The first two RX 7000s with RDNA 3 structure have been launched, so it have to be the very best time to launch rumors of the subsequent structure, RDNA 4. An structure that’s no less than two years away, so any factor mentioned now has a excessive likelihood of being false. Be that as it could, the channel’s announcer RedGamingTech YouTube has claimed to have details about this structure. AMD would be centered on creating a chip with a computing energy that would attain 129 TFLOPS.
The present Navi 31 chip has a computing energy of up to about 62 TFLOPS, in order that would be double the efficiency. In addition, it ensures that the working frequency would be up to 3.5 GHz in contrast to the present 2.5 GHz, so the rumor is starting to indigest me. The Navi 41 chip would have 144 compute items for a whole of 18,432 shaders, in contrast to 12,288 for the Navi 31, however that quantity “is very preliminary.” The Navi 42 would have up to 12,288 shaders, and the Navi 43 up to 6,144.
Those frequencies are absurd in a extremely parallel structure resembling a GPU; It has already been seen within the RTX 40 and RX 7000 that if the utmost frequency is proscribed, efficiency is hardly misplaced however consumption can plummet. With a forecast of utilizing TSMC’s 3nm lithographic course of, I see it as an unlikely rumor until it made a graphics card that exceeded 500W, or AMD did some miracle. -Y it hasn’t executed them with RDNA 3, as a result of even the RTX 4080 with efficiency comparable to the 7900 XTX consumes less-. He mentions a efficiency per watt enchancment of 50-60%, as within the earlier two generations, which would put a new structure 7900 XTX consuming about 235 watts. To give twice the efficiency it would have to go to 470 W. If it achieves this by considerably rising the frequency, it might go above -the frequency-consumption relationship is exponential-.
It is alleged that it would use GDDR7, and it’s the solely a part of the rumor that’s half credible. This kind of reminiscence is in a sophisticated state of growth, so it’s a matter of months earlier than corporations announce their first chips. Probably in direction of the tip of 2023, and the JEDEC standardization would arrive in 2024, which would be when the “RX 8000” will probably be launched, if there aren’t any delays. -what I see possible that the is-.
The remainder of the issues he says I might say proper now: additional ray tracing enchancment, 128 shaders per compute unit which would keep the identical, cache enhancements -What everybody within the trade is doing, which is problematic on the lithography degree—, enhancements to knowledge preloading, there could possibly be adjustments to the way in which to execute arrays (AI) gaining twice the efficiency, and basically “a lot of small and big changes”. There is nothing actually related in all the pieces mentioned within the video.