The NVIDIA Flagship “Ada Lovelace” GPU is rumored to have 18,176 cores, 48GB of memory, 24Gbps speeds, and 800W TBP

Details about the flagship NVIDIA Ada Lovelace GPU SKU have been reported by Kopite7kimi Which makes it look like the ultimate graphics chip ever made.

The NVIDIA Ada Lovelace Flagship GPU is rumored to feature an AD102-450 GPU with 18,176 cores, 48GB, 24Gbps, GDDR6X, 800W TB of memory.

This isn’t the first time that high-end Ada Lovelace GPU storage has been talked about. The previous rumor, which also came from the same leaker, was reported on a Titan-Class graphics card within the Ada Lovelace GPU lineup that will feature some crazy specs. Again, this is not a fully enabled AD102 GPU while the previous variant was mentioned as a 900W TBP SKU with full 18432 CUDA Cores.

According to rumored specs, the graphics card will rock the Ada Lovelace architecture and feature a slightly reductive “AD102-450-A1” configuration, swinging 142 SMs (out of 144 SMs) on 18,432 CUDA cores (out of 18432 CUDA cores). Based on a clock speed of around 3GHz, this graphics card will easily cross the 100 TFLOPs calculating barrier. The graphics card is said to be equipped with 48GB of GDDR6X memory that runs via a 384-bit bus interface.

What’s interesting is that NVIDIA won’t back down from the VRAM specification and features the latest 24Gbps memory modules, delivering up to 1.152TB/s of VRAM bandwidth to the GPU. This represents a 14% increase in memory bandwidth over the current RTX 3090 Ti’s 21Gbps main memory. The upcoming RTX 4090 is also expected to use the same 21Gbps memory with the flagship ‘Ti’ model dying only at 24Gbps.

As for power consumption, the new NVIDIA Flagship AD102 GPU-powered graphics card is going to be insane, with almost double the TBP of the RTX 3090 Ti, which is rated up to 800W. Given that a single 16-pin connector can only provide 600 watts of power, a dual 16-pin connector configuration must be used for this card monster if it becomes a reality. You may use the PG137-SKU0 graphics card

Based on the Ampere lineup, we saw that NVIDIA not only released a Titan graphics card, but essentially replaced the Titan series with a BFGPU-class GeForce RTX lineup. The higher-capacity cards were still released as the RTX A**** workstation lineup which also got full GA102 processing but along with the RTX 3090 Ti and RTX A6000, there was no Titan class. So would a Titan Ada Lovelace GPU make sense, or would this definitional storage end up as the next-generation gaming BFGPU and workstation pioneer? Well, we can’t say for sure but one thing is definitely true, the GPU configuration is really going to be insane in terms of specs, power draw, and price. If the card is available for retail, it will definitely launch after the RTX 4090 which is expected to be revealed later this fall.

NVIDIA GeForce RTX 40 Series graphics cards (rumored):

Photographers card GPU Variable PCB SM modules/cores memory / bus Memory clock / bandwidth TGP power connectors release
NVIDIA Titan A / GeForce RTX 40? AD102-450? PG137-SKU0 142/18176? 48 GB / 384 bits 24 Gbps/1.15 TB/sec ~800 watts 2x 16 teeth to be announced later on
NVIDIA GeForce RTX 4090 Ti M 102 – 350? to be announced later on 144/18432? 24 GB / 384-bit 24 Gbps/1.15 TB/sec ~600 watts 1x 16 teeth to be announced later on
NVIDIA GeForce RTX 4090 M 102 – 300? PG137 / 139 SKU330 128/16384? 24 GB / 384-bit 21 Gbps/1.00 TB/sec ~ 450 watts 1x 16 teeth Fourth Quarter 2022
NVIDIA GeForce RTX 4080 AD103-300? PG13 * / 139 SKU360 80/10240? 16 GB / 256 bit 21 Gbps / 672 Gbps ~ 420 W 1x 16 teeth Fourth Quarter 2022
NVIDIA GeForce RTX 4070 AD104-275? PG141-310 SKU341 56/7168? 10 GB / 160 bits 18 Gbps / 360 Gbps ~ 300 watts 1x 16 teeth Fourth Quarter 2022
NVIDIA GeForce RTX 4060 AD106 – ***? to be announced later on >36/4608? 8 GB / 128 bits to be announced later on ~200 watts 1 x 16 teeth First Quarter 2023

Which NVIDIA GeForce RTX 40 graphics card are you most looking forward to?

Products mentioned in this post

Leave a Reply

%d bloggers like this: