NVIDIA GeForce RTX 4080 – we got to know the details of the AD103 and AD104 cores, which were used in the Ada Lovelace systems

-

NVIDIA GeForce RTX 4080 - we got to know the details of the AD103 and AD104 cores, which were used in the Ada Lovelace systemsDuring the GeForce Beyond conference, NVIDIA focused on the presentation of the top GeForce RTX 4090 graphics card. The flagship of the Ada Lovelace family was equipped with the Ada AD102 graphics chip, offering e.g. over 76 billion transistors. However, no specific details were disclosed about the AD103 and AD104 chips, which power the GeForce RTX 4080 16 GB and GeForce RTX 4080 12 GB respectively. Now we know the missing information.

We learned the details of the construction of the NVIDIA AD103 and AD104 cores, which were used in the GeForce RTX 4080 16 GB and 12 GB cards. It turns out that the AD104 chip offers more transistors than the top GA102.

NVIDIA GeForce RTX 4080 - we got to know the details of the AD103 and AD104 cores, which were used in the Ada Lovelace systems [1]

NVIDIA GeForce RTX 4000 – Card Characteristics, DLSS 3 Tech Details, and Pricing Now and in the Future

The last details about the construction of NVIDIA Ada AD103 and AD104 graphics systems have already appeared on the network. So far we have focused on the top AD102, but we already know everything about it. As for the AD103 and AD104, the new information confirms that both graphics chips have a noticeably smaller surface compared to GA103 and GA104, respectively. The AD103 has an area of ​​378.6 mm² (so smaller than GA104 – 392 mm²), while the AD104 is a relatively small chip with an area of ​​294.5 mm² (therefore slightly larger than GA106 – 276 mm² and much smaller than the aforementioned GA104). This is due to TSMC “4N lithography (by the way – it is a designation for the de facto 5nm process, which has been improved in terms of NVIDIA graphics chips). Despite the reduction in the size of individual chips, the number of transistors has been significantly increased. For example, AD104 has more of them ( !) than the top Ampere GA102 – the difference is 35.8 billion (AD104) vs 28.3 billion (GA102).

NVIDIA GeForce RTX 4090 NVIDIA GeForce RTX 4080 16 GB NVIDIA GeForce RTX 4080 12 GB
Architecture Ada Ada Ada
Graphic layout AD102-300 AD103-300 AD104-400
Lithography TSMC “4N” TSMC “4N” TSMC “4N”
The number of transistors 76.3 billion 45.9 billion 35.8 billion
Surface 608 mm² 378.6 mm² 294.5 mm²
SM blocks 128 76 60
SP units 16384 9728 7680
FP32 power 82.6 TFLOPS 49 TFLOPS 40 TFLOPS
RT cores 128 RT 3rd generation 76 RT 3rd generation 60 RT 3rd generation
Tensor Cores 512 4th generation Tensor 304 4th Generation Tensor 240 4th Generation Tensor
TMUs 512 304 240
ROP units 192 112 80
L2 cache 96 MB 64 MB 48 MB
Base clock 2230 MHz 2210 MHz 2310 MHz
GPU Boost clock 2520 MHz 2505 MHz 2610 MHz
VRAM bandwidth 21 Gbps 22.5 Gbps 21 Gbps
VRAM memory 24 GB GDDR6X 16 GB GDDR6X 12 GB GDDR6X
Memory bus 384-bit 256-bit 192-bit
Capacity 1008 GB / s 720 GB / s 504 GB / s
PCIe Compliant PCIe 4.0 x16 PCIe 4.0 x16 PCIe 4.0 x16
TDP (base) 450 W. 320 W. 285 W.
TDP (Max) 660 Watts 516 W. 366 W.
DLSS Support 3 Yes Yes Yes
Price in PLN PLN 9,399 PLN 7049 PLN 5299

NVIDIA GeForce RTX 4080 12 GB it should be GeForce RTX 4070. It looks bad and bodes badly for next generations

The full list of individual graphics systems is presented in the table below – we compared the parameters of AD102, AD103 and AD104 to GA102, GA103 and GA104, respectively. Meanwhile, the last information concerns the possible high potential of the NVIDIA GeForce RTX 4090 graphics card for overclocking. In the official specification, the manufacturer states that the graphics core clock is 2520 MHz. Meanwhile, on the default settings, without any OC yet, the card in Cyberpunk 2077 was clocked at the level of 2800 to 2850 MHz (last screen in the news). It seems that the 3 GHz breakthrough on the core in the case of the top Ada Lovelace chips should not be a big problem.

NVIDIA AD102 NVIDIA GA102 NVIDIA AD103 NVIDIA GA103 NVIDIA AD104 NVIDIA GA104
Architecture Ada Lovelace Ampere Ada Lovelace Ampere Ada Lovelace Ampere
Surface 608 mm² 628 mm² 378.6 mm² 496 mm² 294.5 mm² 392 mm²
Lithography TSMC “4N ‘ Samsung 8 nm TSMC “4N” Samsung 8 nm TSMC “4N” Samsung 8 nm
The number of transistors 76.3 billion 28.3 billion 45.9 billion ~ 22-23 billion (?) 35.8 billion 17.4 billion
The packing density of the transistors 125.5 MTr / mm² 45.1 MTr / mm² 121.1 MTr / mm² ~ 44 MTr / mm² (?) 121.1 MTr / mm² 44.39 MTr / mm²
SP units 18432 CUDA FP32 10752 CUDA FP32 10240 CUDA FP32 7680 CUDA FP32 7680 CUDA FP32 6144 CUDA FP32
RT units 144 RT 3rd generation 84 RT 2nd generation 80 RT 3rd generation 60 RT 2nd generation 60 RT 3rd generation 48 RT 2nd generation
Tensor Units 576 4th Generation Tensor 336 3rd Generation Tensor 320 4th Generation Tensor 240 3rd Generation Tensor 240 4th Generation Tensor 192 3rd Generation Tensor
Memory Amount 24 GB GDDR6X (RTX 4090) 24 GB GDDR6X (RTX 3090 Ti) 16 GB GDDR6X (RTX 4080 16 GB) 16 GB GDDR6 (RTX 3080 Ti Mobile) 12 GB GDDR6X (RTX 4080 12 GB) 16 GB GDDR6 (RTX 3080 Mobile)
Memory bus 384-bit 384-bit 256-bit 256-bit (RTX 3080 Ti Mobile)
320-bit (general)
192-bit 256-bit

NVIDIA GeForce RTX 4080 - we got to know the details of the AD103 and AD104 cores, which were used in the Ada Lovelace systems [2]

-

NVIDIA GeForce RTX 4080 - we got to know the details of the AD103 and AD104 cores, which were used in the Ada Lovelace systems [3]

Source: NVIDIA, VideoCardz


The article is in Polish

Tags: NVIDIA GeForce RTX details AD103 AD104 cores Ada Lovelace systems

-

NEXT Steam Deck with the list of the most played productions in September. The list includes Cyberpunk 2077
-

-

-

-