Clevo with Framework GPU interconnect standard?

[KabyZen]

So I was watching the Linus Tech Tips videos on his Framework investment and learned that they have developed a modular GPU standard similar to the Dell Alienware (DGFF). The only catch is that they are publishing the standard for anyone to use.

 

My question is that since MXM is dead could Clevo jump on this and start using this standard to bring back modular GPUs? Could we see a revival of the p870? 😃😮

 

Here's the video below:

https://www.youtube.com/watch?v=UeCdBVHYa_8&t=3m49s

[Aaron44126]

The great benefit to Framework's approach is not just the connector, but putting the GPU (+ cooling for it) in its own little enclosure at the back of the laptop.  This enclosure doesn't have to be a fixed size, just a fixed width.  If you want a more powerful GPU, then the enclosure will be bigger and stick out of the back of your laptop a bit more.  What has doomed MXM and other modular GPU attempts in recent years (i.e. Alienware Area-51M) is basically the fact that each GPU generation these days needs more space and power for the highest-end chips, so they often don't fit within the design constraints of laptops even one generation old.  This is why you see MXM cards that are in weird "larger than standard" shapes no longer physically compatible with other laptop models, Alienware failing to offer upgrades to their Area-51M that was billed as somewhat futureproof, and so on.

 

It looks like Framework may have found a solution for this.  The GPU has a standard connection to the motherboard, but the physical size and power hookups are entirely independent, so there's no reason you couldn't slot in a next-gen GPU in a few years even if they had to make the enclosure a little bit bigger to house it.

 

There are still some unanswered questions which will be addressed in the next few months.  (What specific GPUs are they offering, what do power limits look like, what are the specs of the rest of the laptop like CPUs on offer and display resolution, etc...)

 

Framework seems completely open to other manufacturers copying their approach.  You don't even have to use GPUs from Framework.  A GPU AIB like PNY or Gigabyte could offer GPU modules for Framework laptops — all of the necessary specs are already published.  Whether Clevo thinks it makes business sense to use this in their laptops, we will find out.  (I think some laptop manufacturers are fine with requiring users to buy a whole new laptop when they want to upgrade something.)

 

I say, if you like the Framework approach, and Framework has a laptop that will suit your needs, then consider buying one of theirs; it's the best way to make sure that they stick around.  They have something of a proven track record now, having offered 12th gen, 13th gen, and Ryzen 7000 upgrades for their original 11th gen Framework Laptop 13.  Framework Laptop 13 never interested me (too small + no dGPU), but I'm keenly interested in the full reveal of Framework Laptop 16 and I'm sort of bummed that I just bought a laptop last year, otherwise I'd be very interested in getting this one (but for now I'll have to wait a while before upgrading again).

 

(Framework Laptop discussion over on the Framework Laptop thread.)

[Blacky]

I can't entirely agree that GPUs need more and more power with each generation. The high-end models sure do, but if you stick to a particular thermal envelope, you can use the same standard across multiple generations. The limitation will be that if you bought a GTX 1070 at 100W, then for the next generation, you can only upgrade to, say, an RTX 3060 because that one is now 100W.

 

I use an ancient P177SM-A with standard MXM cards. I have upgraded this laptop to a P4200 Max-Q that works around 85W. I can further upgrade to an RTX 4000 Max-Q, which also comes in MXM format and is 85W: https://www.techpowerup.com/gpu-specs/quadro-rtx-4000-max-q.c3427

 

 

[Sniffy]

My worry is that now they've discontinued LGA and MXM laptops, it will be harder to bring them back. IIRC one of the reasons the X170 continued while the Area-51M and GT76 went away is that Clevo had very established channels through which they received support from Intel to implement LGA in laptops because of Clevo's long history of doing so. If that goes away, it might be too much work to get it going again. Plus, in their marketing Intel and some reviewers also have referred to HX as 'desktop processors in a laptop', even though they clearly aren't. This seems to be the path Intel is headed.

 

On their Discord XMG have pretty much said RIP XMG Ultra and moved on.

 

Edit: to answer the OP's question, I don't think Clevo are interested in swappable components anymore. But I hope I'm wrong.

[KabyZen]

Three things I need before I get another laptop.

