Custom Motherboards?

Druegan asks: "I've been rooting around on the net lately checking out all the latest and greatest in new PC parts, plotting out the design for my next build. I'm finding lots of neat stuff, but I can never seem to find a main board that has just the right combination of features. Therefore, I want to Ask Slashdot: Is there any way your esteemed readership knows as to how a person might get a main board custom made?" "I don't know how practical this is, BUT I'm looking for a mainboard that supports a dual processor configuration for the AMD64 FX 55 processor, built around the nVidia nForce 4 chipset. I'd like two full x16 PCI-express slots with support for the nVidia SLI, as well as room for at least 2gb of dual channel DDR, and SATA Raid support. I also am looking to be able to overclock the bejeesus out of the whole mess.

This is only a test case, but there currently is no such mainboard available. I'd like to know if there is some way to get one custom built though, even if it is ridiculously expensive.. (yes, this might fall into the 'more-money-than-brains' dept.)

I'd just like to build the system to see how it'd work."

Comments

Doy

[Apreche]

The reason that you can't get a board that meets your needs is because your needs are unmeetable. IIRC the Athlon 64 FX can't be paired in a dual processor configuration. That's what the Opterons are for.

However, if you come up with requirements for a motherboard that are possible and you want to have one made, good luck. There are many companies that can do it for you for an extremely large pile of money.

Re:Doy

[OAB_X]

It cant be paired, you need the operton 2 series. Its basically the same as an FX, just with it being able to be used in multi-cpu configs.

Re:Doy

[davez0r]

when you said "it's basically the same as the FX", i'm reminded of my dual athlon XP computer from a couple of years back. AMD said that you need athlon MPs for SMP. but the MP and the XP were essentially the same processor. the only differences were that the MP had added testing done, and the XP had some traces cut on its surface to prevent SMP capability. some conductive paint fixed that. is that the case here? or does the opteron actually have additional functionality that the FX physically lacks

Re:Doy

[scheme]

when you said "it's basically the same as the FX", i'm reminded of my dual athlon XP computer from a couple of years back. AMD said that you need athlon MPs for SMP. but the MP and the XP were essentially the same processor. the only differences were that the MP had added testing done, and the XP had some traces cut on its surface to prevent SMP capability. some conductive paint fixed that. is that the case here? or does the opteron actually have additional functionality that the FX physically lacks

Yeah, the 2xx opterons have the capability to setup a single coherent hypertransport link. You need coherent hypertransport links to get processors talking to each other without managling each other's memory.

I don't think that will work

[Jjeff1]

Someone check me on this, but the FX series of chips don't support dual CPU.

And the Nforce 4 isn't a dual CPU chipset. So you'll never find what you're looking for.

Re:I don't think that will work

[OAB_X]

You would need a nForce Pro 4 chipset ofor dual CPU.

Re:I don't think that will work

[MerlynEmrys67]

Actually - the PCI-Express Chipset for AMD (The NForce) IS just a HyperTransport - Device chip... It will work for either a single, a dual, a quad, or for that matter an 8 way motherboard.

You can have two of them in your dual proc system (or 4 of them for that matter) and get your multi 16x.

Don't look for any manufacturer to produce such a thing, and frankly - doing a custom motherboard is probably out of your reach unless your first name is Bill and you don't mind throwing down a few million for a custom job

What's the production run?

[Anonymous Coward]

Unless we're talking about thousands or tens of thousands, it's not economically viable. This is not a system-on-chip design, so it requires considerable debugging. Prototype manufacturing costs are always high as well.

My advice as an electronic engineer: give it up, your idea doesn't make sense.

Re:What's the production run?

[harrkev]

What the parent says is correct, but let me elaborate. Let's assume that you DID have the knowledge and education to undertake this. Here is what you would need...

1) A complete schematic capture system with a really good board layout tool. My company provides one for me, so I do not know how much they paid. But I would guess that for one license, you would have to spend $10K or so. Certainly more than $5K. Note that there are budget packages out there ($2K or so), but you get what you pay for.

2) You will need to get the board manufacturered. Expect at least $2K or so for a handfull of boards (probably closer to $5K). Since you will not get it right on the first try, you will have to get multiple revisions made.

