Nvidia’s waiting RTX 4090 video card can be clocked above 2,750 MHz, The new, expected specs once again come from @kopite7kimi, which is almost single-handedly keeping the rumor mill about Nvidia’s Ada Lovelace generation (RTX 40) running. A new tweet shares more details about the GeForce RTX 4090, along with updated specifications for both the RTX 4080 and RTX 4070.
In addition to some corrections, the clock speed of the alleged RTX 4090 is particularly striking. So far, little or nothing has been released about the effective frequencies of the GPUs, but now those of the RTX 4090 seems to be curing. A base speed of 2,235 MHz is expected, with a boost of up to and including 2,520 MHz. In reality, the GPU could be clocked to at least 2,750 MHz.
Clock speeds themselves don’t tell much about the actual performance, but the frequencies mentioned are significantly higher than Nvidia’s previous offering. Compared to the RTX 3090, the base clock would be no less than 60 percent higher; the boost speed improves by approximately 49 percent.
the fuss around elevated wattages seems to have calmed down: the RTX 4090 probably has a TGP of 450 watts no longer the almost disturbingly high 600 watts or more.
The RTX 4080 also seems to be hardening further. For example, 16 GB of GDDR6X video memory was again mentioned, apparently now also about the 21 Gbps modules that the RTX 3090 Ti introduced. There was no talk about clock speeds, but this time the TGP would be at a slightly higher 420 watts (compared to the RTX 3080 with 320 watts).
Nvidia is still expected to kick off the Ada Lovelace range at the end of 2022, with most likely the above RTX 4090, RTX 4080, and RTX 4070 as the first introduction. With Ada Lovelace, Nvidia omits Samsung’s 8 nm process and most likely switches to TSMC’s so-called 4N wafers (5 nm). However, Nvidia would like to reduce the previously reserved volumes.
For the time being, little concrete has been shown or leaked about the performance ratio between Ada Lovelace and the current Ampere. All signals refer to a higher consumption for Nvidia GPUs, but to what extent the performance drive improves remains guesswork. For that, we still have to wait for an official unveiling, which will most likely only follow in a few months.
Nvidia will likely announce the RTX 4000 series in September or October of this year. The RTX 4060 is expected to be announced at CES 2023.
Model | GeForce RTX 4090 | GeForce RTX 4080 | GeForce RTX 4070 |
---|---|---|---|
Architecture | Ada (TSMC N4) | Ada (TSMC N4) | Ada (TSMC N4) |
GPU | AD102-300 | AD103-300 | AD104-275 |
SMs | 128 | 80 | 56 |
CUDA Cores | 16,384 | 10,24 | 7,168 |
Base clock | 2.235 MHz | nnb | nnb |
Boost clock | 2.520 MHz | nnb | nnb |
Memory speed | 24GB GDDR6X | 18 Gbps | 18 Gbps |
Memory bus | 384-bit | 256-bit | 160-bit |
Bandwidth | 1,008 GB/s | 576 GB/s | 360 GB/s |
Tbsp | 450W | 420W | 300W |
the specifications of the GPUs used of the RTX 4080 and RTX 4070. Both use a different GPU, the AD103, and AD104 respectively. It looks like the RTX 4080 has a 256-bit memory bus and 16GB of gddr6x VRAM, clocked at the same speed as the RTX 4090. The RTX 4070, on the other hand, has a 160-bit bus and 10GB gddr6 VRAM.