The RTX 50 series Blackwell GPU has 92 billion transistors, 4,000 TOPS and four petaflops of AI, three times the previous-generation Ada.

Test the flagship RTX 50-series Blackwell GPU (GB202 / RTX 5090) against NVIDIA's official specifications. The claim has three parts: (a) about 92 billion transistors; (b) about 4,000 AI TOPS, which is the same thing as 'four petaflops of AI'; (c) about three times the AI performance of the previous Ada flagship (RTX 4090). The claim's figures are rounded, so (b) holds if the official figure is within about 10% of 4,000 (roughly 3,600-4,400), and (c) holds if the official ratio is about 2.7-3.3x on NVIDIA's stated comparison basis. TRUE: all three parts hold. MOSTLY TRUE: (a) holds, and one of (b) or (c) is slightly outside its tolerance (within about 20%). MIXED: one part clearly fails while the others hold. MOSTLY FALSE: two parts fail. FALSE: the transistor count and the AI throughput figures are all substantially wrong.

Set before the research began.

1. E1 NVIDIA Blackwell GeForce RTX 50 Series Opens New World of AI Computer Graphics NVIDIA Corporation (investor relations press release) · 6 January 2025 Own record

   “The GeForce RTX 5090 GPU — the fastest GeForce RTX GPU to date — features 92 billion transistors, providing over 3,352 trillion AI operations per second (TOPS) of computing power.”

2. E2 NVIDIA RTX BLACKWELL GPU ARCHITECTURE NVIDIA Corporation · 28 January 2025 Own record

   “The full GB202 chip consists of 92.2 billion transistors and contains 12 GPCs, 96 TPCs, 192 SMs, and sixteen 32-bit memory controllers (512-bit total).”

3. E3 Nvidia announces RTX 5090 for $1,999, 5070 for $549 — plus AI, DLSS 4, and more Tom's Hardware · 7 January 2025

   “The 125 TFLOPS figure is also accompanied by a maximum 4,000 INT8 TOPS, while the RTX 5090 scales that down to 3,400 TOPS — 85% of a fully enabled GB202, give or take.”

4. E4 Nvidia announces RTX 5090 for $1,999, 5070 for $549 — plus AI, DLSS 4, and more Tom's Hardware · 7 January 2025

   “Nvidia also showed "RTX Blackwell" offering 125 TFLOPS of FP32 graphics compute via the shaders, which is 1.5X more than its Ada generation counterpart, while the AI performance will be 3X as high.”

NVIDIA's press release and architecture whitepaper put the RTX 5090 / GB202 at 92 billion (92.2 billion) transistors, so part (a) holds (E1, E2). For part (b), NVIDIA's official RTX 5090 figure is 3,352 AI TOPS, about 16% below 4,000 (E1). Tom's Hardware reports that the 4,000 TOPS figure applies only to the fully enabled GB202 chip, of which the 5090 delivers about 85% (E3). For part (c), Tom's Hardware reports NVIDIA's comparison that RTX Blackwell AI performance is 3X that of Ada (E4). The shipping flagship's AI TOPS falls outside the 10% tolerance but within 20%, so under the criteria the claim resolves mostly true.