Quote from: Bogdan Solca on September 10, 2021, 01:55:38
I actually thought of using MT/s but read somewhere that neither that nor MHz is correct. I can't remember what alternative was offered, seemed like the WCCFTech guys read the same thing and figured I should quote them on that. Now I'm not sure they got it right, so instead of messing things up, I'll just go ahead and use DDR-xxxx.
Technically speaking, hertz is the correct unit. Since you are talking about temporal frequency of occurrence of a certain event. Hertz would also be the correct unit for engine speed rather than "rpm" (1 Hz = 60 rpm). The problem is that it could potentially cause confusion. People are not used to it. And like bell, it can be applied to many different quantities. Meaning, you can't infer what the quantity is just from reading the unit. When I tell you "2 metres," you also don't know whether it's length, width, height, radius, diameter; or whether I'm not talking about wavelength. You need more information.
The alternative could have been baud which is unit of symbol rate from the telecommunications world. The complication with DDR is that it signals both at rising and falling edges of a clock signal. Hence the name (Double Data Rate). And so the symbol rate for a single lane is going to be double the clock frequency (and it's a parallel bus, it sends multiple symbols simultaneously and so you'd need to account for bus width).
Looking at UserBenchmark, the problem isn't that they use hertz; the problem is that they say "clocked at 4400 MHz." The clock is not running at 4400 MHz. DDR5-5600 has a 2.8 GHz clock rate and 5.6 GHz transfer rate. It's not the unit, it's the expression. Together with hertz, it implies it's a clock frequency. Only the magnitude of that number tells you that it can't be clock. Had they used transfers instead of hertz, it would have been clearer what they actually mean. It's an unfortunate case of their inability to express themselves properly.