A December 2024 photograph attributed to Chiphell user harukaze5719 appeared to show NVIDIA’s enormous GB202-300 package surrounded by 16 Samsung GDDR7 memory chips. The image was unofficial, but its central suggestion was substantially correct: the GeForce RTX 5090 would use 32GB of GDDR7 on a 512-bit bus, delivering 1,792GB/s of bandwidth.
NVIDIA later confirmed those specifications when it announced the RTX 5090 on January 6, 2025. However, the photograph did not independently prove every detail reported alongside it—including the package dimensions, die area, VRM design or expected power draw.
As an Amazon Associate I earn from qualifying purchases.
What the leaked RTX 5090 photograph showed
The leak was reported by Notebookcheck on December 26, 2024, after an earlier leaked PCB image from the same source. The follow-up photograph appeared to show a large GPU package with memory packages arranged around it.
Free tools Windows power users keep installed
One-click scans. No signup required.
That wording needs a technical qualification. A GPU package is the complete semiconductor assembly mounted to a circuit board. The GPU die is the silicon inside that package. The surrounding GDDR7 components are separate memory packages mounted around the GPU on the board or package assembly; they are not memory chips physically embedded in the bare silicon die.
#1 Best Overall
- AI Performance: 772 AI TOPS
- OC mode: 2580 MHz Default mode: 2550 MHz(Boost clock)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- SFF-Ready Enthusiast GeForce Card
- Axial-tech fans feature a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
Notebookcheck attributed the image to Chiphell user harukaze5719 and identified the GPU as GB202-300. Because NVIDIA did not authenticate the photograph, those details remained leak-era claims at the time.
Why 16 memory chips suggested 32GB
The photograph reportedly showed 16 Samsung GDDR7 chips. If each package provided 2GB, the total would be:
16 chips × 2GB = 32GB
Sixteen memory packages also fit the logic of a 512-bit interface: sixteen 32-bit memory channels can combine to form a 512-bit bus. The leak-era calculation assumed 28Gbps memory, producing:
28Gbps × 512 ÷ 8 = 1,792GB/s
The chip count alone could not prove the memory speed or electrical interface. Those were inferences based on the expected configuration. NVIDIA subsequently confirmed all three headline figures: 32GB of GDDR7, a 512-bit interface and 1,792GB/s of bandwidth on the official RTX 5090 specification page.
How large was the GB202 die?
The leak report estimated the complete package at approximately 63 × 56mm. It estimated the die itself at roughly 24 × 21mm, or about 744mm2 of silicon area.
Rank #2
- [3352 AI TOPS, 5th Gen Tensor Cores, AI Content Creation] Built for AI-assisted photo and video workflows including upscaling, denoise, background removal, masking, and generative AI creation for faster creator productivity.
- [32GB GDDR7 VRAM, Local LLM Inference, Larger Models] Run local LLM inference and on-device AI tools with massive VRAM headroom for larger models, longer context, and heavier multitasking across AI and creator apps.
- [28 Gbps, 512-bit, 21760 CUDA Cores] High-throughput next-gen memory and core resources for demanding creator projects, complex timelines, large assets, and GPU-accelerated ML experimentation and inference pipelines.
- [Quad-Fan Force, Vapor Chamber, Phase-Change Thermal Pad] Designed for sustained performance under heavy loads with quad-fan cooling, a patented vapor chamber, and a phase-change GPU thermal pad to help lower temps and reduce hotspots.
- [DP 2.1b x3, HDMI 2.1b x2, Bundle GPU Holder] Multi-display ready with up to 4 displays and up to 7680 x 4320 max digital resolution, plus an included GPU Holder to help reduce GPU sag and improve long-term build stability.
Those measurements should still be treated as estimates derived from the leaked image, not as NVIDIA-confirmed dimensions. Package size and die size are different measurements, and the larger package also includes the substrate and other packaging structures.
NVIDIA’s Blackwell architecture documentation describes the full GB202 design as containing approximately 92.2 billion transistors. It does not, by itself, authenticate the photograph or validate its specific physical measurements.
Recommended Free Tools
Leak claims versus the final RTX 5090
| Leak-era claim | Official result | Assessment |
|---|---|---|
| GB202 GPU | RTX 5090 uses GB202 | Broadly validated |
| 21,760 CUDA cores | 21,760 CUDA cores | Confirmed |
| 16 GDDR7 packages | 32GB GDDR7 | Configuration was directionally correct |
| 512-bit interface | 512-bit interface | Confirmed |
| 1,792GB/s bandwidth | 1,792GB/s | Confirmed |
| 500–550W power expectation | 575W TGP | Final rating was higher |
| 24-phase VRM | Not established by the cited official material | Remains unconfirmed |
| 12V-2×6 connector | Adapter or 450W-or-greater PCIe Gen 5 cable guidance | Needs qualification |
The leak also correctly anticipated a CES 2025 reveal. NVIDIA announced the card on January 6, followed by retail availability on January 30, 2025.
