[GPU Daughter Board] mATX-V Form Factor

Mini-ITX boards are now quite powerful and featured. Accomplishing this has rendered many cases somewhat obsolete that were designed with MicroATX (9.6in X 9.6in) motherboards as their target size. The Silverstone ML04, for example:



1/2 height video cards also never became a thing for this current generation. We have 1/2 length now, but that only shortens a potential case in one direction.

What if the space gained from these smaller motherboards could be put to better use?


NOTE: Edited because I can't read >_<

Thoughts? Opinions?

Amazing idea. But I'd go one step further and make it a dual-GPU board with a 16 to 8/8 PCIe lane.
Please send this to AMD, they are the most interested in SFF gaming right now

Using existing mATX standoffs gives this a much better chance of being adopted, as the 'standard' is already in place. Great idea!

One wrinkly with video output handoff (from dGPU through to iGPU) is that this effectively rules out use for VR, due to the additional latency of handing completed frames off over the PCIe bus, and other timing weirdness that entails.
Luckily, because the PCIe backplate will be in a known location relative to the board, it would be easy to add an extra riser for direct outputs, through a rigid PCB to 'normal' rear slots, or through a flex riser to a repositioned slot.

I don't think AMD or the card manufacturers will be interested in such idea. It's something that will only work with itx motherboards. If they really wanted to make a radical change in form factor they would need to change the factory equipment.

If you have such idea for new graphics card form factor you'd need to submit it to PCI-E standard owner rather than gpu manufacturers.

If I had a chance to change the graphics form factor I would go for LGA socket instead of expansion card slot since new AMD units have HBM memory inside the package. This way you would pay less for new graphics and people would change them more often which is good for a chip manufacturer.

Phuncz said:

Amazing idea. But I'd go one step further and make it a dual-GPU board with a 16 to 8/8 PCIe lane.
Please send this to AMD, they are the most interested in SFF gaming right now

Click to expand...

One can hope.

EdZ said:

Using existing mATX standoffs gives this a much better chance of being adopted, as the 'standard' is already in place. Great idea!

One wrinkly with video output handoff (from dGPU through to iGPU) is that this effectively rules out use for VR, due to the additional latency of handing completed frames off over the PCIe bus, and other timing weirdness that entails.
Luckily, because the PCIe backplate will be in a known location relative to the board, it would be easy to add an extra riser for direct outputs, through a rigid PCB to 'normal' rear slots, or through a flex riser to a repositioned slot.

Click to expand...

Right, thats an excellent suggestion. Perhpas the output header could simply loop back to the motherboard itself? And not have to involve the PCIe bus because that would add engineering complication?

SaperPL said:

I don't think AMD or the card manufacturers will be interested in such idea. It's something that will only work with itx motherboards. If they really wanted to make a radical change in form factor they would need to change the factory equipment.

If you have such idea for new graphics card form factor you'd need to submit it to PCI-E standard owner rather than gpu manufacturers.

If I had a chance to change the graphics form factor I would go for LGA socket instead of expansion card slot since new AMD units have HBM memory inside the package. This way you would pay less for new graphics and people would change them more often which is good for a chip manufacturer.

Click to expand...

I completely understand your position, but the big 2 want their video card blood money. They should just make motherboards that can do this, but nVidia in particular is content with their existing business model. Sell big add-in card, make money. To go down the rabbit hole of hoping for them to suddenly get into the motherboard game would be the real pipe dream.

You said GPU manufacturers wouldn't be a target for this? why not. ASUS/Gigabyte both made the Mini-ITX PCB length version of the 970. That's clearly not a reference board. They COULD do something like this if they were ballsy enough, without the need to consult anyone really.

They also like ways to market their stuff:

Isn't mITX 170*170, not 175*175? Doesn't matter, onto the idea itself.

I love it! And I am on your side with this idea being best suited for the actual card manufacturers. They break so many parts of the standard already, I don't see that being the reason for them to not at least try this.

