WePC is reader-supported. Once you purchase by means of hyperlinks on our website, we could earn an affiliate fee. Be taught extra
Final Up to date:
It has been rumored that the following greatest GPU technology from Nvidia is probably not as innovative because it could possibly be. Prolific leaker kopite7kimi has urged the RTX 40 collection will nonetheless function a PCIe Gen 4 interface. This could imply there wouldn’t be an improve from the present technology within the bandwidth allowed. Even with the brand new PCIe Gen 5 16 pin connector that may present graphics playing cards with 600W on every connector.
On the present time, solely Intels Alder Lake helps PCIe 5.0, however AMD is anticipated to carry the identical function to their upcoming launch later within the 12 months. Upgrading from PCIe 4.0 to five.0 would double the out there throughput for the graphics card. It might bounce from 31.508GB/s to 63.015GB/s, assuaging any worries that the connection could also be a bottleneck for the elevated capabilities of the brand new playing cards.
It’s particularly fascinating if true as solely lately in March did Nvidia launch the Hopper GPU that options an improved connector. And it’s not unusual for server-grade playing cards to supply info on what we could get on the consumer-grade. If AMD consists of the brand new connector in its Radeon lineup they’ll get a bonus if they’ll put it to use absolutely.
RTX 40 collection specs
With a possible piece of the puzzle, we are able to put it along with the remainder of the data we at present have. Together with an enchancment within the node course of, there’s potential for MCM packages, though extra more likely to come from AMD initially. Together with that, we may see a lot quicker manufacturing facility clock charges and speeds in reminiscence on the cardboard.
Nevertheless, that is nonetheless early on and never every part is ready in stone, particularly with no acknowledgment from the producer itself. One other message from the leaker is the single-precision (FP32) core config is probably not the identical as Amperes. So the CUDA core depend could not change however a brand new kind altogether could possibly be launched.
|RTX 4090||RX 7900|
|Node course of||TSMC N5||TSMC N5/6|
|Structure||Nvidia Ada/Lovelace||AMD RDNA3|
|Bundle||Monolithic||Multi-chip module (MCM)|
|Mega clusters||12 graphics processors||2×3 shader engines|
|Tremendous clusters||72 texture processors||2×30 RDNA teams|
|Clusters||144 streaming multiprocessors||–|
|FP32 cores||18,432 CUDA cores||15,360 stream processors|
|Clock pace||~2.2 GHz||~2.5 GHz|
|Reminiscence kind||24GB GDDR6X||GDDR6 ?|
|Reminiscence clock & bus||21Gbps 384-bit||256-bit ?|