Without identifying anyone, I can relay that I have heard independently from other sources that nVidia will be supporting consumer GPUs up to at least the RTX 2080, in the MXM format. I don't know how long that support will last, if it will continue in subsequent generations, etc. But those GPUs will be available in MXM to those that hunt around, at least in the near term. This is in striking contrast to what I had heard throughout last year, which was unanimous in saying that nVidia was permanently phasing out MXM for all consumer-grade cards.
To elaborate on this a bit: nVidia has consistently felt uneasy about MXM consumer GPUs, since these compete against board partners as well as (more recently) their own reference designs. They also compromise on the control they have over how the GPUs are implemented, particularly from a cooling and performance context. Finally, their use is pretty uncommon, and thus the sales volume of these cards is quite low. Given all this, nVidia has historically elected to charge a massive premium on these cards to make it worth their while, as well as to counteract some of the negative externalities of MXM support while still providing the format to whoever really wants it.
(To be clear, on the professional side nVidia has steadfastly supported Quadro on MXM for some time, and at reasonable pricing. All indications point to this continuing for the foreseeable future.)
I am relieved that current-gen nVidia cards will land on MXM, because these GPUs are dramatically more space efficient than PCIe cards are. They also provide an opportunity to simplify system cooling solutions, perhaps down to a unified cooler (air or liquid) for the CPU and GPU in some contexts. <5L designs with top-tier GPUs and CPUs all but require this format to be possible; if I had unlimited time and resources, I'd most definitely make a build utilizing MXM to get the absolute best performance density I could. That said, the way nVidia supplies and charges for MXM cards – and the uncertainty around support – means that end users have to eat a huge price premium while simultaneously assuming that there won't be an upgrade path. It's really a shame.
To elaborate on this a bit: nVidia has consistently felt uneasy about MXM consumer GPUs, since these compete against board partners as well as (more recently) their own reference designs. They also compromise on the control they have over how the GPUs are implemented, particularly from a cooling and performance context. Finally, their use is pretty uncommon, and thus the sales volume of these cards is quite low. Given all this, nVidia has historically elected to charge a massive premium on these cards to make it worth their while, as well as to counteract some of the negative externalities of MXM support while still providing the format to whoever really wants it.
(To be clear, on the professional side nVidia has steadfastly supported Quadro on MXM for some time, and at reasonable pricing. All indications point to this continuing for the foreseeable future.)
I am relieved that current-gen nVidia cards will land on MXM, because these GPUs are dramatically more space efficient than PCIe cards are. They also provide an opportunity to simplify system cooling solutions, perhaps down to a unified cooler (air or liquid) for the CPU and GPU in some contexts. <5L designs with top-tier GPUs and CPUs all but require this format to be possible; if I had unlimited time and resources, I'd most definitely make a build utilizing MXM to get the absolute best performance density I could. That said, the way nVidia supplies and charges for MXM cards – and the uncertainty around support – means that end users have to eat a huge price premium while simultaneously assuming that there won't be an upgrade path. It's really a shame.
