The GeForce RTX 4090 will be (at the time of the release of the first wave of products built on the architecture Lovelace) the only desktop card using the largest AD102 GPU. The GeForce RTX 4080 is supposed to run on the AD103 core this time. The GeForce RTX 4090 is expected to have 16,384 active stream processors, which is 56% more than the 10,496 for the GeForce RTX 3090. To do this, the clock frequencies should rise from ~1.7 to ~2.5 GHz, which is a 47% shift.
In total, this would correspond to 2.3 higher arithmetic performance, but the question remains how other parameters will shift, such as the number and performance of ROP units or the effective memory bandwidth, when the bus width remains, memory clocks do not change significantly, but the capacity of the L2 cache increases significantly (which is in the case Lovelace LLC).
The first performance data was provided by leaker kopite7kimi and refers to 3DMark Time Spy Extreme, in which the card should reach “>19000” points. Despite the classic mathematical interpretation, this expression usually means a performance in the range of 19-20 thousand points, at which the GeForce RTX 4090 would achieve an 82-92% shift compared to the GeForce RTX 3090 in this test. This is therefore a performance close to double.
In other applications, the shift may be different, after all, it has long been impossible to say that Nvidia is a company optimizing the performance of its cards for 3DMark – there are currently others. The GeForce RTX 4090 is expected to be released in early autumn this year, meaning either at the end of September or during October. According to some sources, the lower models will be released with a more or less significant gap and the novelty of the family Lovelace so she will be on the market alone for a while.
–