However, there are a few problems this solution entails.
The first is the shape of the PCB. Making PCBs in this V-Shape would have implications for manufacturing (more useless cutoff), shipping (larger box with a lot of wasted space) and stability (PCB will easily flex and bend. Not problematic when installed, but during installation, care is required.)

All these issues can be fixed, though. You could divide the card into two sections that use a high pin count board interconnect to alleviate them quite nicely. Backplates would work quite well, too.

The idea of having a standardised output connector is nice, but I don't see it flying. In the first wave of this idea, you would more likely use the space available to solder four vertical HDMI and display port connectors (one for each slot) while leaving the rest of the area for venting.

Something that is absolutely awesome about this is the additional size you gain for coolers and fans. I don't know what the best solution would be, but it seems that you could have an 80mm fan at the front that blows air through the whole bottom of the card.
A lot of possibilities there.

Watecooling could be a breeze as well. I mean the blocks would be large and expensive, but the builds that would be possible with this would look so damn cool. You could maybe even have AIO coolers for a certain combination of GPU and motherboard.
So many possibilities!
I really like it, but for some reason I doubt it will ever see the light of day. It would probably not be lucrative enough.

iFreilicht said:

Isn't mITX 170*170, not 175*175? Doesn't matter, onto the idea itself.

Click to expand...

WHOOPS! You are 100% correct. thats ok though because you'd need some clearance anyway.

iFreilicht said:

I love it! And I am on your side with this idea being best suited for the actual card manufacturers. They break so many parts of the standard already, I don't see that being the reason for them to not at least try this.

However, there are a few problems this solution entails.
The first is the shape of the PCB. Making PCBs in this V-Shape would have implications for manufacturing (more useless cutoff), shipping (larger box with a lot of wasted space) and stability (PCB will easily flex and bend. Not problematic when installed, but during installation, care is required.)

All these issues can be fixed, though. You could divide the card into two sections that use a high pin count board interconnect to alleviate them quite nicely. Backplates would work quite well, too.

The idea of having a standardised output connector is nice, but I don't see it flying. In the first wave of this idea, you would more likely use the space available to solder four vertical HDMI and display port connectors (one for each slot) while leaving the rest of the area for venting.

Something that is absolutely awesome about this is the additional size you gain for coolers and fans. I don't know what the best solution would be, but it seems that you could have an 80mm fan at the front that blows air through the whole bottom of the card.
A lot of possibilities there.

Watecooling could be a breeze as well. I mean the blocks would be large and expensive, but the builds that would be possible with this would look so damn cool. You could maybe even have AIO coolers for a certain combination of GPU and motherboard.
So many possibilities!
I really like it, but for some reason I doubt it will ever see the light of day. It would probably not be lucrative enough.

Click to expand...

All valid points. The AIO market is a big part of it too. Current GPU AIO solutions ignore VRAM and do really scary shit IMHO. (looking at you NZXT Kraken) The 1/2 way decent ones still use a blower on a heatsink to those components, and just the sealed block on the GPU itself.

My thinking was they could put enthusiast class tall heatsinks on those parts and treat it just like a regular motherboard does, and the GPU itself would get its own heatsink and air cooler. If temps weren't acceptable doing this, then you could layout the board to have two 92mm or 120mm if you were ok with the amount of potential overhang and greatly simplify the process of cooling the Video card components.

Ideally it would work amazing in a 2u tall case, and could even be crammed into a 1u for the crazy.

No more fighting against gravity, brackets, non sense.

If small form factor video cards keep gaining traction, then ok great, but wouldn't it be better to just lay them flat?

I just think its a better use of the space.

In a similar fashion, I've always loved the idea of back-to-back mobo and gpu in a UCFF, kind of like the way the Brix gaming machines do. Except instead of sandwiching the heatsink in between them (like the Brix, in my opinion, stupidly), have both chips facing outward, and mount the whole assembly in the middle of the case, so that each side of the case could have a direct air intake using low profile heatsinks, one on the CPU side, and the other on the GPU side of the case, with a top vent for the air to escape. It'd be nice to see something like a 970M or 980M and a NUC form factor mobo (in order to compete with the Alienware Alpha).

