While waiting for Jen-Hsun Huang’s speech to formalize the Ampere GPUs on September 1, the official specifications of the RTX 30 now seem to be known.
–
You can feel it gently iridescent the news, the arrival of the generation Ampere de Nvidia, with the promise of invigorating our player configurations with a new light? Introducing in particular the second generation of cores dedicated to ray-tracing, and the third generation of Tensor Cores, mainly dedicated to artificial intelligence and precision calculations, the new GeForce RTX 3090, 3080 and 3070 should officially be unveiled on September 1, through an expected conference by Jen-Hsun Huang.
GeForce RTX 3090, 3080 and 3070: 8 to 24 GB of GDDR6X memory
But Ampere and against everything, the first outlines of these future mastodons are already emerging. According to a source cited by VideoCardz, which seems as relevant as it is reliable, Nvidia is preparing to release three new cards during the month of September: the RTX 3090 24 Go, la RTX 3080 10 Go et la RTX 3070 8 Go. Third-party manufacturers would prepare behind the scenes an RTX 3080 with twice the memory, or 20 GB – a first testament to the great modularity of the architecture – but such models have not yet been confirmed.
Besides the support of PCI Express 4.0 and the presence of connectors HDMI 2.1 et DisplayPort 1.4a, the leaked specifications confirm many aspects already mentioned so far. Starting with the fine engraving of the GPU and 7 nm.
The GeForce RTX 3090 This would revolve around the GA102-300 GPU, with 5248 cores and 24 GB of GDDR6X memory, on a 384-bit bus. It would thus deliver a theoretical memory bandwidth of 936 GB / s. As a reminder, the RTX 2080 Ti of Turing generation embeds 4,352 cores with 11 GB of GDDR6 on a 352-bit bus. This card would feature a TGP of 350 W. It is known that the Founders Edition model would require a 12-pin power connector, as we discussed previously, but the rumor also suggests the availability of custom models around two 8-pin connectors.
The GeForce RTX 3080 For its part, it would carry 4352 CUDA cores (… precisely the quantity we have just mentioned of the RTX 2080 Ti, which was the flagship of the previous Turing generation), with 10 GB of GDDR6X. It would benefit from a bandwidth of 760 GB / s on a 320-bit memory bus. Its GA102-200 GPU would be clocked at 1710 MHz (a priori, the highest frequency of this new range, the GA102-300 of the RTX 3090 seeming to be clocked at 1695 MHz) and the card would present a TGP of 320 W. Here again , we are talking about custom models equipped with two eight-pin connectors.
Finally, the GeForce RTX 3070 revolves around the GA104-300 GPU, which the VideoCardz leak does not guarantee that it will be engraved in 7 nm. The card would have 8 GB of GDDR6 (non-X) memory, on a 256-bit bus and with a bandwidth of 512 GB / s. The TGP of the card would be 220 W, which would mark a clear break compared to the other two gondola heads.
Onboard memory, the other argument of the RTX 30?
If all the light has not been shed on Ampere, these first leaks are not lacking in intensity. Apart from the official prices of the new range which are missing (we are talking about from 1580 to 1700 euros for the RTX 3090), it is clear that one of the sinews of war for this next generation lies in the on-board memory.
Between quantities that had never been reached before in this product range, but also unbridled bandwidths and custom models that would seem capable of carrying the double GDDR6X memory planned for the Founders Edition, everything leads to believe that this will be one of the benchmark showdowns between manufacturers in the coming months. Recent news showing the appetite of a certain Flight Simulator at this level, which swallows up more than 16 GB of 8K VRAM, was it heralding a new trend?
–