3) Buy parts. Some items have a minimum buy quantity. You might need to get entire tape-n-reel packs.

4) Get the board built. This is not something that you can do yourself, unless you have $100K equipment lying around. Contract this out. My best guess is between $1000 and $5000 for small quantities. I am not sure about this part, though.

5) Debug. You do have test equipment, I hope. Minimum equipment will be a good logic analyzer and oscope. The minimum that I would recommend would be an Agilent frame with a good analyzer card, pattern generator card, and o-scope card (tektronics make awesome scopes, but I do not like their logic analyzers). Expect to drop at least $30K on this. You can rent, though, if money is tight ;) Now, if you want to make your job easier, you might want to get some specialized equipment. Special analyzers exist which analyze one type of interface. You should probably have one for HyperTransport (assuming AMD), one to analyze the DRAM channels, one to analyze PCI, one to analyze USB, etc. You get the idea. This stuff will be used a lot less, so you should rent it.

6) Let's assume that the hardware works. Did you want a BIOS with that? License it. I have absolutely no idea how much this costs. You will likely have to costomize the BIOS for your board. You might be able to do this with open-source tools. Expect to spend some time on this, though.

7) You want that done this year? While you hold down a day job? Expect to spend $10/day on no-doze and Jolt cola.

As you can see, there is a LOT involved. The only reason that you can get a motherboard for $100 or so is that they make a LOT of them. The first motherboard is incredibly expensive. The second one is dirt cheap.

I believe that the Tyan ....

[the eric conspiracy]

has something very close to what you want in their Thunder K8WE. You need Opterons though because the AMD64 doesn't support dual CPU, and the chipset is the NVIDIA nForce(TM) Professional 2200 and 2050 series, which is needed for dual CPU and PCI-X.

Not a good idea

[photon317]

Aside from what others have already pointed out, namely that your chosen CPU and Chipset don't appear to support dual processors to begin with, the idea of having a custom motherboard made is silly.

Motherboards are extremely complex peices of equipment. An enormous amount of work goes into getting them production-quality, it's a lot more than just wiring the rights pins of the right chips and sockets together. There's all the EMF and heat effects to consider, trace lengths and their effect on signal propogation, etc. Then ocne you have a baord that's even capable of functioning reliably, you have to make a BIOS for it and get all the right parameters tweaked correctly to initialize the board the right way - there's a lot of values tuned by the vendor for the board in question that you never see in your little BIOS setup screen.

Even among commercial boards, as we've seen on review sites, there are varying levels of success at building a rock-solid stable board. It requires an enormous amount of engineering man-hours to go through the design and testing process, and sometimes they still can't get it quite right, and half the boards are a little "flaky" under the wrong conditions.

So even if you wanted to drop some enormous sums of money (very enormous, I would imagine, orders of magnitude more than the cost of any custom built PC), it would be unsupported by other vendors (drivers, etc), and likely be plagued by little one-off problems like so many new boards are. Usually the vendor can see the trends in the problems based on numerous end-user bug reports, and fix it in the BIOS - but with just one user, good luck.

Chances are that if you actually made a competent choice about what motherboard features and components really suit your needs, you'd find they already make it anyways.

Re:Not a good idea

[HeyLaughingBoy]

So, I'd love to hear from some experts about just what level of custom board would be doable for a hobbyist?

Depends on your definition of "hobbyist."
I'm an electrical engineer who also does hardware/software design & development both as a hobby and for pay as a side job. You can build pretty sophisticated hardware these days: it's never been easier.

I use the free version Eagle CAD to do PC boards under Linux, and there are plenty of shops that will build prototypes for under US$100. Most of the important parts manufacturers will either give you free samples, or sell small quantities off their website. My compiler of choice is gcc, also free.

My biggest problem is usually coming up with a cool project that's (a) worth doing, (b) doesn't cost an arm and a leg, and (c) can be done by one person who's married with a small child in less than a lifetime (i.e., 6 months or so :-)

As a result, my most interesting projects tend to be from the people who pay me to design something they need, but have no idea how to do.: generally people with plenty of expertise in fields other than electrical and software engineering.

The stuff I come up with on my own tends to not get finished; OTOH, knowing someone's waiting on me to deliver (and is paying) is a huge motivator to get done on time.