[EDIT]: I suppose that wouldn't be a PCIe ribbon, but an MXM extension cable...?

Blorgon said:

In a similar fashion, I've always loved the idea of back to back mobo and gpu in a UCFF, kind of like the way the Brix gaming machines do. Except instead of sandwiching the heatsink in between them, have both chips facing outward, and mount the whole assembly in the middle of the case, so that each side of the case could have a direct air intake using low profile heatsinks, one on the CPU side, and the other on the GPU side of the case, with a top vent for the air to escape.

Click to expand...

Isn't that kind of what inspired the DanA4 case:



mobo and vid card back to back, essentially.


EIDT: Oh you meant NUC sized mobo, and mobile GPU chips. I understand now. Man that would be tiny.

d0ublethink said:

Isn't that kind of what inspired the DanA4 case

Click to expand...

Yes! Originally, I had a similar design as the A4, but using a 1U powersupply. Also originally, Dan's case was designed for an HDplex power supply. But because the HDplex power supply only goes up to 250W, he changed to an SFX design, allowing for full size GPUs.

His case is gorgeous and I can't wait to order one, but I'm also into going smaller, since I don't have any interest in ultra gaming, nor at anything higher than 1080p, and with the technologies being developed these days, it's becoming easier and easier for me to dream smaller and smaller.

EIDT: Oh you meant NUC sized mobo, and mobile GPU chips. I understand now. Man that would be tiny.

Click to expand...

Ever since the Brix came out, I've been obsessively fooling around with NUC case designs based on low-noise air cooling. And yes, it would be tiny. Smaller than the Alienware Alpha, cooler looking, and more powerful. That's what I want.

But to get back on topic, I agree with earlier comments about a V-shaped PCB being unstable.

What I don't understand is how Dell and Alienware created their Alpha all-in-one PCB, but that this same approach isn't being pursued by other manufacturers. Is UCFF gaming really that much of a niche market still?

Instead of OP's separate GPU PCB, why not offer a mATX-sized PCB with a socketed GPU like the Alpha but for desktop class parts?

How much overlap in non-processing chip components is there between, say, a 960 and a 980? Would changing between the two be as simple as swapping the chip itself, or do you need to change capacitors, memory modules, and stuff like that? I ask because I'm curious as to whether socketed GPUs could ever actually be a thing.

Blorgon said:

Yes! Originally, I had a similar design as the A4, but using a 1U powersupply. Also originally, Dan's case was designed for an HDplex power supply. But because the HDplex power supply only goes up to 250W, he changed to an SFX design, allowing for full size GPUs.

His case is gorgeous and I can't wait to order one, but I'm also into going smaller, since I don't have any interest in ultra gaming, nor at anything higher than 1080p, and with the technologies being developed these days, it's becoming easier and easier for me to dream smaller and smaller.



Ever since the Brix came out, I've been obsessively fooling around with NUC case designs based on low-noise air cooling. And yes, it would be tiny. Smaller than the Alienware Alpha, cooler looking, and more powerful. That's what I want.

Click to expand...

Not to poo-poo your idea, but wouldn't at that point the approach of buying a 970M laptop, and gutting that, mounting the internals into a slim case, and beefing up the cooling a bit. Plus you'd already have a working power brick.

Another interesting question to pose is do you think the GTX 950 will get a 2u 1/2 height variant?

A socketed GPU is unlikely to happen, as GPU lines are so small and change so quickly. E.g. GM200 has two chip variants, the entire Maxwell 2.0 line has 6 variants, and socketed Haswell has 80. It doesn't make sense to have all the compromises of a socketed chip vs BGA with very little to gain from it, particularly with how fast GPU designs can shift. If GPUs were socketed, AMD would have just made every socket incompatible with the release of GPUs with HBM, and will do so again with HBM2.

HBM actually makes things a lot easier since you no longer have to deal with memory. The socket would only need to provide PCIe, power, and outputs. And BIOS I guess.

