Geforce RTX 4000: No PCI Express 5.0?


from Andrew Link
The news is currently making the rounds that the Geforce RTX 4000 does without PCI Express 5.0.

Nvidia should not switch to PCI Express 5.0 with the Geforce RTX 4000. At least that’s what it says”kopite7kimi” on Twitter, which is consistently associated with leaks related to graphics cards (and other) and is seen as a usually reliable source. The reasons can only be speculated on at this point.

PCI-Express 4.0 offers a gross bandwidth of 32 GB/s with 16 lanes and a meanwhile somewhat older test from PCGH was already pointing at PCI Express 3.0 to 4.0that the benefit once played hardly any role. It could be similar with this generational leap, where Ada can’t capitalize on a significant change as long as 16 lanes are used. In addition, Ada, like Ampere, will probably support Resizable-BAR again. It is difficult to say at the moment whether this will be a disadvantage for Microsoft Direct Storage in the long term. One does not rule out that Ada basically masters PCI Express 5.0, since Hopper also does this but does not use it, because that also saves some power consumption with graphics cards, which one assumes will not be completely economical. Meanwhile, ATX 3.0 should fully support the Geforce RTX 300 in order to secure the power consumption. In addition, the implementation of PCI Express 5.0 is more expensive and does not make economic sense if you do not benefit from it.

It will be very interesting to see whether AMD goes the same way. That would at least play into the hands of their own platform. The Ryzen 5000, which uses PCI Express 4.0 and is therefore a bit more cost-efficient as a platform, will still be sold until at least the end of the year.

It is also very interesting that the 144 shader clusters at Ada will probably result in the already suspected 18,432 shaders. That would then be the same formula as for amperes. That was accepted, but given the news that Nvidia was working on the caches, there was of course also the option of changing the formula for the clusters. This does not appear to be the case. On paper, the fastest Geforce RTX 4000 would be roughly twice as powerful as the current generation; However, the first models discussed in the rumor mill suggest that there is a big gap between the top model and one below it and that the second fastest card operates at about the same level as the current top model.

Source: Twitter (1, 2)


source site