Why GB202 has more CUDA cores than the RTX 5090
NVIDIA’s architecture material describes a full GB202 configuration with 24,576 CUDA cores and 192 streaming multiprocessors. The consumer RTX 5090 exposes 21,760 CUDA cores and 170 active SMs.
There is no contradiction. A GPU family can be designed with more units than a particular product activates. The RTX 5090 is a partially enabled implementation of the larger GB202 design, with some units unused or disabled. Therefore, “full GB202 has 24,576 CUDA cores” and “the RTX 5090 has 21,760” describe different levels of the same design.
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4. System Requirements: Minimum 850W PSU with 16-pin 12V-2x6 (12VHPWR) connector required. Verify before purchasing.
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability. Compatibility: 348mm (13.7") length, 3.6 slots, 4.3 lbs. Confirm case clearance and slot spacing. GPU bracket included.
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.6-slot design with massive fin array optimized for airflow from three Axial-tech fans
- Phase-change GPU thermal pad helps ensure optimal thermal performance and longevity, outlasting traditional thermal paste for graphics cards under heavy loads
RTX 5090 versus RTX 4090
| Specification | RTX 4090 | RTX 5090 |
|---|---|---|
| Architecture | Ada Lovelace | Blackwell |
| CUDA cores | 16,384 | 21,760 |
| Memory | 24GB GDDR6X | 32GB GDDR7 |
| Memory interface | 384-bit | 512-bit |
| Memory bandwidth | 1,008GB/s | 1,792GB/s |
| Total graphics power | 450W | 575W |
| Launch starting price | $1,599 | $1,999 |
The wider bus and additional memory are important for bandwidth-heavy games, content creation and some local AI workloads. They do not guarantee a proportional increase in every game: architecture, clocks, cache behavior, software and workload characteristics also matter.
Power, cooling and case requirements
The final RTX 5090 is a 575W graphics card, not a 500–550W product as the leak report expected. NVIDIA recommends an 850W minimum power supply for the Founders Edition. Its guidance supports either three or four PCIe 8-pin cables through the included adapter, or one PCIe Gen 5 power cable rated for at least 450W.
TGP refers to the graphics card’s rated graphics power, not the complete computer’s wall consumption. A high-quality PSU with suitable cabling and adequate transient-response capability remains important, particularly for factory-overclocked partner cards.
NVIDIA lists the Founders Edition at 304 × 137mm, with installation clearance equivalent to a three-slot card and additional room for the power cable. That does not describe every RTX 5090. Add-in-board models from vendors such as ASUS, GIGABYTE, MSI, PNY and ZOTAC can be longer, taller, thicker or more demanding than the Founders Edition. Check the exact card’s dimensions, slot count, PSU recommendation and connector instructions before buying.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What NVIDIA’s performance claims mean
NVIDIA promoted the RTX 5090 as up to twice as fast as the RTX 4090 in selected workloads. That headline depends on the software features and test conditions, particularly DLSS 4 and Multi Frame Generation. It should not be read as a universal doubling of native rasterization performance.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallRank #4
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5080
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
For a meaningful comparison, separate traditional rasterization, ray tracing, DLSS Super Resolution, Frame Generation, Multi Frame Generation and AI workloads using FP4. NVIDIA’s launch figures are the company’s own marketing claims, not independent benchmark results.
Was the leak authentic?
The most defensible conclusion is that the leak was substantially validated by the product NVIDIA later announced, but the photograph itself remained unofficial. Its apparent GB202 identity, 21,760 CUDA-core count and 32GB memory configuration aligned closely with the final RTX 5090.
That does not prove every accompanying claim. The exact 63 × 56mm package measurement, 24 × 21mm die estimate, 744mm2 area, Samsung identification, 24-phase VRM and 500–550W forecast should remain attributed to the leak report or described as unconfirmed. The final official 575W TGP also exceeded the report’s central power estimate.
What the leak means for buyers
- For 4K gaming: the RTX 5090 is a flagship option, but evaluate independent tests and distinguish native performance from DLSS-assisted results.
- For creators and AI users: 32GB of VRAM and 1,792GB/s of bandwidth can be valuable, although software support, model size and quantization still determine what workloads fit.
- For system builders: verify PSU capacity, cable compatibility, case clearance and cooling before selecting a card.
- For price-sensitive buyers: the RTX 5080 launched at $999 with 16GB and a 360W TGP; the RTX 5070 Ti launched at $749 with 16GB and 300W, while the RTX 5070 launched at $549 with 12GB and 250W. These cards trade memory capacity and flagship performance for lower cost and power demand.
- For availability: $1,999 was the RTX 5090 Founders Edition starting price, not a guarantee that every partner model would sell at that price. NVIDIA’s marketplace listing observed for this coverage showed the Founders Edition at $1,999 but out of stock, so current stock and street pricing should be checked before purchase.
The photograph mattered because it made the RTX 5090’s design direction visible before launch: a very large flagship GPU paired with an unusually wide memory subsystem. Its strongest predictions survived official scrutiny, but the image was evidence—not a substitute for NVIDIA’s final specification sheet.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
