Pictures and performance of NVIDIA’s upcoming RTX 3060 Ti graphics card have been leaked from Videocardz (who else). NVIDIA’s RTX 3060 Ti graphics card will have 4864 CUDA cores and will be faster than the older RTX 2080 SUPER and priced at only $ 399 USD. The card is expected to be launched in early December and should make NVIDIA’s mid-spectrum lineup killer.
NVIDIA RTX 3080 Ti images and performance leaked, landing December 2, 2020
NVIDIA GeForce RTX 3060 Ti has GA104-200 GPU with 4864 CUDA cores and 8 GB of GDDR6 memory clocked at 14 Gbps. The bus width will be 256 bits for a total bandwidth of 448 GB / s bandwidth. Since the card will require less power, the TBP should be around 200W with the reference design clocking around 180W. This is a 5W increase over the RTX 2070 at 175W and in line with how we expect the Samsung node to be fair in terms of energy efficiency compared to TSMC’s nodes. It is expected to have an MSRP of $ 399 and will launch on December 2nd.
Founder’s edition variant has the now well-known metallic black aesthetic from RTX 3080 and RTX 3090, but both fans are at the top of the card with a small cut in the back plate that is visible on the back. The overall shape and form of the founder’s edition is very consistent with the RTX 3000 series. The fan towards the end of the card will push the air through the heat sink and out to the bottom – which is an interesting design philosophy compared to graphics cards from earlier this decade.
In terms of benchmarks, the RTX 3060 Ti is faster than the RTX 2080 SUPER and should trade with the RTX 2080 Ti. It is faster not only in shadow performance but also reproduction (calculation) and beam tracking and destroys the older RTX 2060 SUPER it replaces, completely. Of course, as with any benchmark comparison, always keep in mind that you have to wait for third-party results before making any decision.
Keep in mind that the reason NVIDIA’s RTX 3090 and RTX 3080 cards do not scale linearly in performance is due to bottlenecks from game engines and binaries that are not designed to handle as much kernel. We know that the hardware and driver stack scales linearly due to the performance of these cards in software such as vRAY and Octane – which are designed to handle a huge amount of graphics power. This means that when you scale down kernels, performance will does not scales down linearly. In fact, I fully expect the RTX 3060 Ti to beat the RTX 2080 clock to the clock! Assuming there are slightly fewer watches than the RTX 2080 Ti, it should easily trade blows with the previous flagship.
The yields for the first few batches of the Samsung 8nm node are lower than expected and expensive. The company expects the yield (and thus the prices of wafers) to improve significantly in the coming months. To protect its shareholders, the company has been careful with its orders in the first few batches and has placed significantly higher volume orders for later batches. This problem was further exacerbated by bots scraping the entire volume from the start. We expect bulk orders to start hitting by the end of the year (assuming once again that miners obviously don’t eat it all).
NVIDIA GeForce RTX 30 Series ‘Ampere’ Graphics Card Specifications:
|Graphics card name||NVIDIA GeForce RTX 3050||NVIDIA GeForce RTX 3050 Ti||NVIDIA GeForce RTX 3060||NVIDIA GeForce RTX 3060 Ti||NVIDIA GeForce RTX 3070||NVIDIA GeForce RTX 3070 Ti?||NVIDIA GeForce RTX 3080||NVIDIA GeForce RTX 3080 Ti?||NVIDIA GeForce RTX 3090|
|GPU name||Ampere GA107||Ampere GA106?||Ampere GA106?||Ampere GA104-200||Ampere GA104-300||Ampere GA102-150||Ampere GA102-200||Ampere GA102-250||Ampere GA102-300|
|Process node||Samsung 8nm||Samsung 8nm||Samsung 8nm||Samsung 8nm||Samsung 8nm||Samsung 8nm||Samsung 8nm||Samsung 8nm||Samsung 8nm|
|Die size||TBA||TBA||TBA||395.2 mm2||395.2 mm2||628.4 mm2||628.4 mm2||628.4 mm2||628.4 mm2|
|Transistors||TBA||TBA||TBA||17.4 billion||17.4 billion||28 billion||28 billion||28 billion||28 billion|
|TMUs / ROPs||TBA||TBA||TBA||152/80||184/96||232/80||272/96||328/112||328/112|
|Tensor / RT cores||TBA||TBA||TBA||152/38||184/46||232/58||272/68||328/82||328/82|
|Base ur||TBA||TBA||TBA||1410 MHz||1500 MHz||TBA||1440 MHz||TBA||1400 MHz|
|Boost Clock||TBA||TBA||TBA||1665 MHz||1730 MHz||TBA||1710 MHz||TBA||1700 MHz|
|FP32 Compute||TBA||TBA||TBA||16.2 TFLOPs||20 TFLOPs||TBA||30 TFLOPs||TBA||36 TFLOPs|
|RT TFLOPs||TBA||TBA||TBA||32.4 TFLOPs||40 TFLOPs||TBA||58 TFLOPs||TBA||69 TFLOPs|
|Tensor TOPs||TBA||TBA||TBA||TBA||163 TOPs||TBA||238 TOPPER||TBA||285 TOPPER|
|Memory capacity||4 GB GDDR6?||6 GB GDDR6?||6 GB GDDR6?||8 GB GDDR6||8 GB GDDR6||10 GB GDDR6X?||10 GB GDDR6X||20 GB GDDR6X||24 GB GDDR6X|
|Memory speed||TBA||TBA||TBA||14 Gbps||14 Gbps||TBA||19 Gbps||19 Gbps||19.5 Gbps|
|Bandwidth||TBA||TBA||TBA||448 Gbps||448 Gbps||TBA||760 Gbps||760 Gbps||936 Gbps|
|Price (MSRP / FE)||$ 149?||$ 199?||$ 299?||$ 399 US?||$ 499 US||$ 599 US?||$ 699 US||$ 899 US?||1499 US $|
|Start (availability)||2021?||2021?||2021?||November 2020?||October 29||Q4 2020?||September 17||January 2021?||September 24th|