RTX 3050s may have 4 or 8GB of memory

The well-known leaker @ kopite7kimi revealed the specifications of NVIDIA’s much talked about GeForce RTX 3050, an entry-level graphics card based on the Ampere architecture. NVIDIA is said to have originally planned to put the GA107 GPU inside the GeForce RTX 3050. However, the latest update from kopite7kimi claims that the chip maker has backtracked on the decision and will reportedly use the GPU instead. GA106. This is a good choice, since the GA106 is already used for the GeForce RTX 3060, so NVIDIA could potentially retrieve the non-standard dies for the GeForce RTX 3060 and plug them into the GeForce RTX 3050.

The GeForce RTX 3050 will come in two versions with different memory endowments and core counts, according to the leaker. The GA106-150 die is specifically designed for the GeForce RTX 3050 8GB, while the GA106-140 will go in the GeForce RTX 3050 4GB. Logically, the eight GB variant will be the more powerful of the two. The GA106 GPU measures 276mm² and hosts up to 30 multiprocessor (SM) streams, for a total of 3,840 CUDA cores. The GeForce RTX 3060 has 28 enabled SMs, only two fewer than the full die. To ensure a fair margin of separation, even in view of a potential GeForce RTX 3050 Ti in the future, the 8GB and 4GB GeForce RTX 3050s will likely be equipped with 20 and 18 SMs, respectively. This would equate to 2,560 CUDA cores on the 8GB variant and 2,304 CUDA cores on the 4GB variant.

GeForce RTX 3060 GeForce RTX 3050 8GB GeForce RTX 3050 4GB Architecture GA106 GA106 GA106 Process Technology Samsung 8N Samsung 8N Samsung 8N Transistor (Billion) 12 12 12 Die Size (mm²) 276 276 276 SM / CU 28 20 18 GPU Core 3,584 2,560 2,304 Tensor Core 112 80 72 RT Core 28 20 18 Base Clock (MHz) 1,320? ? Boost Clock (MHz) 1.777 ? VRAM Speed ​​(Gb / s) 15 14 14 VRAM (GB) 12 8 4 VRAM bus width 192 128 128 ROP 48? ? TMU 112 80 72 TFLOPS FP32 (Boost) 12.7? ? TFLOPS FP16 (Tensor) 51 (102)? ? RT TFLOPS 24.9? ? Bandwidth (GB / s) 360 224 224 TDP (Watts) 170 90 90 Launch Date Feb 21 Jan 22 Jan 22 Introductory Price $ 329? ? Like the GeForce RTX 3060, the GeForce RTX 3050 will presumably use GDDR6 memory. However, the latter may use lower clocked memory chips, perhaps 14Gb / s instead of the 15Gb / s that NVIDIA uses on the former. We should also expect a smaller memory interface, perhaps 128-bit, for a maximum memory bandwidth of 224GB / s.

According to the latest rumors, NVIDIA may introduce the GeForce RTX 3050 at CES 2022 , while it should not debut on store shelves until January 27.