Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesNVIDIA’s most powerful consumer Ampere graphics cards were built on a process that looked old on paper: Samsung 8nm. The GeForce RTX 3090 and RTX 3080 used Samsung’s customized 8N NVIDIA Custom Process, yet the resulting GA102 GPU packed 28.3 billion transistors into a 628.4 mm² die.
That apparent contradiction explains Ampere’s design better than the phrase “old 8nm” does. Samsung 8N was less dense than contemporary TSMC 7nm in the relevant comparisons, but it was not simply an obsolete off-the-shelf process. NVIDIA combined it with a radically expanded architecture, faster memory and a much higher power budget to produce one of the most aggressive consumer GPU generations of its era.
As an Amazon Associate I earn from qualifying purchases.
The short answer
The consumer GeForce RTX 30-series Ampere chips, known as the GA10x family, were manufactured on Samsung’s customized 8N process. That includes the GA102 chip used by the GeForce RTX 3090 and RTX 3080.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →But not every Ampere GPU used Samsung 8N. NVIDIA’s much larger data-center GA100 was manufactured by TSMC on 7nm. The split shows that NVIDIA did not reject TSMC 7nm outright; it used different manufacturing strategies for different products and markets.
#1 Best Overall
- AI Performance: 767 AI TOPS
- OC mode: 2632 MHz (OC mode)/ 2602 MHz (Default mode)
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Axial-tech fan design features a smaller fan hub that facilitates longer blades and a barrier ring that increases downward air pressure
- A 2.5-slot design maximizes compatibility and cooling efficiency for superior performance in small chassis
Samsung 8N imposed trade-offs in density, power and die area. Ampere overcame those limits through architectural scale: more execution resources, redesigned ray-tracing and Tensor hardware, GDDR6X memory on the high-end cards, and a willingness to consume substantially more power.
What “8N NVIDIA Custom Process” actually means
NVIDIA’s GA102 documentation identifies the process as Samsung 8nm 8N NVIDIA Custom Process. The “custom” designation matters. This was not necessarily an ordinary Samsung consumer process used without modification; it was a version optimized for NVIDIA’s GPU design and requirements.
The number in a modern process name is also not a literal measurement of every transistor feature. Samsung’s 8nm and TSMC’s 7nm were competing node labels from different foundries, not directly comparable rulers. A useful process comparison has to consider practical transistor density, voltage and frequency behavior, leakage, yield, wafer economics, capacity and design compatibility.
So calling Samsung 8N “old” is only accurate in a relative sense: it was less advanced by the contemporary node labels and density comparisons than TSMC N7. That does not make it unusable, poor or technologically irrelevant. It offered enough capability for NVIDIA to build enormous GPUs and meet its performance targets.
Which Ampere GPUs used Samsung 8N?
| GPU family | Representative products | Process |
|---|---|---|
| GA102 | GeForce RTX 3090, RTX 3080 | Samsung 8nm 8N NVIDIA Custom Process |
| GA104 | GeForce RTX 3070 and related products | Samsung 8N |
| GA10x | Consumer GeForce RTX 30-series generally | Samsung 8N custom process |
| GA100 | NVIDIA A100 data-center accelerator | TSMC 7nm |
The distinction between GA102 and GA100 is essential. “Ampere used Samsung 8nm” is a useful shorthand for the consumer generation, but it is wrong as a statement about the entire architecture.
The RTX 3080 and RTX 3090 both used GA102, although the RTX 3080 disabled more of the chip. NVIDIA’s full GA102 specification lists 10,752 CUDA cores, while the RTX 3090 shipped with 10,496 active CUDA cores. The RTX 3080 shipped with 8,704.
Why GA102 was such a huge chip
NVIDIA’s GA102 whitepaper lists:
- 28.3 billion transistors
- 628.4 mm² of die area
- Up to 10,752 CUDA cores
- 84 second-generation RT cores
- 336 third-generation Tensor cores
- A 384-bit memory interface
- A reference architecture power target of 300 W
Using the published transistor and die-size figures, GA102 averages roughly 45.0 million transistors per mm². That is a derived estimate, not a directly quoted density specification, and it should not be treated as a universal rating for Samsung 8N. Still, it illustrates the central compromise: NVIDIA fit an enormous amount of logic onto a very large die rather than relying on maximum process density alone.
