Although the new generation of graphics from Nvidia will not be released until the new year, details about the upcoming models that will be released in the first months of 2025 have already begun to leak. During the week, there was news about the GeForce RTX 5090 as the most powerful model, but especially about the cheaper GeForce RTX 5070 Ti and GeForce RTX 5070, which may be the cards that will interest the most players due to the “high-end” prices.
The website VideoCardz citing a leading leaker and nicknamed Kopite7kimi came up with a report on the planned parameters of the GeForce RTX 5070 Ti on Thursday, which was then supplemented by a separate article from the website BenchLife.
GeForce RTX 5070 Ti
This card will be the third most powerful model of the RTX 5000 generation, and according to their information, it is based on the GB203 chip – allegedly in the GB203-300 variant. So this card will still use the second largest chip as in the second highest model RTX 5080 (it has GB203-400). Both cards use the same reference PCB (higher model is PG147 SKU 45, RTX 5070 Ti model is PG147 SKU 60).
The RTX 5070 Ti will include 8960 shadersi.e. 70 SM (while the fully active GB203 GPU configuration should have 84 SM / 10,752 shaders). Unlike the computing units, however, the memories will not be truncated. The GPU provides a physically 256-bit bus, and the RTX 5070 Ti model will have the full memory width and capacity 16 GBwhile the RTX 4070 Ti only had 12 GB of memory, because it was based on the third smaller AD104 chip with 60 SM (7680 shaders) and the 192-bit bus was the maximum possible with the given silicon.
The GeForce RTX 5070 Ti should use GDDR7 memory, and thanks to this it will achieve a decent increase in throughput (with an effective frequency of 28.0 GHz, it would reach 896 GB/s with a 256-bit bus). In terms of raw performance, the RTX 5070 Ti with 8960 shaders will have a sixth more (16.7%) more compute units than the RTX 4070 Ti.
Other potential performance gains can come from architectural improvements increasing performance to 1 MHz per unit (one could say IPC), but also from frequency increases. Although we don’t yet have information on what the clocks in the specs will be, rumors have suggested that the Blackwell GPUs will reach high numbers.
However, consumption is also expected to increase – the RTX 4070 Ti has a TDP of 285 W, while the new GeForce RTX 5070 Ti model should, according to information from Benchlife, have a consumption of 350 W (by the way – that was the consumption of the highest model RTX 3090 four years ago and it was practically a record at the time; the trend of consumption of graphics cards is not exactly a positive development). It is not entirely clear whether it is the so-called Maximum Power Limit or by the value that Nvidia will state as TDP or more precisely TGP (total consumption of the card). Benchlife writes that it is TGP, which would be the worse (meaning higher real consumption) of the two variants.
GeForce RTX 5070
While Kopite7kimi has not yet revealed the parameters of the VideoCardz website for the GeForce RTX 5070, they have been published by the Benchlife website. This card already has a third smaller GB205 chip in the GB205-300 variant (this is the successor to the AD104 in the previous generation – Nvidia skipped the GB204 designation for some reason). This GPU already only has 192-bit memory, so the RTX 5070 will only carry 12 GB memory like the RTX 4070 model before it. But it could reportedly still be GDDR7, so throughput could be up to 672GB/s at an effective clock of 28.0GHz (or even more if higher).
They will be active on the chip 6400 shaderswhich is 50 SM blocks. GPU frequencies are again not known. The number of units alone represents a potential performance increase of 8.7% (the RTX 4070 only has 5888 shaders / 46 SMs), to which again architectural improvements and higher clocks can add something. According to Benchlife, this model should also have increased consumption, TDP is said to be 250 Wwhile the RTX 4070 model was only at 200 W.
Non-reference GeForce RTX 4070 Ti Super graphics
GeForce RTX 5090: PCIe 5.0 confirmed, very large silicon
GeForce RTX 5090 specs were leaked earlier (along with RTX 5080 specs). The RTX 5090 has 512-bit GDDR7 memory with a capacity of 32 GB and a throughput of 1792 GB/s (28.0 GHz effective frequency) and 21,760 shaders (170 SM blocks out of 192 physically present on the GB202 chip). The TDP is said to be 600W, which will be powered by a single 12V-2×6 contactor cable (modified 12+4pin 12VHPWR).
Information about the size of its GB202 chip (this is the most powerful version of Blackwell’s gaming GPU series) has recently appeared for this model. According to the leaker with the nickname MEGAsizeGPU, it should have an area of 744 mm². For comparison: the AD102 in the GeForce RTX 4090 is only 608 mm² and the GA102 of the Ampere generation had 628 mm². The BGA case measures 6.3 x 5.6 cm. Whether, according to the information so far, it should still be manufactured by the 4nm process.
In addition, another leaker hongxing2020 also showed a photo of the connector that will fit the card into the PCIe x16 slot. The contacts on it have a specific appearance, which seems to be related to the fact that this GPU will already support PCI Express 5.0, so on newer platforms (Ryzen 7000/9000, Intel Core 12th generation and later) it will finally be possible to use a higher throughput of the connection to the processor . The information that the RTX 5090 will be the first GeForce to be able to “Gen 5” appeared earlier, but this image perhaps proves it.
Most cards will be released in the first three months of 2025
According to the Benchlife website, Nvidia will release the GeForce RTX 5090 (and 5090D for China), GeForce RTX 5080 and GeForce RTX 5070 Ti as well as the RTX 5070 in the first quarter of 2025. The GeForce RTX 5060 Ti and GeForce RTX 5060 as lower-end mainstream models do not yet have a date like this concretely confirmed and probably won’t come out in the first third of the year. According to Benchlife, it is likely that they will be released in April instead of the previously teased March.
Source: www.cnews.cz
