Although the supply and price of RTX 30 series graphics cards seem to be a little normal recently, it is no surprise that NVIDIA should launch a new RTX 40 series codenamed Ada Lovelace at the end of this year. According to Kopite7kimi, the top-level complete RTX 40 series The version AD102 chip product is rumored to be equipped with 48GB GDDR6X memory, and the TGP power consumption will reach 900W, which is much higher than the previously rumored RTX 4090; but rumors also indicate that some products may be NVIDIA’s knife test products, not necessarily the last will become the commercial version.
**90 PG137/139-SKU330, AD102-300, 24G 21Gbps, 600W
– kopite7kimi (@ kopite7kimi) April 11, 2022
At present, it is believed that NVIDIA will announce RTX 4090, RTX 4080 and RTX 4070 in the early stage. RTX 4090 will be equipped with 24GB GDDR6X and 600W TGP. As for RTX 4080, it will use AD103 chip, equipped with 12GB GDDR6X memory and about 450W TGP. The entry-level version is also played by the 70 series RTX 4070, with AD104 chip and 12GB GDDR6, TGP is 300W.
In fact, there is another full-fat AD102 SKU with 900W TGP, 48G 24Gbps GDDR6X, 2*16pin and higher frequency.
But no one knows whether it will become an actual product. Because the test board of AD102 has more than two 16pin connectors, so everything is possible.– kopite7kimi (@ kopite7kimi) April 27, 2022
It is rumored that NVIDIA is currently testing the highest-end AD102 chip, which is also the top chip that will be used in the RTX4090 series; after NVIDIA released the RTX 3090, NVIDIA also readjusted its product line and regarded the RTX x090 series positioned on BFGPU as the previous Titan’s Products of the same level, but do not have the workstation functions such as ECC and virtualization of the previous Titan, and are more clearly positioned at the consumer level rather than cross-border products, but also said that customers who need advanced functions of these workstations can only choose the official acceleration card product.
The unconfirmed product in this outgoing test is the RTX 4090 Ti, which is said to use the same AD102 chip as the RTX 4090, but with a more complete core and a faster and more 24Gbps than the RTX 4090 The 48GB GDDR6X memory, etc., TGP will reach 900W in one fell swoop, and it will also be equipped with two 16Pin slots, but it has not yet been determined whether this product is already in the product planning, and it may also be the final product at the end of the product like the current RTX 3090 Ti. product.
▲The first wave of Ada Lovelace is said to also provide three product lines including RTX 4090, RTX 4080 and RTX 4070
Aside from the uncertain RTX 4090 Ti, the top product RTX 4090, which is almost certain to be unveiled in the first wave of Ada Lovelace products, will use the AD102 chip, but the design concept will be a more streamlined basic architecture than the current Ampere architecture, thereby accommodating more More SM and CUDA cores, rumored to have up to 144 SMs and 18,432 CUDA Cores, much more than the 10,752 CUDA Cores of the RTX 3090 Ti; in addition, the cache capacity is also the focus of the Ada Lovelace architecture, AD102 will be equipped with 96MB The L2 of the RTX 3090 series has increased a lot compared to the 6MB of the GA102 of the current RTX 3090 series. Although the performance of the RTX 4090 is expected to increase dramatically compared to the RTX 3090, the TGP of up to 600W means that both power consumption and heat generation are even more amazing.
Relatively afraid of the unattainable RTX 4090, RTX 4080 and RTX 4070 should be more interesting products for many players; RTX 4080 will use AD103 core, and chips below this level will also be used for mobile versions; AD103 The number of SMs will be Maintaining the same 84 as GA102, the desktop version of TGP will also be maintained at around 350W to 450W, but the cache will be increased to 64GB, and with 256bit 16GB GDDR6X memory.
As the first wave of the most affordable RTX 4070, the AD104 chip will be used, equipped with 60 SMs and 7,680 CUDA Cores, the L2 cache will also have 48MB, the memory will be equipped with 192bit 12GB GDDR6, and the TGP will be set around 300W.