Has anyone looked into using PICMG 1.3 single board computers? There are half-length boards that are the size of a short GPU (e.g. ITX 970). I also found something called butterfly backplane which would effectively allow the kind of layout where both the CPU and GPU face the same direction but sadly those are only available in full-length.

d0ublethink said:

Not to poo-poo your idea, but wouldn't at that point the approach of buying a 970M laptop, and gutting that, mounting the internals into a slim case, and beefing up the cooling a bit. Plus you'd already have a working power brick.

Click to expand...

The point is that it would be the size of a NUC. Hell, it would be a NUC. A laptop with the required internals would cost an arm and a leg, and I wouldn't even be using the chassis or the keyboard or the screen. I might not even use the included RAM, if I wanted to use something else. I also don't really understand your comment about the power brick. Laptop motherboards are strangely shaped, the CPUs are generally soldered down (I don't follow the laptop industry too closely; is there such thing as a socketed laptop?)... the list goes on and on.

Here's a picture to give a rough idea of the size of a NUC, in case you're not familiar with the form factor:

vipz said:

HBM actually makes things a lot easier since you no longer have to deal with memory. The socket would only need to provide PCIe, power, and outputs. And BIOS I guess.

Has anyone looked into using PICMG 1.3 single board computers? There are half-length boards that are the size of a short GPU (e.g. ITX 970). I also found something called butterfly backplane which would effectively allow the kind of layout where both the CPU and GPU face the same direction but sadly those are only available in full-length.

Click to expand...

PICMG 1.3 isn't really suited for SFF applications. Yes, the board itself is very small, but it's not able to operate on it's own and I don't even think you get PCIe out of them. Might be wrong though. If something like this thing in the picture exists and it had PCIe x16 on one side and the SHB slot on the other, it might be interesting.
It's hard to tell. the picture you posted is an unlinked thumbnail.

Blorgon said:

Yes! Originally, I had a similar design as the A4, but using a 1U powersupply. Also originally, Dan's case was designed for an HDplex power supply. But because the HDplex power supply only goes up to 250W, he changed to an SFX design, allowing for full size GPUs.

His case is gorgeous and I can't wait to order one, but I'm also into going smaller, since I don't have any interest in ultra gaming, nor at anything higher than 1080p, and with the technologies being developed these days, it's becoming easier and easier for me to dream smaller and smaller.

Ever since the Brix came out, I've been obsessively fooling around with NUC case designs based on low-noise air cooling. And yes, it would be tiny. Smaller than the Alienware Alpha, cooler looking, and more powerful. That's what I want.

Click to expand...

That's actually not quite true. The HDPlex can deliver up to 400W when actively cooled, which is what he did with the default A4. But it was concluded that the SFX design would offer better part compatibility and less hassle on his side, so he decided to make that first, at least that's how it seemed to me.

Smaller than the Alpha and more powerful sounds like quite a challenge. But I wish you the best of luck!

Blorgon said:

The point is that it would be the size of a NUC. Hell, it would be a NUC. A laptop with the required internals would cost an arm and a leg, and I wouldn't even be using the chassis or the keyboard or the screen. I might not even use the included RAM, if I wanted to use something else. I also don't really understand your comment about the power brick. Laptop motherboards are strangely shaped, the CPUs are generally soldered down (I don't follow the laptop industry too closely; is there such thing as a socketed laptop?)... the list goes on and on.
[...]

Click to expand...

Nope, socketed laptops don't exist as far as I know. And yes the boards have odd form factors. But you could be getting a very darn slim PC out of one of those.

iFreilicht said:

PICMG 1.3 isn't really suited for SFF applications. Yes, the board itself is very small, but it's not able to operate on it's own and I don't even think you get PCIe out of them. Might be wrong though. If something like this thing in the picture exists and it had PCIe x16 on one side and the SHB slot on the other, it might be interesting.
It's hard to tell. the picture you posted is an unlinked thumbnail.

Click to expand...

