Ada Lovelace is Nvidia’s next consumer GPU architecture. After the hack, in which large amounts of data were released from Nvidia, SemiAnalysis in collaboration with the analyst locuza expected dimensions and layout of the moulds.

It is noteworthy that there is a lot of L2 cache and the patterns are slightly smaller. Parts can be made smaller as this generation is manufactured on TSMC’s N4 5nm node. This makes it possible to place more cache in less space. The top model, named AD102 according to its current name, measures 611.3 square millimeters. The GA102 used in the RTX 3090 is 628mm.2nd

The amount of L2 cache has been increased from 64 to 96 megabytes, coincidentally the same as the L3 Infinity Cache on AMD’s Navi 22 GPU. This should compensate for the relatively small increase in memory. A maximum of 24 gigabytes of gddr6x will be used at 21 Gbps over the same 384-bit bus. With 5 additional GPU processing clusters, the number of cuda cores has been increased by 70 percent.

At 379.69 square millimeters, the AD103 is a big step smaller and has as many cuda cores as the GA102. It has 64MB of cache and 16GB of ram on a 256-bit bus. In addition, the improvement in clock frequencies due to the newer process, which should surpass the 3090 Ti in performance. Like the GA103, this variant can only be used on laptops using lower power limits.

The AD104 is slightly smaller with a surface area of ​​300.45 mm.2nd† There is a 192-bit memory bus to which 12 gigabytes of g6x memory can be connected. By significantly increasing the clock speed, the performance level of the 3090 at which 350 or 400 watts can be burned can be achieved. With an adjustable TDP of 90 to 135 W and less luxury memory configuration, this GPU can also be used in laptops.

The AD106 is expected to deliver the highest volumes just like the previous generation Pascal, Turing and Ampere. It has 32 MB of cache and 8 GB of memory at 128 bit. The smallest version is the AD107, with an estimated area of ​​145.54 square millimeters. The number of PCIe lanes has been reduced to 8 and performance should far exceed that of integrated GPUs.

The biggest advance is clearly in the top models, with lower-position variants that carry the performance of the previous generation into a cheaper class. In order to squeeze as much gigahertz and performance from each chip as possible, power consumption is expected to be quite high. Lovelace is expected to launch in the third quarter of this year and will likely only be used for consumer graphics cards.

Source: Hardware Info

Previous articleOnePlus Ace: design anticipated by a leaked video?
Next articleChina lockdown forces Apple to delay shipping of MacBooks

LEAVE A REPLY

Please enter your comment!
Please enter your name here