.
• PCB sourcing and manufacture.
• Manufacturing techniques and cost.
• Component bandwidth and interconnect requirements.
• EMI shielding and induction/capacitive dynamics in PCB design.
• High frequency signaling design.

This list is just for starters. I have the design finished. I know how I plan on making each motherboard, I have five different variants, I have three of the five PCB designs done, I have BOM lists ready for each variant, I have the BIOS nearly finished, I have started working on the video cards I want as a tie-in product to the board, I have three mounting solutions/cases designed and I have a solution to the PSU needing to be unprecedentedly massive.

I plan on making this for myself no matter the public interest and my designs are so far beyond any enterprise or consumer level that I'm not a threat to either market. For any given enterprise solution this motherboard is far outside the acceptable envelope of cost of ownership and cost of deployment. For any consumer this motherboard has an extremely low performance to cost ratio. The demand for this I predict to be near-zero, marginal at best.

There are, however, the few rare exceptional people out there that see the value in a god killer and would love a computer that costs upwards of $30k and replaces all other computers on a network of any size. The god killer can run XCP and with PCI pass-through you could host a Windows VM to play any high-end video game on your eight graphics cards in Crossfire (the god killer does not support SLI and fuck you for not wanting to run a full AMD rig you communist dog) while jamming as much porn onto your 50+ HDDs, all downloaded from your eight-way GbE trunk with TOE.

Unless you're able to connect to the internet at a tier 3 internet hub or better this is more then a single network should need. This is all per motherboard. There's even a variant that's nothing but GbE TOE NICs with enough overhead to allow a network of these motherboards to connect on a nearly latency free backbone. The numbers have all been balanced so bandwidth is unrestricted by any bottlenecks. The HT transport the CPUs use, the RAM speeds and the SAS/SATA interconnects are balanced so that a file that would require the full bandwidth of the RAM for one second could be loaded into ram from disk in one second.

This is a 200 GB file we're talking about. If you split your drives properly in a system made on a god killer and organized your data storage properly there would not even be load times. You could run an entire VM image directly from RAM with a half TB of RAM and load that in two seconds. (OK, two and a half seconds.)

I'm out of the design phase. Don't get it twisted. I'm just acting coy about it.

I think I've gone through 5 different types of nerdgasm reading your description.