A couple of days ago we saw that things were not looking good for the 20 series RTX, a generation of graphics cards that will be succeeded this year by the 30 series RTX and that, according to the latest information, going to age as badly as the GTX 600 and GTX 700 did based on Kepler.
I explain this comparison, since although it is interesting it could be correct perhaps not everyone has come to understand it. The Kepler architecture was designed to work well with DirectX 11, but its support with DirectX 12 was quite poor. This allowed them to offer excellent performance in games based on the first API, but with the increasing DirectX 12 and Vulkan adoption, coupled with the lack of improvements at the driver level, its performance has declined significantly.
There is an example that I think perfectly illustrates GTX 780 Ti, a graphics card that at the time cost $ 699, it performs less than current low-end graphics cards in certain games, such as the RX 550, which surpasses it in DOOM Eternal, and the Radeon HD 7970, a contemporary of its generation that cost much less than that, doubles it in performance.
Saying that the 20 series RTX are going to age as badly as Kepler is, as we see, a very serious statement and with many implications that we must analyze. First of all, it must be made clear that refers only to ray tracing performanceThat is, the RTX 20 series will age as poorly as Kepler but only in ray tracing workloads.
It makes sense, we’ve already seen that Turing is not capable of optimally moving current games with limited ray tracing, furthermore, to a single effect unless complemented by DLSS 2.0, a smart rescaling that renders at a lower resolution than native that we have selected. If this is confirmed, the GeForce RTX series 20 will age poorly in ray tracing, although its performance without such technology in DirectX 12 and Vulkan It will still be very good.
RTX 20 series vs. RTX 30 series: higher IPC, more memory and higher overall performance
The source of this information, MyDrivers, has also collected several slides in which we can see some of the supposed keys to Ampere architecture. In them we see very important and very interesting things about the GeForce RTX 30 series, among which we can highlight:
Ampere will be a big evolution against Turing, and it will come manufactured in 8nm and 7nm process.
It will have more shaders, more tensor and RT cores (double versus Turing) and higher frequencies (1,900 MHz minimum).
DLSS 3.0 technology will be the great standard bearer for this generation, and it will work natively in games that use TAA.
Improvements at the CPI level that will be between 10% and 20%. It will double the L2 cache and improve cache and VRAM latency.
The RTX 3060 will outperform the RTX 2080 Ti in ray tracing, but not in rasterization and shading. It will not be, strictly speaking, more powerful.
Goodbye to the GTX series, even low-end graphics cards will have tensor and RT cores.
The RTX 3080 Ti will be able to move games in 8K resolution with ray tracing.
Graphics memory is increased, going to 10 GB on the RTX 3080 and 12 GB on the RTX 3080 Ti.
Ampere architecture is expected to be presented on May 14 taking advantage of the GTC 2020 online celebration, but this will only be a first step. NVIDIA still has a lot of stock of the 20 series RTX, so we will first see the debut of HPC solutions in May. The presentation of consumer graphics cards will not take place until September, and could be a paper release, that is, without actual availability until End of the year.
The 20 series RTX are not going to disappear overnight, that is very clear, in fact it is possible that NVIDIA maintains it as a “cheap” alternative to the 30 series RTX. Keep in mind that a technological leap like the one who promises to score Ampere could serve as “Excuse” to raise prices to a level never seen before, which would make the 20 series RTX continue to make some sense.