High-Power (600-Watt) PCIe 5.0 Power Connector for Next-Gen GPUs Pictured, Rumored to Debut with NVIDIA GeForce RTX 3090 Ti

[Tsing]

Image: Igor’s Lab



Igor’s Lab has shared diagrams of the new power connector that will supposedly debut with NVIDIA’s GeForce RTX 3090 Ti graphics card. Representing PCI-SIG’s new “High Power Connector” (H+) standard, the connector is a new, non micro-fit design that features a total of 16 pins, four of which are located at the bottom. The new standard can support up to 600 watts of power for next-generation graphics cards.







From Igor:



The new 12VHPWR connector […] does not fit the PCI Express 2×3 and 2×4 Auxiliary Power connectors. The power supply pins of the 12VHPWR connector have a spacing of only 3.0 mm, while the contacts...

Continue reading...

 

[Brian_B]

I thought nVidia already had their new connector?

 

[Grimlakin]

I thought nVidia already had their new connector?

Perhaps I mis read but this is a new power connector for additional power to be delivered over the pcie bus for demanding cards like video cards.

 

[Peter_Brosdahl]

I thought nVidia already had their new connector?

They did but this is something a little different. It was already rumored earlier last week that a 16-pin connector could be coming, and this somewhat is but instead 12+4. I'm sort of happy about it. Having 3x8-pin connectors was definitely getting complicated. Since these cards could easily pull upwards of 500W, my Suprim 3090 can do around 450W when overclocked, it makes sense they're planning for the future. What I'm not happy about is another connector that the industry hasn't adopted yet. Simplicity and efficiency don't matter when you have to get either custom cables or adapters unless the cards come with them and even then it'll probably still have to be split across 2 or 3 8-pin ports at the PSU.

The one takeaway I have from this is that we could end up seeing the PSU port revisions somewhere between PCIe 5.0 and PCIe 6.0 with more mainstream adoption by or after PCIe 6.0.

 

[Grimlakin]

I have to wonder at the leaps that we are seeing with bus bandwidth. Specifically along the lines of PCIE. I know for PCI-X (a commercial version.) that the bandwidth was a huge deal for I/O. I'm wondering what the big gain here is for PCIE. I we stick with the doubling for easy math PCIE 4.0 was 2x as a fast as PCIE 3.0 and we are JUST starting to see things that need PCIE 4.0 for anything. Maybe... and that really is debateable. Now we are going to introduce PCIE 5.0 that should be a 2x improvement over PCIE 4.0. Considering 4.0 was 65GB a second and 5.0 would be 128GB a second. Short of enterprise use... where do we need that level of bandwidth in the consumer space? MAYBE in the prosumer space for people that are running home NAS that want an all NVME setup. (No reason you couldn't split out a PCIE 6.0 lane to support 4 PCIE 4.0 lanes or one PCIE 4.0x4 card per PCIE 6x bus lane. So I suppose that removes limits for prosumers.

And now the ability to deliver 600w of PCIE power over the bridge.

I fully expect to see 'consumer' cards with PCIE 4.0 or MAYBE 5.0 slots on the boards, and PROsumer with PCIE 5.0 if not 6.0 slots on the board. Parts for the prosumer will approach Enterprise grade hardware in costs because that is what it will really need to be.

Then about 4 years later these will make it down to the Consumer grade hardware as the new hotness.

That's just my opinion though. I know many of us here are prosumers with threadripper or Xenon based systems. Not me... I'm in the high end consumer market because I'm not doing anything scientific or pro design with my system.

Would love to see a discussion about the future of computers and what we think it will mean going forward. I mean... I KNOW we will be wrong. But hey maybe we will be right too about a few things!

 

[Brian_B]

Perhaps I mis read but this is a new power connector for additional power to be delivered over the pcie bus for demanding cards like video cards.

6+2 is commonly called a PCIe connector because it augments the power available to a PCI card, not because it gets power from the bus. I am assuming that is also what this article is referring to.

 

[Grimlakin]

6+2 is commonly called a PCIe connector because it augments the power available to a PCI card, not because it gets power from the bus. I am assuming that is also what this article is referring to.

You're probably right here. Just what I read and I guessed hoped was that motherboards would be able to handle 600 watt power delivery in the future. And that would make for some clean systems. Side mount USB headers, Side mount or coverable front panel control headers. No additional power going over the board even to the video card. Super clean with NVME drives.

See I think that would be kinda awesome but I don't think Motherboard vendors want that kind of power going through their PCB's.

 

[Brian_B]

You're probably right here. Just what I read and I guessed hoped was that motherboards would be able to handle 600 watt power delivery in the future. And that would make for some clean systems. Side mount USB headers, Side mount or coverable front panel control headers. No additional power going over the board even to the video card. Super clean with NVME drives.

See I think that would be kinda awesome but I don't think Motherboard vendors want that kind of power going through their PCB's.

Hmm.. Intel CPUs are already going at nearly 300W+ stock, before we look at something like Threadripper, those are where all those VRM phases are aimed at. Throw in overclocking at that balloons rapidly.

RAM isn't huge - 10W on the upper end, maybe a bit more if your overclocking.

PCI slots are 75W each. I don't know that most motherboards are really designed to actually accomodate full power delivery to all PCI slots simultaneously though.

USB isn't much, but there's usually a high power port or two on a motherboard, and those are upwards of 10W each. And the rest of all the auxillaries - NICs, WiFi, audio, SATA, chipset controller, etc... a few more watts here and there.

I think high motherboards are already handling 600W+, across the entire board. Mid-tier motherboards might stretch and hit 600W.

But 600W to just a single PCIe port -- I suppose it could be possible with the right engineering - the CPU socket is already doing it in high end cases, but it takes a pretty beefy VRM setup right next to the socket to get there... although for GPUs, the VRMs usually live on the video card, so possible. Especially if they stepped up the voltage past 12V, which is how USB is getting higher power draws in those high power USB C ports.

 

[serpretetsky]

But 600W to just a single PCIe port -- I suppose it could be possible with the right engineering - the CPU socket is already doing it in high end cases, but it takes a pretty beefy VRM setup right next to the socket to get there... although for GPUs, the VRMs usually live on the video card, so possible. Especially if they stepped up the voltage past 12V, which is how USB is getting higher power draws in those high power USB C ports.

Yeah I think in my mind it just doesn't make sense. Generally you deliver the power to the component that needs it as directly as possible. Routing the power from the PSU through the motherboard for the GPU just doesn't make sense. It's an extra power connector. It's another power plane or a thicker power plane (and ground). It's another point of ohmic losses. It's another cost. Just plug it straight into the GPU instead.