Nvidia Details GeForce RTX 4090 FE PCB: 23 Phases and Clean Power

News
By Anton Shilov
last updated

Nvidia improves GeForce RTX 4090 FE's overclocking potential with an all-new VRM design.

Nvidia GeForce RTX 4090 Founders Edition
(Image credit: Nvidia)

Nvidia on Wednesday revealed the printed circuit board (PCB) design of its GeForce RTX 4090 graphics card and disclosed some of the peculiarities of its new product. As it turns out, the company completely redesigned the card's voltage regulating module (VRM) by adding a PID controller with a feedback loop to ensure proper power supply to the GPU and maximize its overclocking potential.

When fully assembled, Nvidia's GeForce RTX 4090 Founders Edition graphics card resembles its predecessor as it continues to use an extensive dual-fan cooling system featuring a unibody design. But the new product has several distinctive differences outside and, more importantly, inside.

Just like previous generation Founders Edition graphics cards, the new GeForce RTX 4090 FE uses a very sophisticated yet compact PCB with a triangular cut on its back (or its right side, depending on how you look at it) that is meant to maximize the efficiency of Nvidia's flow through thermal design (which is further boosted by bigger fans).

(Image credit: Nvidia)

The PCB has a single 12+4-pin 12VHPWR additional PCIe Gen5 power connector that can deliver up to 600W of power to the device, yet default BIOS settings do not allow Nvidia's GeForce RTX 4090 FE to draw more than 450W. The board also comes with a 20+3-phase VRM (20 phases for the GPU, 3 phases for memory), which is a slight improvement from the 18+3-phase power circuitry of the GeForce RTX 3090/3090 Ti FE.

In addition to more power phases, Nvidia implemented a PID controller with a feedback loop to reduce current spikes and drops, effectively stabilizing power delivery to its GeForce RTX 4090 Founders Edition graphics card, reports TechPowerUp. Nvidia says that its new VRM enables a substantially lower power management response time without quantifying it in absolute figures. A stable and clean power supply is crucial for stability under high loads and overclocked conditions. The green company calls this feature 'power transient management,' which might be one of the ways it manages to push its AD102 graphics processor to 3 GHz without overvolting it and using extreme cooling methods.

Nvidia

(Image credit: Nvidia)

In addition to enhancing the cooling system with bigger fans that provide up to 20% more airflow and redesigning power delivery, Nvidia also uses new GDDR6X memory chips with its GeForce RTX 4090 FE that consume less power. While we can only speculate how the power consumption of DRAMs featuring PAM-4 signalings was reduced, we suggest that Micron uses a thinner fabrication process to make these chips, which automatically lowers their power consumption. Meanwhile, with colder memory, the cooling system will better cool down the GPU, which means extra overclocking headroom.

Nvidia's GeForce RTX 4090 Founders Edition graphics cards will go on October 12 and will likely join the best graphics cards, at least for those willing to pay $1,599 per unit.

(Image credit: Nvidia)
Anton Shilov
Anton Shilov
Freelance News Writer

Anton Shilov is a Freelance News Writer at Tom’s Hardware US. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

See more GPUs News
17 Comments Comment from the forums
  • -Fran-
    I'd imagine that the partners don't have access to this, so all AIB cards will have crappy power management as always?

    I mean, nVidia is talking about their FE card here, not about this being part of the reference design the AIBs have access to, right?

    Regards.
    Reply
  • spongiemaster
    -Fran- said:
    I'd imagine that the partners don't have access to this, so all AIB cards will have crappy power management as always?

    I mean, nVidia is talking about their FE card here, not about this being part of the reference design the AIBs have access to, right?

    Regards.
    It's not a question of access, but of cost. Some AIB's cheap out, some don't. Choose wisely.
    Reply
  • -Fran-
    spongiemaster said:
    It's not a question of access, but of cost. Some AIB's cheap out, some don't. Choose wisely.
    That's the point of a "reference design". That sets the bar for how low they can go. And that's also the problem with nVidia makings statements like these using the FE as an example. They're implying underneath that all partner cards will be like the FE models, which is untrue.

    I would love to be wrong though, but the 3K series launch made it abundantly clear that FE is not being used by partners as a reference or even getting proper guidance from nVidia. Specially on the power side. Do not forget the 3080 power issues at launch due to boost behaviour set as FE cards when the "true" reference design didn't have the same power delivery structure/fortitude.

    Regards
    Reply
  • spongiemaster
    -Fran- said:
    That's the point of a "reference design". That sets the bar for how low they can go. And that's also the problem with nVidia makings statements like these using the FE as an example. They're implying underneath that all partner cards will be like the FE models, which is untrue.