The backplane is required. I don't think a "1U butterfly" style backplane (SHB and PCIe on opposite sides) exists at half-length, but here's a half-length one that has a 16x and a 4x slot on the same side:



Here's a full-length 1U butterfly backplane:



Here's a < 5 liter chassis:

Good enough Are you sure that case is under 5L? It looks a bit larger than that to me.

Honestly I only took the LxWxH figures from the site but if you look at just how much area the expansion slots take up in the back it doesn't seem unreasonable.

iFreilicht said:

Good enough Are you sure that case is under 5L? It looks a bit larger than that to me.

Click to expand...

232x232x90mm is 4,84L

Blorgon said:

The point is that it would be the size of a NUC. Hell, it would be a NUC. A laptop with the required internals would cost an arm and a leg, and I wouldn't even be using the chassis or the keyboard or the screen. I might not even use the included RAM, if I wanted to use something else. I also don't really understand your comment about the power brick. Laptop motherboards are strangely shaped, the CPUs are generally soldered down (I don't follow the laptop industry too closely; is there such thing as a socketed laptop?)... the list goes on and on.

Here's a picture to give a rough idea of the size of a NUC, in case you're not familiar with the form factor:

Click to expand...

I am familiar with the NUC. If portability was the goal then it just seems like a lot of the work is already done in the laptop. Yeah they aren't modular and are pricy. I see your point. To me, portable = battery. That's just me though. I get that.

It boils down to what each person wants from SFF and what their needs are. I just want best in class cpu/gpu in a 2u height chassis, to me 90mm height is fine. Regular power supplies are fine. It's just if you look around this forum you have all these people trying to engineer around certain limitations that just don't need to exist anymore. If I want to drop 600$ or more on an extremely high end video processing unit, why is my only option a fucking 12 inch long tank turn sideways with a crappy crammed heatsink/blower.

If the GPU mfgs insist they need that much pcb space, then let people lay it flat in an actual form factor. So there can be competition and advancement, just as there has always been. And won't threaten their core business model to boot.

Maybe it's even possible to utilize a small riser board like the Asus maximus does to get it all on an ITX length pcb. Even better. But flat is still where it should go. Your not targeting SLI/Crossfire with these ITX Cards anyway.

iFreilicht said:

That's actually not quite true. The HDPlex can deliver up to 400W when actively cooled, which is what he did with the default A4. But it was concluded that the SFX design would offer better part compatibility and less hassle on his side, so he decided to make that first, at least that's how it seemed to me.

Click to expand...

You're right, it was more about compatibility. But I do think a lot of people were concerned initially with pushing that little unit so hard. At least I was. But I'm glad he went in the direction he did, for what its worth.

Smaller than the Alpha and more powerful sounds like quite a challenge. But I wish you the best of luck!

Click to expand...

Ha, well, I'm really just fantasizing right now. More of a "wouldn't this be awesome, why isn't it already being done," kind of thing, as opposed to "I'm planning on doing this."

Nope, socketed laptops don't exist as far as I know. And yes the boards have odd form factors. But you could be getting a very darn slim PC out of one of those.

Click to expand...

Slim, yes. And that'd be an interesting project, certainly. If I ever have $3000+ to spare, I'll give it a go!

vipz said:

Honestly I only took the LxWxH figures from the site but if you look at just how much area the expansion slots take up in the back it doesn't seem unreasonable.

Click to expand...

SaperPL said:

232x232x90mm is 4,84L

Click to expand...

Ahhh sorry guys I didn't see the picture was a link. 4.84L is awesome! I'm a bit worried about cooling the CPU, but the fact they can fit it all in that tiny box is astonishing already.

Blorgon said:

[...]

Slim, yes. And that'd be an interesting project, certainly. If I ever have $3000+ to spare, I'll give it a go!

Click to expand...

Yeah my thoughts exactly

Bumping this thread because I was thinking of PICMG 1.3/SHB Express (the standard has both names, I'll use SHBe for expediency) for other reasons, then stumbled on the half-length variant's existence, and realized the implications for SFF.

I'll note that measurements are coming from version 0.9 of the SHBe spec, not the current release. That one's behind a paywall it seems.

