Smart Memory could then be as efficient as cross-CPU EPYC memory access. Which is pretty good.
totally incompatible: https://www.itworldcanada.com/article/amd-zen-3-processors-w...
erm, we means totally compatible!!1: https://www.tomshardware.com/news/amd-reverses-course-will-e...
"chipset" nowadays is just a pcie device providing legacy ports (USB 2.0, additional USB 3, SATA, HDA audio controller, PCIe bridge, I2C, SPI, LPC, bunch of GPIOs).
Makes me wonder if that's related to the improved L3 cache access of Zen 3?
Leaked 3DMark raytracing benchmarks showed Big Navi lagging far behind Ampere so I wonder how that's going to bear out in real games.
Nvidia did that by enabling DLSS and RTX in everything, that's how they ended up with "Up to 2 x 2080 performance" which in practice only seems to be the case with Mincecraft RTX running on 30 fps instead of 15 fps.
I mean, I get that you can’t enable these on AMD 6000 series because the card doesn’t support these well.
Not every game supports RTX, those that fully rely on it are still performing so badly that a 15 fps increase can be marketed as "double 2080 performance", while saying absolutely nothing about the much more relevant rasterization performance.
Using DLSS can be misleading as picture quality is very difficult to objectively compare. But common sense makes the idea of "More performance with more details!" just sounds too good to be true, even with ML magic involved. Reminds me much more about something like Temporal Filtering: Giving more performance, but a slight cost of picture quality that might not be too noticeable with the extra raw pixels trough upscaling.
I.e. just because games on the consoles can support ray tracing now doesn't mean they actually support it well. I still think it's a generation or 2 away from running well (as a current 2080 Ti owner)
I wouldn't be that worried if I were nvidia-- catering to the whales is good business, but I think we're looking at amd winning the lion share of the market