Large dies are difficult and expensive to manufacture. A larger die means fewer potential chips per wafer, and each die has more area that can encounter a defect. Yield and cost therefore become increasingly important as the design approaches the practical limits of the process.
Rank #2
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5060
- Integrated with 8GB GDDR7 128bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Ampere’s architectural brute force
The process node was only one part of the performance equation. NVIDIA changed the structure of the streaming multiprocessor so that Ampere could execute twice as many FP32 shader operations per clock as the equivalent Turing design, according to NVIDIA’s GA102 documentation.
That change helped NVIDIA advertise up to 30 FP32 TFLOPS for the RTX 3080, compared with 11 TFLOPS for the comparable Turing GPU cited in the whitepaper. Theoretical throughput is not the same as game performance, however. Real applications depend on workload, clocks, memory behavior, software and bottlenecks elsewhere in the system.
Ampere also brought newer RT cores for ray-tracing workloads and third-generation Tensor cores for AI operations. On the high-end cards, GDDR6X memory supplied substantially more bandwidth than the GDDR6 used on many previous GPUs. The RTX 3090 combined that bandwidth with 24 GB of memory and a 384-bit bus, while the launch RTX 3080 used 10 GB of GDDR6X.
The result was a system-level performance increase created by several decisions at once:
- The process allowed more transistors than Turing’s 12nm consumer design.
- The architecture added more execution capacity and changed FP32 throughput.
- RT and Tensor hardware became more capable.
- GDDR6X increased memory bandwidth on flagship models.
- NVIDIA accepted higher power consumption to sustain the resulting performance.
Why NVIDIA used Samsung instead of TSMC 7nm for GeForce
The public evidence supports the manufacturing split, but not one definitive explanation for the commercial decision. It is too strong to say that TSMC had no capacity, that Samsung won solely because it offered cheaper wafers, or that NVIDIA was forced into the arrangement.
The likely decision involved several factors:
- Capacity and timing: A foundry relationship is valuable only if the required process can deliver chips in the schedule and volume a product launch demands.
- Wafer economics: A very large consumer GPU makes wafer cost, usable die count and yield especially significant.
- Customization: NVIDIA could tune Samsung’s process for its own design goals rather than treating the node label as the entire product.
- Target performance: NVIDIA may have judged that Samsung 8N could deliver the desired gaming performance when paired with Ampere’s architecture and power envelope.
- Product segmentation: Consumer GeForce cards and data-center accelerators have different price structures, workloads, memory systems and manufacturing priorities.
The existence of GA100 on TSMC 7nm is the clearest counterexample to the idea that NVIDIA simply could not or would not use TSMC 7nm. It used that process for a different Ampere product family, while the consumer GA10x chips followed another path.
What Samsung 8N cost Ampere
Power
The RTX 3080 and RTX 3090 delivered major performance gains, but they were not low-power designs. NVIDIA’s GA102 material lists a 300 W reference target for the architecture. That figure should not be applied to every RTX 30-series card: lower-tier GPUs, partner boards and later variants can have different power specifications.
Free tools Windows power users keep installed
One-click scans. No signup required.
Higher power enabled higher sustained performance, but it also shifted the engineering burden to the rest of the card. Power delivery, voltage behavior, board layout and cooling all became more demanding.
Rank #3
- Powered by the NVIDIA Blackwell architecture and DLSS 4
- Powered by GeForce RTX 5070 Ti
- Integrated with 16GB GDDR7 256bit memory interface
- PCIe 5.0
- WINDFORCE cooling system
Thermals and physical size
The high-end cards required large heatsinks and aggressive fan designs. NVIDIA’s Founders Edition cards introduced a new cooling layout, while GDDR6X memory added another important thermal consideration. The RTX 3090’s large, triple-slot design was not an aesthetic accident; it was part of the hardware required to move heat away from a very powerful GPU and its memory.
