NVIDIA GeForce RTX 3090

The NVIDIA GeForce RTX 3090 is an enthusiast-class graphics card launched on September 1, 2020, as part of the RTX 30 series based on the Ampere architecture. Branded as the "BFGPU" (Big Ferocious GPU), it was positioned as a professional-grade GPU for creators, data scientists, and AI researchers. At $1,499, it became legendary for its 24 GB of GDDR6X VRAM.

Beyond gaming, the RTX 3090 found a second life in AI and machine learning. With 24 GB VRAM and powerful CUDA compute, it became the go-to budget workhorse for running large language models, democratizing access to serious GPU compute for thousands of independent researchers and developers.

Built on the GA102 GPU (Samsung 8 nm 8N process), the die contains 28.3 billion transistors on 628 mm² - one of the largest consumer GPU dies ever produced.

Feature	RTX 3090	RTX 3090 Ti
CUDA Cores	10,496	10,752
Tensor Cores (3rd gen)	328	336
Boost Clock	1,695 MHz	1,860 MHz
Memory	24 GB GDDR6X	24 GB GDDR6X
Memory Bandwidth	936 GB/s	1,008 GB/s
TDP	350 W	450 W

The third-gen Tensor cores added support for sparse computation, TF32, and BF16 precision formats. The 24 GB GDDR6X using PAM4 signaling delivered 19.5 Gbps per pin, essential for 8K content creation and massive AI model inference.

Unveiled on September 1, 2020, at $1,499, the RTX 3090 launched on September 24, 2020. The RTX 3090 Ti followed on March 29, 2022, with 10,752 CUDA cores, faster memory, and a 450W TDP at $1,999.

The launch coincided with a perfect storm: COVID-19 supply disruptions, a cryptocurrency mining boom, scalpers, and a global chip shortage. RTX 3090 prices on resale markets exceeded $3,000-$4,000 throughout 2021. The generation also saw the end of Nvidia's relationship with EVGA, which exited the GPU market entirely in September 2022.

The RTX 3090 experienced a bizarre price resurgence in 2024-2025 as the AI boom accelerated. Used prices climbed from $700-$900 in early 2024 back to $1,200-$1,500 by mid-2025, making it the most sought-after used GPU in the AI community.

With 24 GB VRAM, a single RTX 3090 could comfortably run 13B-34B parameter models at usable speeds. A 7B model in 4-bit requires ~4-5 GB, a 13B model ~8 GB, a 30B model ~18 GB. The RTX 4090 also had 24 GB but cost significantly more, and the RTX 4080 had only 16 GB. Two 3090s in parallel became a cost-effective alternative to enterprise GPUs for larger models.

At launch, the RTX 3090 delivered 60-80% faster than the RTX 2080 Ti and 10-15% faster than the RTX 3080 at 4K. In AI inference: Mistral 7B at ~50-60 tokens/second, LLaMA 2 13B at ~30-40 t/s (4-bit), Stable Diffusion XL at ~6-8 iterations/second. FP32 performance of ~35.6 TFLOPS made it a highly capable compute card.

Notable AIB partners included ASUS (ROG Strix, TUF Gaming), EVGA (FTW3, Kingpin - cancelled), MSI (Gaming X Trio, Suprim), Gigabyte (AORUS Xtreme), PNY, Zotac, Colorful, Inno3D, and Palit. The Founders Edition featured Nvidia's distinctive "flow-through" cooling design with a 12-layer PCB and vapor chamber.

• The Great GPU Shortage - Cards impossible to find at MSRP for 18+ months. Nvidia was sued by shareholders.
• 12VHPWR Connector (3090 Ti) - Introduced the 16-pin connector that later became infamous with the RTX 4090's melting issues.
• EVGA Capacitor Issue - Early cards had poor PCB soldering causing crashes. EVGA issued replacements.
• NVLink Removal - Two 3090s could not pool memory as one 48 GB device, a feature the 2080 Ti had.
• Used Market Scams - Fake refurbished cards, BIOS-flashed mining farm cards, and degraded memory became common pitfalls.

The RTX 3090's legacy is split between gaming and AI. In gaming, it was an expensive halo product. In compute and AI, it became perhaps the most important consumer GPU of its era - the card that let thousands of independent developers, students, and researchers run serious AI workloads on a budget. Production ceased in mid-2022, but by 2025 the used market had stabilized around $800-$1,200, maintained by insatiable AI demand.

Card	VRAM	Relative AI Perf	Used Price (2025)
RTX 3090	24 GB	1.0x (baseline)	$800-$1,200
RTX 4090	24 GB	1.6x	$1,400-$1,800
RTX 4080 Super	16 GB	1.3x	$900-$1,100
RX 7900 XTX	24 GB	0.5x (CUDA-dependent)	$700-$900