 

Socketed Ryzen

Framework-style DGFF

Unlocked BIOS

 

Come on Clevo dont fail me now.

[1610ftw]

7 hours ago, Sniffy said:

My worry is that now they've discontinued LGA and MXM laptops, it will be harder to bring them back. IIRC one of the reasons the X170 continued while the Area-51M and GT76 went away is that Clevo had very established channels through which they received support from Intel to implement LGA in laptops because of Clevo's long history of doing so. If that goes away, it might be too much work to get it going again. Plus, in their marketing Intel and some reviewers also have referred to HX as 'desktop processors in a laptop', even though they clearly aren't. This seems to be the path Intel is headed.

 

On their Discord XMG have pretty much said RIP XMG Ultra and moved on.

 

Edit: to answer the OP's question, I don't think Clevo are interested in swappable components anymore. But I hope I'm wrong.

 

I agree with you assessment of Clevo's stance on this and I think that MSI would be in the best position to do this:

 

They seem to be better connected than Clevo

They have done LGA before and only have stopped with Comet Lake so one gen before Clevo. Before that they had three LGA laptop chassis starting with first gen Skylake.

They also produce motherboards and separate GPUs so they do not only have contacts with the big three for special laptop chips.  That would help implementing a modular solution and getting support from Intel and AMD that is comparable to the support in their desktop motherboards.

 

They have the most elaborate bios options of any laptop these days when it comes to overclocking and undervolting.

 

I have for some time said that next gen CPUs would be perfect for an LGA chassis with Intel processors and a GPU module that is swappable from the outside would be much less of a support nightmare although it may be a challenge to produce a module with built in cooling or alternatively to attach such a module to a unified heat sink that all manufacturers prefer these days.

 

 

[Shark00n]

Quite like the idea and its potential.

 

Let's talk in 3 or 4 generations time 🤣

[KabyZen]

On 3/29/2023 at 12:46 PM, 1610ftw said:

 

I agree with you assessment of Clevo's stance on this and I think that MSI would be in the best position to do this:

 

They seem to be better connected than Clevo

They have done LGA before and only have stopped with Comet Lake so one gen before Clevo. Before that they had three LGA laptop chassis starting with first gen Skylake.

They also produce motherboards and separate GPUs so they do not only have contacts with the big three for special laptop chips.  That would help implementing a modular solution and getting support from Intel and AMD that is comparable to the support in their desktop motherboards.

 

They have the most elaborate bios options of any laptop these days when it comes to overclocking and undervolting.

 

I have for some time said that next gen CPUs would be perfect for an LGA chassis with Intel processors and a GPU module that is swappable from the outside would be much less of a support nightmare although it may be a challenge to produce a module with built in cooling or alternatively to attach such a module to a unified heat sink that all manufacturers prefer these days.

 

 

But MSI is...MSI...they are sinusoidal in their performance lol.

[1610ftw]

1 hour ago, KabyZen said:

But MSI is...MSI...they are sinusoidal in their performance lol.

 

True, but can't be worse than Clevo who could be said to be flatlining for the second generation in a row with the X370 😄

[ymsv]

On 3/29/2023 at 11:51 AM, Sniffy said:

My worry is that now they've discontinued LGA and MXM laptops, it will be harder to bring them back. IIRC one of the reasons the X170 continued while the Area-51M and GT76 went away is that Clevo had very established channels through which they received support from Intel to implement LGA in laptops because of Clevo's long history of doing so. If that goes away, it might be too much work to get it going again. Plus, in their marketing Intel and some reviewers also have referred to HX as 'desktop processors in a laptop', even though they clearly aren't. This seems to be the path Intel is headed.

 

On their Discord XMG have pretty much said RIP XMG Ultra and moved on.

 

Edit: to answer the OP's question, I don't think Clevo are interested in swappable components anymore. But I hope I'm wrong.

 

Not only from Intel , but also from Nvidia which does not want to support MXM anymore . And you will need a support from video cards makers and I think also their " approval " for the product . 

[1610ftw]

4 hours ago, ymsv said:

 

Not only from Intel , but also from Nvidia which does not want to support MXM anymore . And you will need a support from video cards makers and I think also their " approval " for the product . 

