100% Original Wholesale Brand new AD104-400 AD104-250/AD104-251 ADA lovelace type

Feature	AD104-400 (e.g., RTX 4070 Ti)	AD104-250 (Lower SKU)	AD104-251 (Revised)	Notes
Architecture	Ada Lovelace	Ada Lovelace	Ada Lovelace	Same architecture
Process Node	TSMC 4N	TSMC 4N	TSMC 4N	No node change
CUDA Cores	7,680 (Full AD104)	~5,888 (Disabled Cores)	~6,144 (Partial Enable)	Bin-determined cuts
RT Cores	3rd Gen (60 cores)	~46 (Scaled down)	~48	Proportional to CUDA cores
Tensor Cores	4th Gen (240 cores)	~184	~192	For DLSS/AI workloads
Base Clock	2.31 GHz	~2.1 GHz	~2.2 GHz	Lower TDP targets
Boost Clock	2.61 GHz	~2.4 GHz	~2.5 GHz	Potential binning gains
VRAM	12GB GDDR6X	12GB GDDR6	12GB GDDR6X	-251 may use faster memory
Memory Bus	192-bit	192-bit	192-bit	Same bus width
Bandwidth	504 GB/s	432 GB/s (GDDR6)	504 GB/s (GDDR6X)	-251 matches -400
TDP	285W	~200W	~220W	Efficiency tweaks in -251
PCIe Interface	PCIe 4.0 x16	PCIe 4.0 x16	PCIe 4.0 x16	No PCIe 5.0 support
NVLink Support	No	No	No

1. AD104-400 (Flagship Variant)

Overview

The AD104-400 is the fully-enabled version of NVIDIA’s AD104 die, representing the pinnacle of the company’s mid-range Ada Lovelace offerings. This GPU powers high-performance consumer cards like the RTX 4070 Ti and RTX 4070 Super, delivering exceptional 1440p and entry-level 4K gaming performance.

Key Specifications

• CUDA Cores: 7,680 (Full AD104 implementation)
• RT Cores: 60 (3rd Generation)
• Tensor Cores: 240 (4th Generation, supports DLSS 3.5)
• Memory: 12GB GDDR6X (192-bit bus, 504 GB/s bandwidth)
• Clock Speeds:
  • Base: 2.31 GHz
  • Boost: 2.61 GHz (up to 2.8 GHz with OC)
• TDP: 285W (reference design)

Performance & Use Cases

• Gaming: Dominates 1440p ultra settings, capable of 4K with DLSS
• Content Creation: Excellent for Blender, Premiere Pro, and DaVinci Resolve
• AI Acceleration: Strong Stable Diffusion and local LLM performance

Special Features

• Full Ada Lovelace architecture benefits
• Superior cooling solutions (vapor chamber in FE cards)
• Best-in-class efficiency for its performance tier

---

2. AD104-250 (Cut-Down Variant)

Overview

The AD104-250 represents a cost-optimized version of the AD104 die, with disabled cores and potentially lower-grade memory. This variant is typically used in OEM systems, mobile workstations, or budget-oriented desktop GPUs.

Key Specifications

• CUDA Cores: ~5,888 (23% reduction from full die)
• RT Cores: ~46 (3rd Generation)
• Tensor Cores: ~184 (4th Generation)
• Memory: 12GB GDDR6 (192-bit bus, 432 GB/s bandwidth)
• Clock Speeds:
  • Base: ~2.1 GHz
  • Boost: ~2.4 GHz
• TDP: ~200W

Performance & Use Cases

• Gaming: Solid 1440p performance, may require DLSS for demanding titles
• Professional Workloads: Suitable for CAD and light rendering
• Efficiency: Better suited for compact systems and mobile applications

Special Notes

• Likely used in “cut-down” retail SKUs or OEM-only models
• Memory downgrade to GDDR6 helps reduce costs
• Power efficiency makes it viable for SFF builds

---

3. AD104-251 (Revised Variant)

Overview

The AD104-251 is believed to be a refined version of the AD104 silicon, potentially offering better binning and minor architectural improvements. This variant might power refreshed models like hypothetical “Super” series cards.

Key Specifications

• CUDA Cores: ~6,144 (20% reduction from full die)
• RT Cores: ~48 (3rd Generation)
• Tensor Cores: ~192 (4th Generation)
• Memory: 12GB GDDR6X (192-bit bus, 504 GB/s bandwidth)
• Clock Speeds:
  • Base: ~2.2 GHz
  • Boost: ~2.5 GHz
• TDP: ~220W

Performance & Use Cases

• Gaming: Bridges gap between AD104-250 and AD104-400
• Content Creation: Improved performance over -250 variant
• Efficiency: Better performance-per-watt than full -400 die