NVIDIA launches GeForce RTX 3060
NVIDIA officially launches its first GA106 desktop graphics card.
Nearly three months after the launch of GeForce RTX 3060 Ti, NVIDIA adds a new graphics card to the series. Board partners have been eagerly awaiting this release because GeForce cards at this price range are always very popular. This year, however, AIBs are met with unprecedented demand, which will drive prices a lot higher.
First GeForce RTX 30 with 12GB memory
NVIDIA GeForce RTX 3060 is the fourth installment in the Ampere series. This graphics card is equipped with GA106 GPU, which is already available under RTX 3060 Laptop GPU. The desktop model, however, has fewer cores. NVIDIA decided to introduce this SKU with 3584 CUDA cores, which means that 256 cores are disabled from the full GPU. Surprisingly, this model features 12GB GDDR6 memory. This is more than RTX 3060 Ti’s 8GB and even more than RTX 3080’s 10GB. NVIDIA’s intentions here are unclear, but rumor has it, NVIDIA is also planning a 6GB model, which should be more affordable.
NVIDIA GeForce RTX 3060 Reference PCB, Source: VideoCardz
Crypto mining limiter
NVIDIA GeForce RTX 3060 is the first model to feature an anti-crypto algorithm, which is built into the firmware and software. NVIDIA says that this will not affect any other workload, such as gaming or rendering, and that it will impossible to hack. The crypto-limiter will automatically detect if the graphics card is used for cryptocurrency mining, which will then half the hash rate. This should technically make the graphics card less interesting to miners. Whether that is the case, we don’t know yet.
Resizable BAR
The new SKU also is the first to introduce ResizableBAR support. This technology allows the CPU to access the full VRAM. AMD has introduced this technology under its own name of Smart Access Memory. NVIDIA originally communicated to AIBs that it will try to implement this feature with a driver, but eventually announced that other RTX 30 models will require a BIOS update. The new firmware is now expected to be available next month. For the time being, RTX 3060 is currently the only NVIDIA graphics card to support it.
NVIDIA GeForce RTX 3060 reviews will be posted later today. The sales will begin shortly after. Do note that NVIDIA has not released a Founders Edition of the graphics card, which means that reviews will feature AIB cards only.
| NVIDIA GeForce RTX 3060 Series | ||||
|---|---|---|---|---|
| VideoCardz.com | GeForce RTX 3060 Ti | GeForce RTX 3060 | GeForce RTX 2060 SUPER | GeForce RTX 2060 |
| Picture |  | |  |  |
| Architecture | NVIDIA Ampere | NVIDIA Ampere | NVIDIA Turing | NVIDIA Turing |
| GPU | 8nm GA104-200 | 8nm GA106-300 | 12nm TU104-410 | 12nm TU106-300 |
| Base Clock | 1410 MHz | 1320 MHz | 1470 MHz | 1365 MHz |
| Boost Clock | 1665 MHz | 1777 MHz | 1650 MHz | 1680 MHz |
| CUDA : TMU : ROPs | 4864 : 152 : 80 | 3584 : 112 : 48 | 2176 : 136 : 64 | 1920 : 120 : 48 |
| Tensors : RTs | 152 : 38 | 112 : 28 | 272 : 32 | 240 : 30 |
| Memory | 8GB GDDR6 | 12GB GDDR6 | 8GB GDDR6 | 6GB GDDR6 |
| Memory Clock | 14 Gbps | 15 Gbps | 14 Gbps | 14 Gbps |
| Memory Bus | 256-bit | 192-bit | 256-bit | 192-bit |
| Memory Bandwidth | 448 GB/s | 360 GB/s | 448 GB/s | 336 GB/s |
| TGP/TBP | 200W | 170W | 175W | 160W |
| MSRP | 399 USD | 329 USD | 399 USD | 349 USD |
| Release Date | December 2nd, 2020 | February 25th, 2021 | July 9th, 2019 | January 14th, 2019 |
NVIDIA GA106 GPU
NVIDIA GA106 processor has three GPU Processing Clusters, each with five Texture Processing Clusters. The full die has 30 Streaming Multiprocessors, but the RTX 3060 only has 28 enabled. Each SM has one RayTracing Core, which means it will have 28. The number of Tensor cores has decreased from RTX 2060’s 240 to 112, however, those are of a newer, third generation.
NVIDIA GA106 GPU Diagram, Source: VideoCardz
NVIDIA is still expected to launch entry-level GA107 GPU later this year, however, for this moment, the lineup features four GPUs:
| NVIDIA Ampere GPUs | ||||
|---|---|---|---|---|
| VideoCardz.com | GA100 | GA102 | GA104 | GA106 |
| Picture (not to scale) |  |  |  |  |
| Fabrication Node | 7nm TSMC N7 | 8nm Samsung | 8nm Samsung | 8nm Samsung |
| Transistors | 54.2B | 28.3B | 17.4B | |
| Die Size | 826 mm2 | 628 mm 2 | 392 mm 2 | ~276 mm 2 |
| GPCs | 8 | 7 | 6 | 3 |
| TPCs | 64 | 42 | 24 | 15 |
| SMs | 128 | 84 | 48 | 30 |
| CUDAs | 8192 | 10752 | 6144 | 3840 |
| Tensors | – | 336 | 192 | 120 |
| RTs | – | 84 | 48 | 30 |
| Memory Bus | 3072-bit | 384-bit | 256-bit | 192-bit |
| Memory Technology | HBM2 | GDDR6X | GDDR6 | GDDR6 |
GPC – GPU Processing Cluster, TPC – Texture Processing Cluster, SM – Streaming Multiprocessor