Do you still measure FSB in the hundreds?

Many people have been doing just that for the last 15-20 years. A hundred or more.



What is Front Side Bus (FSB, system bus)?

A bus that provides a connection between an x86 / x86-64-compatible CPU and internal devices. Its base frequency is used, with a multiplier, by the processor.



The entire toolkit (I'm familiar with HwInfo64 and CPU-Z) is geared towards this (hundreds). But now, I have a processor on which I see the bus frequency of 25 MHz.



image




HwInfo64
image



And everything seems to be honest, a couple of years ago they switched to AMD ( Precision Boos t), they are more convenient to set the upper limit of the operating frequency for the CPU. But, nevertheless, all (HwInfo64 and CPU-Z) continue to show the operating frequency based on 100 MHz ?! Therefore, we see a very suspicious CPU operating frequency. With the declared 1500-1000MHz, the processor works strangely at 400-600MHz. Prokolchik.



image



CPU-Z
image



Moreover, this factor also affects the frequency of memory operation, at least its parameters displayed in tools.



image



Of course, thoughts immediately flew towards ProcHot and ThermalThrottling. But no, they were fine.



Usually the frequency multipliers are integer, the maximum that could be achieved is 0.5. This is where I noticed that the multiplier is somehow suspiciously fractional in CPU-Z. We see and .2, and .3, and .4. This will not fit into the control register. This means that this multiplier is not real, but "synthetic". It is not read from iron, but "magically" is calculated and adjusted to all of us any 100 MHz. It's sad. Indeed, in the documentation, all multiplier values ​​for AMD are integer, and are based on 25 MHz. And the multipliers there are much higher than those attached to a hundred. Both 90 and 130 are found there. And this is not a ceiling at all.



Those wishing to delve into the MSR registers
MSR

β€” C001006[4-B] [7:0]

β€” C0010063 [2:0]

β€” C0010293 [7:0] [13:8]



All right, these "fake" operating frequency numbers were only on my hardware. But no, they also get out on reference boards (Bilby) from AMD. And the release date of these processors on the market was the first quarter of 2020.



Moreover, performance tests do not show any drawdown. The operating frequency is as stated 1500-1000MHz.



The mess with frequencies is complemented by the fact that the processor settings contain all the frequencies of 25 and 100 MHz. And even a little more))). So, for example, for REFCLK there is also a frequency of 27 MHz. Moreover, it is declared as active at the moment after RESET. However, all the time during the execution of UEFI, 25MHz is active. But the focus, in Windows, we see again in registers ... 27MHz!



Although, documentation from AMD, this is a completely different story. AMD never had enough time for it. We have what we have. And we rejoice in that.



So what was the article about? Believe it or not, I want a toolkit that shows the truth about hardware. So I'm complaining about life. Although with such documentation, which pleases processor designers, it will take more than one year to wait for it. Or maybe you know such a toolkit? Then share your links in the comments!



PS: and by the way, can anyone know why the bus frequency is shown as "floating"? In what place it is read or on the basis of what it is "synthetically considered". After all, it does not cost 100 or 25 MHz, but floats up to -3%, but this is not SpreadSpectrum either. SpreadSpectrum is solely responsible for "external" devices.



All Articles