RTX 40 series may only feature PCIe Gen 4 connector
Will the new cards not max out the current connector?

It has been rumored that the next best GPU generation from Nvidia may not be as cutting edge as it could be. Prolific leaker kopite7kimi has suggested the RTX 40 series will still feature a PCIe Gen 4 interface. This would mean there wouldn’t be an upgrade from the current generation in the bandwidth allowed. Even with the new PCIe Gen 5 16 pin connector that can provide graphics cards with 600W on each connector.
At the current time, only Intels Alder Lake supports PCIe 5.0, but AMD is expected to bring the same feature to their upcoming release later in the year. Upgrading from PCIe 4.0 to 5.0 would double the available throughput for the graphics card. It would jump from 31.508GB/s to 63.015GB/s, alleviating any worries that the connection may be a bottleneck for the increased capabilities of the new cards.
It’s especially interesting if true as only recently in March did Nvidia release the Hopper GPU that features an improved connector. And it’s not uncommon for server-grade cards to provide information on what we may get on the consumer-grade. If AMD includes the new connector in its Radeon lineup they can get an advantage if they can utilize it fully.
RTX 40 series specifications
With a potential piece of the puzzle, we can put it together with the rest of the information we currently have. Along with an improvement in the node process, there is potential for MCM packages, although more likely to come from AMD initially. Along with that, we could see much faster factory clock rates and speeds in memory on the card.
However, this is still early on and not everything is set in stone, especially with no acknowledgment from the manufacturer itself. Another message from the leaker is the single-precision (FP32) core config may not be the same as Amperes. So the CUDA core count may not change but a new type altogether could be introduced.
RTX 4090 | RX 7900 | |
---|---|---|
Node process | TSMC N5 | TSMC N5/6 |
Architecture | Nvidia Ada/Lovelace | AMD RDNA3 |
Package | Monolithic | Multi-chip module (MCM) |
Estimated size | ~600mm² | ~800mm² |
Mega clusters | 12 graphics processors | 2×3 shader engines |
Super clusters | 72 texture processors | 2×30 RDNA groups |
Clusters | 144 streaming multiprocessors | – |
FP32 cores | 18,432 CUDA cores | 15,360 stream processors |
Clock speed | ~2.2 GHz | ~2.5 GHz |
Memory type | 24GB GDDR6X | GDDR6 ? |
Memory clock & bus | 21Gbps 384-bit | 256-bit ? |
Interface | PCIE Gen4 | ? |
Power | 600W | 500W |
WePC is reader-supported. When you buy through links on our site, we may earn an affiliate commission. Learn more