Will we ever see a GPU socket on a MBO?

[MadDuke]

Hi guys.

 

I'm just dropping some inspiration that I've got about a thing that I would personally like to see and think it would be cool. 

 

Seeing as GPU's with HBM memory combinations on a single package are a thing. What do you think about the possibility of having a GPU socket together with a CPU socket?

The upfront cost on a MBO would be higher (seeing as you would need additional VRMs for power delivery towards the GPU), and connector wise you would only need to put the PCI-E power connectors let's say next to the 24 pin connector or somewhere along those lines.

 

All the downsides on the "size" problem already exist on the CPU.  If you wish to use big ass air tower coolers. You already need to have a clearance in the case for those so the same space would be needed for the GPU part.  Or going "low profile" with AiO solutions and whatnot. 

 

I've had a PC since the 1990 and assembled my own since the 1999 PIII era. I've found that in the last 10 years or so I've gradually went the route of MicroATX and now Mini-ITX . I've simply haven't had the real need to have anything else connected to the PC itself other than the GPU for some time.  (literally, only thing I need missing on my mini-ITX is a 10 gig NIC). I think most of the people really also don't use anything else on their PC.  And the era of me wanting and/or needing a TV card or 15 years ago popular Skystar satellite cards, PCI audio etc. are a non-issue anymore for me.

 

What are your thoughts on that?

[GoldenLag]

For x86?

 

 

Nope, never. 

 

Simply because memmory and HBM and so much more.

 

Edit: and the memmory controllers required.

[dgsddfgdfhgs]

we already have it, the pcie slot

 

 

[MadDuke]

2 minutes ago, GoldenLag said:

For x86?

 

 

Nope, never. 

 

Simply because memmory and HBM and so much more.

Don't know what x86 has to do with it and/or memory as the GPU  memory is on the same package as the GPU.  Two stacked HBMs of 16 gigs on Vega cards is plenty to start with with even "last gen" hardware wise.  

 

 

 

 

[Labeled]

Probably not, it would complicate compatibility. WIth PCIe you don't need to worry about having the right socket or good enough power delivery, I think if this ever did become a thing GPU performance variance would become a nightmare.

 

You'd also have to get a more specific motherboard, Intel + Nvidia, or Intel + AMD, or AMD + AMD, and AMD + Nvidia, or in the future AMD + Intel, or Intel + Intel.

[MadDuke]

Yeah.   I mean.  It's a conversation starter really.  I don't expect anyone doing it really unfortunately.  But I would be happy as a kid to have a single high performance CPU and GPU with a lets say a big MBO monoblock cooling all the VRMs together with a GPU and CPU

 

[GoldenLag]

4 minutes ago, MadDuke said:

Two stacked HBMs of 16 gigs on Vega cards is plenty to start with with even "last gen" hardware wise.  

You also need to deal with memmory controllers and VRM design aswell as pincount. 

 

There is a lot that needs to be taken care of.

 

4 minutes ago, MadDuke said:

Don't know what x86 has to do with it and/or memory as the GPU  memory is on the same package as the GPU.

Well i believe embedded systems have some semi-custom solutions. 

 

 

X86 i kinda wrong, but it was mostly to rule out semi-custom.

 

Also GDDR is a pain to stack onto the same substrate that also needs to be slotable. 

 

Also custom Board partners wouldnt be that happy.

[greenhorn]

Due to the different form factors of GPUs and the different memory types I don't think so. I could only imagine a GPU manufacturer teaming up with some motherboard manufacturer to create a product together (motherboard + GPU combo), but then the GPU will likely be fixed. This is essentially what we had years ago when onboard graphics weren't integrated into the CPU but consisted of a discrete chip on the motherboard.

[MadDuke]

1 minute ago, greenhorn said:

Due to the different form factors of GPUs and the different memory types I don't think so. I could only imagine a GPU manufacturer teaming up with some motherboard manufacturer to create a product together (motherboard + GPU combo), but then the GPU will likely be fixed. This is essentially what we had years ago when onboard graphics weren't integrated into the CPU but consisted of a discrete chip on the motherboard.

