NVIDIA GeForce RTX 4070 variant to feature AD103 GPU

NVIDIA AD103 GPU for RTX 4070

NVIDIA might be preparing to use salvage dies for (currently) the lowest tier SKU in RTX 40 lineup.

Thus far, we have not heard about any Ada GPUs being repurposed for other SKUs, but is about to change. According to NVIDIA hardware leaker @kopite7kimi, the company has now set plans to introduce AD103 GPU for RTX 4070.

As we know, the full AD103 GPU is currently not used by any graphics card on the market, but the RTX 4080 is the closest we got with 9728 CUDA Cores and just 5% of the GPU disabled. The rumored RTX 4070 with AD103 GPU would be a different story. This SKU has 5888 CUDA cores, so should NVIDIA use the bigger GPU instead, this would mean 42% of the GPU partially disabled.

Board partners should not encounter issues with the GPU swap, as AD103 and AD104 are pin compatible. The AD103 GPU will be heavily cut and will use lower TDP, so they can easily adopt the same cooling designs. Questions that may arise would be related to L3 cache size, ROP count and encoding capabilities, as those features are usually affected by GPU swapping.

One should note however that GeForce desktop cards are no stranger to mixed GPU specs. Those are commonly sold on the Chinese market, and we have seen plethora of RTX 3070 and 3060 designs with non-standard GPUs. This may explain why NVIDIA is not displaying the exact GPU names on their official website, but only provides this information to reviewers. However, many board partners who use mix GPUs often create separate variants, just to make sure customers know what card they are buying.

NVIDIA GeForce RTX 4070 vs. 3070 Series
VideoCardz.com	RTX 4070 Ti	RTX 4070 “AD103”	RTX 4070	RTX 3070 Ti	RTX 3070
Picture
Board-SKU	PG141-SKU331	TBC	PG141-SKU344	PG143-SKU15	PG142-SKU10
Architecture	Ada (TSMC 4N)	Ada (TSMC 4N)	Ada (TSMC 4N)	Ampere (SAMSUNG 8N)	Ampere (SAMSUNG 8N)
GPU	AD104-400	AD103-???	AD104-250	GA104-400	GA104-30X
L2 Cache	48 MB	TBC	36 MB	4 MB	4 MB
CUDA Cores	7680	5888	5888	6144	5888
RT Cores	60	46	46	48	46
Base Clock	2310 MHz	TBC	1920 MHz	1575 MHz	1500 MHz
Boost Clock	2610 MHz	TBC	2475 MHz	1770 MHz	1725 MHz
Max FP32 Compute	40 TFLOPS	TBC	29 TFLOPS	22 TFLOPS	20 TFLOPS
Memory	12 GB G6X	12 GB G6X	12 GB G6X	8 GB G6X	8 GB G6
Memory Bus	192-bit	192-bit	192-bit	256-bit	256-bit
Memory Speed	21 Gbps	21 Gbps	21 Gbps	19 Gbps	14 Gbps
Memory Bandwidth	504 GB/s	504 GB/s	504 GB/s	608 GB/s	448 GB/s
TDP	285W	TBC	200W	290W	220W
Interface	PCIe Gen4 x16	PCIe Gen4 x16	PCIe Gen4 x16	PCIe Gen4 x16	PCIe Gen4 x16
MSRP	799 USD	599 USD	599 USD	599 USD	499 USD
Release Date	January 5, 2023	TBC	April 13, 2023	June 10, 2021	May 8, 2021

Source: @kopite7kimi