We’ve been hearing rumors that NVIDIA’s next-gen flagship based on the Ada Lovelace microarchitecture will offer incredible performance, at a cost though. The power consumption of the RTX 4080 and 4090 will reportedly go as high as 450-500W, with some AIB variants going even further. According to a report from Igor, this will be facilitated by the new 12VHPWR power connector.
As part of the next-gen PCIe Gen 5 standard, the PCI SIG has defined a new “High Power Connector” or H+ for graphics cards with a power requirement of up to 600W. The 12VHPWR connector is a major departure from the existing PCI Express 2×3/2×4 Auxiliary Power connectors. While the pins of the former have a spacing of 3.0 mm, the latter has a larger spacing of 4.2 mm. Therefore, they won’t be compatible with the same hardware and require changes on both the PSU and the GPU PCB.
The 12VHPWR power connector can supply up to 55A of continuous power to the graphics card via its 12V power rail with a maximum power of 600W. The PCI SIG specifies a pin current capability (excluding sideband contacts) of 9.2 A per pin/position with a limit of 30 °C T-rise above ambient temperature at + 12 VDC with all twelve contacts energized. This results in 55.2 amps in one direction for the 12-volt power rail or 662.4 watts.
Taking the tolerances and safety precautions into place, up to 600W of power can be safely supplied to the add-in board (a reduction of 11% over the max possible). The added power consumption may seem like a step back for GPUs, but the new connector will simplify plug and PCB design quite significantly. NVIDIA’s RTX 40 series graphics cards, most notably the RTX 4080 and 4090 should leverage this new power connector along with support for PCIe Gen 5.
Meanwhile, the RTX 3090 Super/Ti should stick to multiple 8-pin molex connectors (three?). It’s unclear whether AMD will make the jump with its RDNA 3 (Radeon RX 7800 XT/7900 XT) graphics cards as well, considering that they’ll adopt an MCM design. Both vendors may retain the older connectors on lower-end cards to avoid unnecessary PSU upgrades for budget users.
Previous coverage on the RTX 4080/4090:
From what we’ve heard, Lovelace should easily offer twice as much performance as the contemporary Ampere parts, with an FP32 core count of up to 18,432. The AD102 flagship is expected to feature a total SM count of 144 distributed across 12 GPCs. That’s a 71% gain in raw compute performance over the GA102. Even if TSMC’s N5 node is 30% more power-efficient than Samsung’s 8nm LPP node, we’re looking at a sheer increase of around 80% in shader count. That should easily result in a power draw at least 30-50% more than the top-end (RTX 3080/3090) Ampere offerings.
NVIDIA’s Hopper Data Center graphics cards are said to be based on an MCM design. According to @kopit7kimi, every GPC (Graphics Processing Cluster) in the GH100 will consist of three CPCs which in turn will feature three TPCs (Texture Processing Cluster), further sub-divided into two SMs. It’s worth noting that we are expecting a separate architecture for the gaming and data center markets, with Ada Lovelace coming to the former as a monolithic design and Hopper with its MCM design headed to the next-gen Tensor core accelerators.