 

Intel seems like a less controlling company than Nvidia but still they helped neither Clevo nor MSI to allow users to update from Comet Lake to Rocket Lake.

[Bidelloman]

I would not mind having a dtr with a cpu as efficient as the 7800x3d, giving more headroom for the gpu, whether mxm or interconnect bga

[KabyZen]

On 4/5/2023 at 3:19 PM, Bidelloman said:

I would not mind having a dtr with a cpu as efficient as the 7800x3d, giving more headroom for the gpu, whether mxm or interconnect bga

A computer that Prema would probably not have to tweak...

[Csupati]

Is there any chance that someone will make an Egpu case for the Framework, like the Alienware Graphics Amplifier, or the Thunderbolt version?

I know with the Pcie 3.0 X8 it has bottleneck at the connection, but with the 4.0 X8 is same as the 3.0 X16 , or im waiting for the TB5 release and EGPU .

[Aaron44126]

3 hours ago, Csupati said:

Is there any chance that someone will make an Egpu case for the Framework, like the Alienware Graphics Amplifier, or the Thunderbolt version?

I know with the Pcie 3.0 X8 it has bottleneck at the connection, but with the 4.0 X8 is same as the 3.0 X16 , or im waiting for the TB5 release and EGPU .

Framework themselves indicated that they would release an eGPU enclosure that could fit one of their laptop GPUs. It would be Thunderbolt. Possibly not until they have an upgraded GPU to offer, so people have something to use their “old” one for.

[Csupati]

5 minutes ago, Aaron44126 said:

Framework themselves indicated that they would release an eGPU enclosure that could fit one of their laptop GPUs. It would be Thunderbolt. Possibly not until they have an upgraded GPU to offer, so people have something to use their “old” one for.

Im not talking about laptop GPU , ive thinked about full size desktop GPU. 

Its not better then any gaming laptop if you put a laptop gpu in there . And it will have limitations at power delivery,  but with the egpu enclosure you have only data transfer at the DGFF connector .

[Aaron44126]

Ok, I see what you're saying, you want to attach a desktop GPU to the laptop via the DGFF port.

 

It seems like this would be completely possible for "anyone" to build because the spec for the ports is public, but I haven't heard anything about such a contraption being made.  A couple of "gotchas" occur to me.

• The Framework 16's "empty" expansion bay module which just has fans in it still uses the DGFF ports.  The fan power and speed sensors go through those ports.  So, such an adapter would have to take the form of an expansion bay module with fans AND some kind of data output connector for the eGPU.
• It would still be somewhat cumbersome because to detach at the DGFF ports because you would have to power the system off, snap out the keyboard, and remove the screws holding the DGFF connector in place.  (...This could be "solved" if the data connector between the expansion bay module and the eGPU could be easily detached from the outside.)

Err.  Heck.  They already have an expansion bay module that has two NVMe connectors inside.  Hasn't someone made a NVMe-to-GPU adapter already...?  (Granted, it would be limited to four PCIe lanes unless you could somehow use both NVMe connectors at the same time.)

[Csupati]

1 hour ago, Aaron44126 said:

Ok, I see what you're saying, you want to attach a desktop GPU to the laptop via the DGFF port.

 

It seems like this would be completely possible for "anyone" to build because the spec for the ports is public, but I haven't heard anything about such a contraption being made.  A couple of "gotchas" occur to me.

• The Framework 16's "empty" expansion bay module which just has fans in it still uses the DGFF ports.  The fan power and speed sensors go through those ports.  So, such an adapter would have to take the form of an expansion bay module with fans AND some kind of data output connector for the eGPU.
• It would still be somewhat cumbersome because to detach at the DGFF ports because you would have to power the system off, snap out the keyboard, and remove the screws holding the DGFF connector in place.  (...This could be "solved" if the data connector between the expansion bay module and the eGPU could be easily detached from the outside.)

Err.  Heck.  They already have an expansion bay module that has two NVMe connectors inside.  Hasn't someone made a NVMe-to-GPU adapter already...?  (Granted, it would be limited to four PCIe lanes unless you could somehow use both NVMe connectors at the same time.)

Nvme slots using X4 speed , its very very bottleneck,  and there is no solution in present days for dual Nvme slot Egpu case. 