So, full-length SHBe follows the ISA (as in IBM PC/AT) standard for card sizing (and, importantly, component positioning - I'll get into why this matters later), meaning full-length is 13.33" or 338.58 mm (longer than a full-length PCI/PCIe card, even, which is 312 mm), and full-height is 4.98" or 126.39 mm not including the slot bracket (taller than PCI/PCIe, at 111.15 mm - good for tall gaming GPUs). Half-length still follows the ISA standard, except length follows the PCI standard of 6.60" or 167.64 mm - Mini-ITX length just about. This nicely ties into SFF applications, for very, very obvious reasons... but the added height over a normal PCIe card means that tall GPUs are more likely to fit. (Asus DirectCU Mini cooler, anyone?)

So, SHBe half-size offers the following: PCIe x16 (or bifurcated configurations), PCIe x4 (or 4x PCIe x1), and ATX-compatible power input. Full-size adds more power input, USB/SATA/Ethernet breakouts, and optional 32-bit PCI support on the last connector.

AFAIK, once you've supplied power to the backplane (which is often ATX power), the SHB (System Host Board, the name for the CPU board in a SHBe system) is otherwise self-contained - it's a full PC on one board, really.

Now, why is ISA component positioning important? It means that SHBe is mirrored from PCI - the component side of the board is the top side (from the perspective of a normal tower case). This means that a SHB can be above a GPU in the backplane, and the component sides (and therefore HSFs) are facing away from each other, just like in the DAN A4 SFX.

However, the example SHBe backplanes have the SHB in the bottom slot of the backplane, and the Advantech backplanes follow this example. (The spec does explicitly allow the SHB slot to be in any position, though, and the examples in the spec are merely examples.) This means that the GPU cooler will be facing the CPU HSF, which is definitely bad unless you're watercooling.

In any case, I've been thinking about a layout for a SHBe gaming case today, and came up with the following:

PSU in bottom of case, either a Flex-ATX unit mounted flat or a SFX-L unit mounted on its side will do, depending on power demands
SHB takes 1.6" of width (and SHBs are designed for 2 slots worth of space and the resulting cramped cooling layout of having cards immediately above it - the Skylake ones seem to stick to the 65 W TDP parts as a result), PCIe GPU takes 1.6" of width, add .3" to make it 3.5" wide for a bit more cooling space (you're still running a very low profile cooler unless you make it wider)
7" deep chassis - this should be deep enough to take the backplane, SHBe card, and a Mini-ITX GPU
Backplane has the SHB slot in position 2, and a PCIe x16 slot in position 3, with a M.2 2280 on the backside (PCIe x4 only, because SATA isn't broken out onto the backplane in half-size SHBe), taking ATX power from the PSU
About 5.6" of height needed for the SHB (with slot bracket), add 1.6" for Flex-ATX PSU, 5" for SFX-L PSU, if Flex-ATX PSU is used, add 0.8" height for drive clearance under the PSU

Upshot, you've got an 8" H x 3.5" W x 7" D chassis - 3.2 liters - with Flex-ATX, or 10.6" H x 3.5" W x 7" D chassis - 4.3 liters - with SFX-L, and that gets you at least a 65 watt if not 91 watt CPU and a R9 Nano or GTX 970. Take it out to the DAN A4 SFX's width at 4.4" and you get better cooling, and the case is still 4.0 liters in Flex-ATX or 5.4 liters in SFX-L. Or, put a standard SFX power supply in up at the front of the case beside the SHB ala DAN A4 SFX, let's say the case is 6" tall for clearance (it'd be damn tight though), and you get support for longer GPUs in a 6" H x 4.4" W x 12" D case (may need to be deeper for connector clearance) - 5.2 liters.

Rather than placing a miniITX board in a microATX chassis, with an unwieldy GPU 'card' filling in the remaining board space, I would propose the following&#8230;

A miniITX board with no PCIe 3.0 x16 slot on the frontside, but a MXM slot on the backside, along with a M.2 slot for the system drive.