Yep.  Had the Abit nf7-s v2.0.       Nforce 2 was the name I believe as an example.  

 

 

 

[Labeled]

5 minutes ago, MadDuke said:

Yeah.   I mean.  It's a conversation starter really.  I don't expect anyone doing it really unfortunately.  But I would be happy as a kid to have a single high performance CPU and GPU with a lets say a big MBO monoblock cooling all the VRMs together with a GPU and CPU

 

The biggest problem is it hurts modularity.  It would also inflate prices of motherboards a lot considering they'll essentially have the GPU board embedded into it. So now you're splitting the market even more, PCIe based boards and these new GPU socket boards. Monoblocks for these would be incredibly more specific thus being A LOT more expensive.

 

So beefing up the motherboards will cost more, creating a socket and an interface will cost more. This only results in more expensive products. 

[MadDuke]

That is pretty much a given of course in the start

[mr moose]

I think we will see memory and SSD integration first.    GPU is slightly more complicated and has too many variables in end use cases to be viable.

[Mira Yurizaki]

10 hours ago, MadDuke said:

What are your thoughts on that?

It's impractical because of the amount of PCB real-estate needed to support a GPU and its power delivery circuitry. This would easily eat into the expansion slot space.

 

Besides, this would also require NVIDIA, AMD, and possibly Intel to agree on a single socket. Otherwise you'll be left with boards that can only use a specific vendor's GPU. Given AMD and Intel can't even agree on a CPU socket, it makes no sense to try socketed GPUs on a motherboard.

[MadDuke]

7 hours ago, Mira Yurizaki said:

It's impractical because of the amount of PCB real-estate needed to support a GPU and its power delivery circuitry. This would easily eat into the expansion slot space.

 

Besides, this would also require NVIDIA, AMD, and possibly Intel to agree on a single socket. Otherwise you'll be left with boards that can only use a specific vendor's GPU. Given AMD and Intel can't even agree on a CPU socket, it makes no sense to try socketed GPUs on a motherboard.

Of course    I'm talking about it being standardized.   And yes. I know it's freaking far fetched (to say the least about it), but it was simply an exercise in "form factors" which we don't have at the moment (and won't see)  and that I would personally like to see.   I honestly (personally) as mentioned before don't need a single PCI-E slot at all. 

[Tristerin]

17 hours ago, MadDuke said:

Yeah.   I mean.  It's a conversation starter really.  I don't expect anyone doing it really unfortunately.  But I would be happy as a kid to have a single high performance CPU and GPU with a lets say a big MBO monoblock cooling all the VRMs together with a GPU and CPU

 

This is actually a good topic/question imho - Ive been wondering...why is the personal PC still in its original form (basically) after all of these technological advancements in other more crucial areas of technology.  My guess has been that its because people don't like change or to many choices so this keeps it easier and for the plebs even easier.  

 

17 hours ago, Labeled said:

Probably not, it would complicate compatibility. WIth PCIe you don't need to worry about having the right socket or good enough power delivery, I think if this ever did become a thing GPU performance variance would become a nightmare.

 

You'd also have to get a more specific motherboard, Intel + Nvidia, or Intel + AMD, or AMD + AMD, and AMD + Nvidia, or in the future AMD + Intel, or Intel + Intel.

I agree to this assessment as upgrade paths would be super limited, therefore a niche market, and niche markets aren't a great thing if even a few percent of your base falls out.

[Mira Yurizaki]

43 minutes ago, MadDuke said:

Of course    I'm talking about it being standardized.   And yes. I know it's freaking far fetched (to say the least about it), but it was simply an exercise in "form factors" which we don't have at the moment (and won't see)  and that I would personally like to see.   I honestly (personally) as mentioned before don't need a single PCI-E slot at all. 

If you don't need a PCIe slot, then why would a GPU socket be any different?

[Drak3]

Yes. We already have.

 

It's called MXM. It's expensive and not nearly as good as GFX cards.