NVIDIA RTX 4090, RTX 4080, RTX 4070 Final Specs Leak Out: Up to 16,384 Cores, 24GB GDDR6X VRAM, 2.75GHz Boost Clock, and 450W TGP [Report]

Areej Syed

July 6, 2022

The almost final specifications of NVIDIA’s next-gen RTX 40 series graphics cards have surfaced. Shared by our usual tipster kimi, we now know pretty much everything about the RTX 4070, RTX 4080, and the RTX 4090. According to the source, the GeForce RTX 4090 will leverage the AD102-300-A1 die with 16,384 enabled cores and 24GB of GDDR6X memory running at 21Gbps across a 384-bit bus. The GPU core will have a base clock of 2,235MHz and a boost of 2,520MHz with an effective in-game clock of 2,750MHz.

https://twitter.com/greymon55/status/1543866846385823744

The GeForce RTX 4090 should be priced $1,999, same as the RTX 3090 Ti but offer close to 2x the performance in gaming workloads. We’re talking about a single precision throughput of roughly 90 TFLOPs at a TGP of 450W. Keep in mind that the fully-enabled AD102 die packs a whopping 18,432 FP32 cores. The RTX 4090 Ti will get the honor to showcase it and offer up to 100 TFLOPs of single precision performance. On the downside, the TGP is also likely to peak at 600W.

GPU	GA102	AD102	RTX 4090	AD103	RTX 4080	RTX 4070 Ti (AD104)	RTX 4070
Arch	Ampere	Ada Lovelace		Ada Lovelace		Ada Lovelace
Process	Sam 8nm LPP	TSMC 5nm		TSMC 5nm		TSMC 5nm
GPC	7	12	11	7	7	5	5
TPC	42	72	64	42	40	30	30
SMs	84	144	128	84	80	60	60
Shaders	10,752	18,432	16,384	10,752	9,728	7,680	7,680
TP	37.6	~100 TFLOPs?	83 TFLOPs	~50 TFLOPs	47 TFLOPs?	~35 TFLOPs	35 TFLOPs?
Memory	24GB GDDR6X	48GB GDDR6X	24GB GDDR6X		16GB GDDR6X		12GB GDDR6X
L2 Cache	6MB	96MB	72MB	64MB		48MB
Bus Width	384-bit	384-bit		256-bit		192-bit
TGP	350W	600W	450W	450W	285-340W	300W	285W
Launch	Sep 2020		Sept 22?		Sept 22?		Q1 2023?

Moving to the more “affordable” RTX 4080, the smaller AD103-300-A1 die comes into the picture. It’ll feature 10,240 FP32 cores alongside a 16GB GDDR6X VRAM buffer clocked at 21Gbps via a 256-bit bus. The GPU core will be clocked at around 2.5GHz (base) with the boost possibly even exceeding 3GHz on AIB cards. The RTX 4080 will have a respectable TGP of 420W, all the while beating the RTX 3090 by at least 50%. It should be priced close to $999. The full-fledged AD103 die is unlikely to be featured in any SKU.

Finally, moving to the RTX 4070, we are looking at the even smaller AD104 die. It’ll feature 7,168 shaders paired with 10GB of GDDR6 memory clocked at 18Gbps across a 160-bit bus. The base clock should be similar to the RTX 4080, hovering in the 2.5-3.0GHz mark. It should have a TGP of 300W. We may or may not see an RTX 4070 Ti based on the AD104 die as the fully enabled core won’t be much faster than the 4070. A wider bus width and faster memory may do the trick though. The RTX 4070 will probably cost the same as the RTX 3070 at $499.

The RTX 4090, 4080 and 4070 can feature up to 96MB, 64MB, and 48MB of L2 cache, respectively. However, it’s likely that NVIDIA will trim chunks of the SRAM along with the disabled GPCs. These 4nm GPUs are slated for a September/October launch, starting with the RTX 4080 and 4090. The mythical 4090 Ti (with 600W TGP) won’t be launching this year, and like its predecessor will make a late cycle appearance.