I imagine the Asus Maximus IX Impact motherboard, with a Pascal MXM GPU & M.2 NVMe SSD (yes, Asus would come to its senses and move the U.2 connectors to the mATX & larger ROG boards & return the M.2 slot to the M9I), could make for a highly compact, yet very powerful gaming rig. Could also be a nice compact VR rig.

The difference being that any mATX case would be compatible with the proposed solution, while no existing case would be compatible with the mITX with MXM on the back. Unless you can make a watercooling solution about 5mm thick from block to tube or have a case that has a lot of room on the back.

Also, MXM isn't a very popular format, only available to OEMs and resellers at a much higher price for basically only mobile chips (excluding the one GTX 980 for ~1250$).

Phuncz said:

The difference being that any mATX case would be compatible with the proposed solution, while no existing case would be compatible with the mITX with MXM on the back. Unless you can make a watercooling solution about 5mm thick from block to tube or have a case that has a lot of room on the back.

Also, MXM isn't a very popular format, only available to OEMs and resellers at a much higher price for basically only mobile chips (excluding the one GTX 980 for ~1250$).

Click to expand...

So, we go from an as-yet created L-shaped custom GPU 'card', one designed to fill in the 'dead space' left by mounting a mITX motherboard into a mATX chassis; and that is okay…

But we mention integrating an already existing format (MXM) onto the backside of a mITX board, and that is inconceivable…?!?

As for the argument of space constraints for a MXM behind the motherboard, how would this really be different than the 'sandwich' arrangement one finds in, say, the A4-SFX…?

Yes, there are not a lot of MXM GPU boards available to the general market; but I am quite sure there are WAY more than this proposed mATX Fill-in GPU assembly…

Just stop and think about a high-end (theoretical ROG M9I, of course) mITX board with GPU & system drive being MXM & M.2 backside units… Throw in a HDPLEX PSU, with the internal 'power brick', and you have a very compact, yet (probably) quite powerful gaming rig!

Blorgon said:

(I don't follow the laptop industry too closely; is there such thing as a socketed laptop?)

Click to expand...

iFreilicht said:

Nope, socketed laptops don't exist as far as I know.

Click to expand...

Since this thread was bumped, let's just nip this one in the bud.

Yes, socketed mobile CPUs do exist, however I think Intel has moved away from doing that, probably because few people upgrade their laptops in this manner, plus there are electrical and packaging benefits (both important with laptops) of soldered on CPUs.

Check out this example:
Intel® Core™ i7-3820QM Processor
(8M Cache, up to 3.70 GHz)

Package Size 37.5 x 37.5mm (rPGA988B); 31.0 x 24.0mm (BGA1224)
Sockets Supported FCBGA1224, FCPGA988

Click to expand...

This model was available both socketed and soldered, and you can see a package size difference. Here's how you can read the specs:

FC = Flip Chip, die is on top of the package. Mobile chips usually have bare die, while desktop chips usually have a heat spreader.
BGA = Ball Grid Array, which means soldered on. The connectors on the CPU would basically have metal balls on the bottom which contacts the solder points on the motherboard PCB.
PGA = Pin Grid Array, which have pins on the bottom of the CPU. Mobile socketed CPUs kept going this route, I think due to packaging size. The socket would have a screw which locked the CPU in. Very low profile.
LGA = Land Grid Array, which are the pads on desktop CPUs.
1224, 988, etc. = The number of pins/pads/connectors on the CPU.

You'll find mobile PGA chips in legacy products, but I don't recall seeing any current product made that way.

Haswell-M is the last generation of Intel socketed mobile CPUs, and Haswell-H is the soldered version of Haswell-M, for comparison.

The closest comparison is between i7-4700MQ and i7-4700HQ. Both are 2.4-3.4 GHz 4C8T CPUs with HD Graphics 4600 (although MQ is 0.4-1.15 GHz, versus HQ being 0.4-1.2 GHz) at 47 W TDP. Package is actually the same size, although HQ has more balls than MQ has pins, which will affect current delivery capability negatively for MQ.