    I would love to be wrong though, but the 3K series launch made it abundantly clear that FE is not being used by partners as a reference or even getting proper guidance from nVidia. Specially on the power side. Do not forget the 3080 power issues at launch due to boost behaviour set as FE cards when the "true" reference design didn't have the same power delivery structure/fortitude.

    Regards
    FE isn't a reference model. Nvidia isn't implying anything. They're telling us about their FE model.
    Reply
  • JarredWaltonGPU
    -Fran- said:
    I'd imagine that the partners don't have access to this, so all AIB cards will have crappy power management as always?

    I mean, nVidia is talking about their FE card here, not about this being part of the reference design the AIBs have access to, right?

    Regards.
    I believe with Nvidia's partners, they have a few options:

    Buy the reference PCB from Nvidia along with the GPU, VRAM, and various other bits. About the only things missing are the cooler, fans, and RGB.
    Roll their own PCB, based off Nvidia's design documents. Nvidia still has to validate the final design, so this is quite expensive.
    There's probably some places that make a variety of PCBs loosely based off the reference design, and an AIB could pick something like this and then ask for a few tweaks.
    Note that the reference design isn't actually the FE board, however. I'm pretty sure there's a different reference that doesn't have the cut outs and a few other differences from the FE board.
    Reply
  • -Fran-
    spongiemaster said:
    FE isn't a reference model. Nvidia isn't implying anything. They're telling us about their FE model.
    I know. That's the whole point I'm making here: nVidia talking about the FE power delivery does not mean the AIB cards will have the same power delivery setup. It's misleading.

    JarredWaltonGPU said:
    I believe with Nvidia's partners, they have a few options:

    Buy the reference PCB from Nvidia along with the GPU, VRAM, and various other bits. About the only things missing are the cooler, fans, and RGB.
    Roll their own PCB, based off Nvidia's design documents. Nvidia still has to validate the final design, so this is quite expensive.
    There's probably some places that make a variety of PCBs loosely based off the reference design, and an AIB could pick something like this and then ask for a few tweaks.Note that the reference design isn't actually the FE board, however. I'm pretty sure there's a different reference that doesn't have the cut outs and a few other differences from the FE board.
    I know the FE isn't the "reference design". That's the point I'm trying to make precisely. nVidia is not showing the "capabilities" of the "bare minimum" of* the 4090's power delivery looks like from neither their own Ref Model or whatever their partners come up with, but their special boutique version built by themselves. That's a very important and relevant distinction to make here.

    And #3 looks a tad redundant with #2, unless they go to Foxconn or BYD, which already has made nVidia PCBs, but I doubt they'd be allowed to give them the same PCB as the FE cards? Or even a reference design model by nVidia? Although that seems plausible if nVidia actually tells AIBs which OEM worked with them making prototypes (if they do; I dunno) and such.

    It would be really interesting to have a bit more insight how nVidia gives AIBs the "bare minimum" a PCB can have and/or be. Assuming those 3 are not the only options, that is.

    Regards.
    Reply
  • wifiburger
    neat, all you have to do now if you ever get one is not play with the pcie5 power connector / cable too much

    apparently there's a huge failure with them at this stage due to high power usage; not a really good design, bending them, insert/remove / insert they catch fire / melt :ROFLMAO:
    Reply
  • artk2219
    wifiburger said:
    neat, all you have to do now if you ever get one is not play with the pcie5 power connector / cable too much

    apparently there's a huge failure with them at this stage due to high power usage; not a really good design, bending them, insert/remove / insert they catch fire / melt :ROFLMAO:
    Honestly getting an extender and just using that would save the wear and tear on the receptacle, youd still need to worry about the cable though. I also see many of these being adapted from three 8 pin connectors instead of being a native cable. That being said, that spec DEFINITELY needs some work, only rated for 40 connections? Do test power supplies or test cards not exist for? You could easily go through 5 connections just putting the thing together, initial test, test boot, oh shoot forgot this cable, man its easier to do this without the gpu in there, maybe i should install this other card first, oh damn thats where the SSD goes. Granted you dont always have to unplug for those, but 40 is a pitiful amount of rated connections, thats like one batch of cards for a system builder. If you're using a test card then you really want to use a dongle instead of plugging it in directly every time. Eh, i guess we'll see how that shakes out.
    Reply
  • newtechldtech
    The big Question is : do we really need RTX 4090 for gaming ?? we have enough fps already for majority of games. unless we move to 8K screens I dont see any use for RTX 4090 that will make a difference in game play. I think that RTX 4080 is the wall unless we move to K resolution.
    Reply
  • TJ Hooker
    artk2219 said:
    That being said, that spec DEFINITELY needs some work, only rated for 40 connections?
    Where have you seen that the spec for the 12VHPWR connector is only 40 connect/disconnects?
    Reply