stuff and nonesense
Honorable
There are chips which hit higher frequencies within the same architectures, CPUs are binned to identify these chips.Are there cpus that due to their architecture can hit higher clockspeeds than other cpus with less voltage? Eg. Zen 4 can hit ~2ghz extra clocks at similar voltages to zen 1. Therefore a clockspeed comparison does not help you to determine efficiency. That's all im saying.
Your Zen 1 to Zen 4 comparison holds, and taking it at face value, Zen 1 can clock at 4.1GHz, single thread I see 5.8 on my Zen 5 chip. 1.7GHz extra clock speed. You say it does this at similar voltages so let’s run with this.
Each generation has increased IPC, by how much doesn’t matter. I will assume 10%. The chip with the the higher IPC in this simplified example does 1.1 times more work at the same clock frequency. Engineers attempt to maintain or lower the voltages and power envelope for the parts. In this dumbed down example, assuming the frequency and voltages are the same the newer chip is 10% more efficient. It gets 10% more work done at the same power. Power x time is energy. In the same time it can get 10% more work done or it takes 10% less time to do the same thing for a given set of voltage/current/clock speed.
Smaller transistors need fewer electrons to trigger the cmos triggering event, less voltage this reduces power, 14nm to 4nm is approximately 3 so I’ll use 3. Applying a voltage to the gate of the cmos transistor requires approximately 1/3 of the electrons required to saturate the gap between the drain and the source. No current flows between the gate and the transistor substrate, an electric field energises the substrate, connecting the source and the drain (where the current flows) same in both the 14nm and 4nm, but less power is needed - everything is smaller, (read up on field effect transistors).
The lower power requirements of the smaller components offsets the increased frequency to maintain the power envelope (previous post, look at the eetimes article, dynamic power section. It even has the maths….)
The retail chips we buy are in very general terms the worst of the silicon produced. Server parts and “professional” parts are, where designs are common across professional and retail, better binned chips. Same silicon, different quality. LTT compared a number of 7000 series chips
No chips are identical, there is variation across individual wafers, there is variation between wafers. There are performance bins in which each manufactured chip will fit. A poor chip for a given bin will require more voltage to maintain a frequency, a good one will require less. If you are lucky then you get a part which is close to the upper end of the performance bin for the part you buy… a 7700x that performs closer to a 7800x instead of an imaginary 8 core 7600x.
Going back, efficiency is energy used to complete a task. From the previous post :
Energy is power over time e = pt joules
Power in the fet application is given from Capacitance, voltage, frequency and probability of a state change (look at the article)
Capacitance and the probability of a state change are constant for a given application within the designed specification, voltage and clock speed change. Clock speed directly effects efficiency. Voltage directly effects frequency and power.
if you can find it look at the I486dx 25, 33, 50, dx2 50, 66 you would be looking for the data sheets. The dx2 parts might have a lower IPC due to being clock doubled,
The dx 50 was on a 50MHz memory bus and may show slightly higher IPC than the dx2 50.
Look at their power consumption, look across the range, the dx chips came without heatsinks, the dx50 was marginal, the dx2 66 needed a heatsink…. More power was being consumed.
Do some research for yourself. It will help you form a cogent argument.