News GeForce RTX 3050 Rumors: Baby Ampere with 2304 Cores Due in 2021

[Admin]

RTX 3050 Specs Rumored on Twitter, 2304 CUDA cores, with ray tracing and DLSS support.

GeForce RTX 3050 Rumors: Baby Ampere with 2304 Cores Due in 2021 : Read more

 

[cryoburner]

If we apply that math to the supposed RTX 3050, the RTX 3050 would perform just above a GTX 1660 Ti / Super and just below an RTX 2060.

If we went by core counts alone, a 1650 SUPER should perform within about 9% of a 1660 SUPER. However, even at 1080p, it's closer to 25% behind that card. A lot of that likely comes down to the reduced memory bandwidth resulting from reducing the number of VRAM chips. Seeing as Nvidia hasn't really been pushing much more VRAM this generation, with the 3070 having the same amount as the 2070 before it, it's very possible that the 3050 will have 4GB.

A 3050 with 2304 Ampere cores and 6GB of VRAM could potentially perform in-between a 1660 SUPER and a 2060, but with 4GB, it might not perform any better at rasterized rendering than the 1660 SUPER, and the limited VRAM would likely hurt performance more in demanding titles, especially moving forward. From an RT-capability standpoint, the 30-series cards don't really add much RT performance relative to the rasterized performance they deliver, so RT performance would most likely be a bit below that of a 2060 as well, which is already kind of borderline in terms of usefulness with raytraced lighting effects enabled.

 

[Pytheus]

I don't understand why they need to release new entry level cards, just lower the price point of the previous generation to align with their relative performance.

 

[cryoburner]

I don't understand why they need to release new entry level cards, just lower the price point of the previous generation to align with their relative performance.

It's not likely to be as profitable for them to move those older cards down to lower price points, and the cards would also require more power to run. From the cost perspective, they have moved to a newer process node with the 30-series, along with an updated architecture, so they can get more graphics chips of a given performance level out of a single wafer. And as I pointed out above, they will likely adjust the amount of VRAM as well.

Take for example the RTX 3070, a card that performs roughly similar to a 2080 Ti in today's games. The graphics chip that goes into the 3070 is almost half the size of the one in the 2080 Ti, and VRAM has been reduced from 11GB to 8GB. With the 3070 having a suggested starting price that's half that of the 2080 Ti, it's only natural for them to look for ways to reduce manufacturing costs. That goes even more so for entry-level cards, where the profit margins tend to be smaller to begin with.

 

[Pytheus]

it's only natural for them to look for ways to reduce manufacturing costs. That goes even more so for entry-level cards, where the profit margins tend to be smaller to begin with.

I don't disagree. I don't know what it costs to develop and tool for the new chips verses what it costs them to produce the older chips which their supplier is already tooled for. In the case of the 2080ti I can understand eliminating it completely since newer parts give the same performance, but producing gimped versions of the new chips just to create a value line seems counterproductive and adds development cost. I'm sure they can work out something to drop a 2070 Super by 100 bucks.