HTH :-)

Tyan Thunder K8WE

[DeadBugs]

http://www.tyan.com/products/html/thunderk8we.html

- Dual AMD Opteron(TM) 200 series processors
- DUAL PCIe x16 with FULL SPEED x16 lanes
- (8) DIMMs for Reg'd DDR400 memory
- U320 SCSI and SATA-II with NVRAID(TM)
- Dual Gigabit Ethernet with ActiveArmor(TM)
- FireWire and USB 2.0 ports

I can

[Anonymous Coward]

Give me lots of money and time. I will need to learn all about the processor's power requirements, the connection to all the other chips, their power requirements, their datasheets, until I know everything about every part that's gonna be needed.

Then you need to pay a licence for a schematic-capture to PCB suite with simulation and auto-routing. I think a few 100K$ for the Cadence suite should do it. Oh, you thought that someone who thinks two-layer boards are a challenge and uses Eagle could do it? Nope.

You'll need someone to procure small quantities of the parts you want, that is, get to know all the reps in your area and squeeze them for samples. In between the badgering phone calls, you still have to order all the other parts, while trying to find a local PCB assembler to handle your parts. Don't forget to supply the pick-and-place file and keep in mind the constraints when building your parts library. Oh, you thought you can just buy parts and the software automatically draws them for you with the IBIS model already connected? Nope.

Advise the fab of your PCB ahead of time of the layer count and size of the PCB so they can at least get the materials and open up a time slot for you. When you start the layout, you should be able to estimate a completion time, so it's not a problem, right? You did think of the stack up BEFORE laying out, right? You didn't just use any thickness you felt like, right? You know the difference between pre-preg and core, right?

Anyways, once you've done all the placement, assigned all the properties to the nets (unless you did all that on the schematic and configured the packager to forward the properties to the board), set up the DRC rules and routing areas and keepouts, and defined your via technology, and routed the critical paths, you can unleash the auto-router.

While Specctra is churning away, you can go back to the BOM. Are all your parts going to ship at the same time? Will you receive them at the same time? Will you ship them to your assembler or will you want to look at them first? Anyways, you should be able to tell them when to open up a slot for assembly so they can schedule it.

After the auto-router is done, it's clean up time. Check the artwork carefully. Create the assembly drawings, mechanical drawings, drill files, pick-and-place and IPC files. You *will* want to electrically test the PCB before tossing it to your assembler, right?

So the layout is done, you can start sending gerbers to your fabricator for DFM checking, and you can send the pick and place and solder stencil files to the assembler. While the DFM checking is done, you can extract the layout into Signal Explorer and run some simulations to see if there are any signal integrity issues. What's that? You want to run those tests BEFORE actually getting the board built??

Well, it's already been a year since the project started, and no one can wait anymore....

Dude, what you're asking for is the single most ridiculous thing I've ever heard on /. Even more outlandish than the "can I use a laptop LCD on a desktop?"-type questions.

Re:I can

[John_Booty]

"Dude, what you're asking for is the single most ridiculous thing I've ever heard on /."
-----

I agree completely.

I've been posting on Slashdot for around 7 years and I think this is my first "me too"-style post. Please forgive me. It's just that this question was so. fucking. ridiculous. Literally the stupidest thing I've ever seen posted on Slashdot.

Go to Taiwan

[Matt_Bennett]

Seriously- if you wanted to build your own custom motherboard, do what the big guys do- go to one of the motherboard makers in Taiwan. You'll need them because they already have relationships with all the CPU and chipset companies. You (or the company that makes the motherboard) would very likely have to use something already released, or enter into some *very* restrictive NDAs to get pre-release silicon, and that would be only if you can convince the silicon manufacturers that they should use up some of their very limited pre-release silicon on you and your project.

It will cost millions and take 6 months to a year to get a good, stable system. Making a motherboard is far from the "just plug it in" stage, particularly when you get to new-ish technology (say, stuff released in the last 3 years). Everybody is still learning how to connect things up right and view the secrets of the development and testing as proprietary.

On the other hand, there is a big problem in what you're asking for- as far as I am aware, The chipset you specify doesn't support 2 x16 PCI-E slots, and opening it up to other chipsets, there are none that offer capability for 2 x16 slots (available now)