Every egpu method (except TB ) has to be powered off to remove cabels and connectors . 

For sensors,  the old MSI GT60/70 laptops there was unlocked bios for the empty mxm slot problem.  Maybe someone will unlock it,  or even  Frameworks will add option to the bios .

[Developer79]

On 3/28/2023 at 7:45 PM, KabyZen said:

So I was watching the Linus Tech Tips videos on his Framework investment and learned that they have developed a modular GPU standard similar to the Dell Alienware (DGFF). The only catch is that they are publishing the standard for anyone to use.

 

My question is that since MXM is dead could Clevo jump on this and start using this standard to bring back modular GPUs? Could we see a revival of the p870? 😃😮

 

Here's the video below:

https://www.youtube.com/watch?v=UeCdBVHYa_8&t=3m49s

Yes, it is a pity that the MXM format is dying out. For this reason I thought a few years ago to create a possibility in the P870xx to continue to use the MXM port for the latest graphics cards. I have posted my tests in the P870TM thread with the second development stage. So the newest GFX generation is integrable in the P870xx!

[Csupati]

1 hour ago, Developer79 said:

Yes, it is a pity that the MXM format is dying out. For this reason I thought a few years ago to create a possibility in the P870xx to continue to use the MXM port for the latest graphics cards. I have posted my tests in the P870TM thread with the second development stage. So the newest GFX generation is integrable in the P870xx!

Itsan awesome work, but with Thunderbolt 5 everything will change in  egpu gaming,  

"Thunderbolt 5 will deliver 80 gigabits per second (Gbps) of bi-directional bandwidth, and with Bandwidth Boost it will provide up to 120 Gbps"

So its gonna do the same performance as the mxm egpu version,  but didnt require any screwing or dissasembly.

[Developer79]

12 hours ago, Csupati said:

Itsan awesome work, but with Thunderbolt 5 everything will change in  egpu gaming,  

"Thunderbolt 5 will deliver 80 gigabits per second (Gbps) of bi-directional bandwidth, and with Bandwidth Boost it will provide up to 120 Gbps"

So its gonna do the same performance as the mxm egpu version,  but didnt require any screwing or dissasembly.

It's a simple calculation: Thunderbolt is only as fast as the weakest link in the transmission. It would have to go directly via the CPU. But it probably goes via the chipset. I have, for example, my solution directly connected to the CPU. This allows 16 lanes. On Thunderbolt you would need a distribution chip that manages the 16 lanes. This also costs time. Besides, it would only be a laptop, if not so much stuff is standing around outside 🙂 My solution integrates the desktop GFX!

[Developer79]

On 3/28/2023 at 7:45 PM, KabyZen said:

My question is that since MXM is dead could Clevo jump on this and start using this standard to bring back modular GPUs? Could we see a revival of the p870?

I currently see only my solution that allows to use the MXM port of the P870xx with 16 lanes and integrates a latest GFX generation. With all others everything is outside and more cumbersome. Also the tests from PCIe 3.0 to PCIe 4.0 is not relevant in games! There is pratically no notable difference between the two generations. Also the jump between the CPU generations does not show a huge jump in games. The 13th gen is a bit faster in games, but the 9th gen can still keep up well. It also depends on the motherboard itself, etc. It puts it into relative terms!!!

[Csupati]

24 minutes ago, Developer79 said:

I currently see only my solution that allows to use the MXM port of the P870xx with 16 lanes and integrates a latest GFX generation. With all others everything is outside and more cumbersome. Also the tests from PCIe 3.0 to PCIe 4.0 is not relevant in games! There is pratically no notable difference between the two generations. Also the jump between the CPU generations does not show a huge jump in games. The 13th gen is a bit faster in games, but the 9th gen can still keep up well. It also depends on the motherboard itself, etc. It puts it into relative terms!!!

With TB4 they managed to bypass the PCH and other interconnection,  its directly wired to the cpu.

[Developer79]

3 hours ago, Csupati said:

With TB4 they managed to bypass the PCH and other interconnection,  its directly wired to the cpu.

I think any redirection reduces speed. This is critical!

[jaybee83]

welp, one can dream...i like the way that OP it thinking. however, in the end, Clevo is now following the money trail, thin n light BGA crap FTW! 😢