NVIDIA marketed the RTX 3090 as a “BFGPU,” emphasizing its size and capability for extreme gaming and creator workloads, including 8K HDR gaming. Its physical scale reflected the underlying combination of a huge die, high board power and a large memory subsystem.
Density and manufacturing economics
A less dense process requires more silicon area for a given transistor budget than a denser alternative would. With a 628.4 mm² GA102 die, that matters. More area can mean fewer potential dies per wafer and greater exposure to manufacturing defects.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
That does not prove Samsung 8N made every RTX 30-series card unprofitable or inefficient. It explains why density is one of several criteria—alongside yield, wafer price, capacity, frequency, leakage and schedule—that determine whether a process is commercially attractive.
Efficiency and clock headroom
A denser or otherwise more advanced process might have offered better efficiency or frequency headroom, but there was no retail TSMC 7nm GA102 against which to make a controlled comparison. It is therefore impossible to state as fact that a hypothetical TSMC-built GeForce chip would have been faster, cooler or cheaper.
NVIDIA claimed up to a 1.9× improvement in power efficiency over Turing at the same performance level. That is a claim made under NVIDIA’s stated methodology, not a universal independent result that applies equally to every game, resolution or card.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why the RTX 3090 felt “monstrous”
The RTX 3090’s reputation came from the whole package:
Windows 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 reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minute- A 628.4 mm² GA102 die
- 28.3 billion transistors on the chip
- 10,496 active CUDA cores
- 24 GB of GDDR6X memory
- A 384-bit memory interface
- A high power target and substantial cooling system
- A near-full configuration of the large GA102 design
It was not the full GA102 configuration—the complete chip had 10,752 CUDA cores—but it was close. NVIDIA positioned it as a Titan-class product for extreme gaming and creator workloads rather than merely a conventional step above the RTX 3080.
Rank #4
- Powered by the NVIDIA Blackwell architecture and DLSS 4 OC mode: 2640MHz/Default mode: 2610MHz (Boost Clock)
- Military-grade components deliver rock-solid power and longer lifespan for ultimate durability
- Protective PCB coating helps protect against short circuits caused by moisture, dust, or debris
- 3.125-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
The RTX 3080 was also important to the story because NVIDIA put a cut-down version of this very large chip into a lower product tier. That helped create a dramatic generational performance jump, but it also meant the card inherited much of GA102’s power and thermal character.
Was Samsung 8N a mistake?
There is no evidence-based yes-or-no answer that ignores the alternatives. Samsung 8N was not the theoretically densest contemporary option, and its use contributed to the need for a large die and an aggressive power envelope. But it was capable enough for NVIDIA to produce exceptionally fast consumer GPUs on schedule and at the intended product scale.
Calling the decision a failure also ignores what Ampere achieved. The RTX 3090 and RTX 3080 were powerful because process technology, architecture, memory, board design and cooling were engineered as one system. Samsung 8N did not create the performance by itself, but neither did it prevent NVIDIA from delivering a major leap over Turing.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →The fairest conclusion is that NVIDIA traded some density and likely efficiency headroom for a process and manufacturing strategy that supported its consumer launch. It then compensated with architectural scale and power. Whether that was the optimal business decision cannot be proven from the existence of the finished cards alone.
The accurate way to describe Ampere’s manufacturing story
NVIDIA’s consumer Ampere GPUs were not “7nm GPUs,” and it is misleading to describe Samsung 8N as simply ordinary, obsolete 8nm technology. The precise description is more revealing:
Consumer Ampere used a customized Samsung 8N process to build unusually large GPUs, while NVIDIA relied on architectural expansion, faster memory and high power limits to achieve its performance targets. The data-center GA100 used TSMC 7nm instead.
That is why the RTX 3090 could be both a product of a comparatively less aggressive node and one of the most physically extreme consumer graphics cards of its generation. Its monstrous character came from the amount of silicon, logic, memory and power NVIDIA was willing to deploy—not from the process label alone.
Recommended Free Tools
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.
