Falcosoft wrote on 2025-06-17, 17:03:

Hi, I'm sorry to ask this, but would you repeat the Core i5 tests with the correct 1 core turbo speed corrected? According to th […]
Show full quote

Hi,
I'm sorry to ask this, but would you repeat the Core i5 tests with the correct 1 core turbo speed corrected?
According to this Intel page the correct 1 core max turbo is 2700 MHz.
https://www.intel.com/content/www/us/en/produ … ifications.html

I did not change the turbo speed manually but used the default values your program provided. Is your program capable of changing the turbo speed?

Start me up wrote on 2025-06-17, 17:10:

Falcosoft wrote on 2025-06-17, 17:03:

Hi, I'm sorry to ask this, but would you repeat the Core i5 tests with the correct 1 core turbo speed corrected? According to th […]
Show full quote

Hi,
I'm sorry to ask this, but would you repeat the Core i5 tests with the correct 1 core turbo speed corrected?
According to this Intel page the correct 1 core max turbo is 2700 MHz.
https://www.intel.com/content/www/us/en/produ … ifications.html

I did not change the turbo speed manually but used the default values your program provided. Is your program capable of changing the turbo speed?

No. It has been written in the 1st post:

It's important to note that I could not find a generic and reliable way to determine the actual 1 core turbo frequencies of diff […]
Show full quote

It's important to note that I could not find a generic and reliable way to determine the actual 1 core turbo frequencies of different modern processors. I could only detect the CPU speed by using the TSC which usually only gives back the base speed. So you have to manually type the proper 1 core turbo frequency of your CPU before the benchmark to get one of most important benchmarks results:
That is the 1GHz normalized pixels/millisecond value.
With this value you can easily compare the execution efficiency of different generations of CPUs with different working speed/MHz values.
So please, use CPU-Z (https://www.cpuid.com/softwares/cpu-z.html) or read your CPU's datasheet to determine the correct 1 core turbo frequency value.

But it seems we have a bigger problem:
According to these results the 1 core speed was definitely not 2700 MHz. I would say it was not even 1800 MHz. Your Core I5 CPU must have run at a lower P-state...

Can you run CPU-Z to determine the actual speed of your CPU while the benchmark is running?

Falcosoft wrote:

But it seems we have a bigger problem:
According to these results the 1 core speed was definitely not 2700 MHz. I would say it was not even 18000 Mhz. Your Core I5 CPU must have run at a lower P-state...

Intel Speed Step is deactivated in the BIOS so I guess that the CPU is running at a constant speed of 1.8 GHz.

Falcosoft wrote on 2025-06-17, 17:12:

Can you run CPU-Z to determine the actual speed of your CPU while the benchmark is running?

I am sorry, but I don't have CPU-Z installed.

I executed the "read time stamp counter"-instruction, waited about 100 seconds and then executed it again. The value increased by 179 634 879 342.

Start me up wrote on 2025-06-17, 17:17:

Falcosoft wrote:

But it seems we have a bigger problem:
According to these results the 1 core speed was definitely not 2700 MHz. I would say it was not even 18000 Mhz. Your Core I5 CPU must have run at a lower P-state...

Intel Speed Step is deactivated in the BIOS so I guess that the CPU is running at a constant speed of 1.8 GHz.

Hi,
I do not think so here is the result of the same 3rd gen I7 at 1800 MHz (set by ThrottleStop). The results are more than 2x better than yours (actually 1800/800 = 2.25 better). According to this your CPU must be running at about 800 MHz. (That is the minimum allowed by 3rd gen chips).

Start me up wrote on 2025-06-17, 17:24:

Falcosoft wrote on 2025-06-17, 17:12:

Can you run CPU-Z to determine the actual speed of your CPU while the benchmark is running?

I am sorry, but I don't have CPU-Z installed.

I executed the "read time stamp counter"-instruction, waited about 100 seconds and then executed it again. The value increased by 179 634 879 342.

Intel Core CPUs have an invariant TSC. That is the TSC value does not change with lower P-state/Speedstep values.
This way the operating system can use TSC as a reliable timer source (it does not have to worry about timer anomalies).
So the TSC always reports the base clock even if your CPU using boost for higher clocks or low power states for energy saving.

Falcosoft wrote on 2025-06-17, 17:26:

According to this your CPU must be running at about 800 MHz. (That is the minimum allowed by 3rd gen chips).

No, I don't think so. Your program uses the read time stamp counter instruction to read the speed and it got about 1.8 GHz. Also, my manual test with the 100 seconds (see previous post) resulted in the speed of about 1.8 GHz.

It is possible that the processor simply cannot process faster if the operating system is set to support only a single core on a multi core system. I heared rumors that the other 3 cores were running at 100% load. But I can't validate these rumors.

Falcosoft wrote on 2025-06-17, 17:32:

Intel Core CPUs have an invariant TSC. That is the TSC value does not change with lower P-state/Speedstep values.
This way the operating system can use TSC as a reliable timer source (it does not have to worry about timer anomalies).
So the TSC always reports the base clock even if your CPU using boost for higher clocks or low power states for energy saving.

Are you sure? That doesn't make much sense.

Start me up wrote on 2025-06-17, 17:36:

Falcosoft wrote on 2025-06-17, 17:26:

According to this your CPU must be running at about 800 MHz. (That is the minimum allowed by 3rd gen chips).

No, I don't think so. Your program uses the read time stamp counter instruction to read the speed and it got about 1.8 GHz. Also, my manual test with the 100 seconds (see previous post) resulted in the speed of about 1.8 GHz.

It is possible that the processor simply cannot process faster if the operating system is set to support only a single core on a multi core system. I heared rumors that the other 3 cores were running at 100% load. But I can't validate these rumors.

As I said the TSC cannot be trusted to get the actual working frequency in case of Intel Core CPU's. I'm 100% sure it is running below 1 GHz. There is an older version of throttleStop that maybe can run on Win 2000.
I try to find it. Then you should try it.

Are you sure? That doesn't make much sense.

Yes , it is 100% sure. There is even a CPUID feature bit introduced to this new 'Invariant TSC' feature:
https://community.intel.com/t5/Intel-ISA-Exte … port/m-p/772125

@Edit:
I have found the old version of ThrottleStop that is working with my WinXP. Maybe it can run on Win2000 also. You should check the FiD values.
FID x 100 is your actual frequency in MHz. I suspect your FID is fixed at 8.

Start me up wrote on 2025-06-17, 16:54:

Here are my results of an Intel Atom and an Intel Core running at slightly different clock rates. The screenshots include the measurement readings for the power consumption.

I have uploaded your Intel Atom result since it seems to be correct.

I think you are right about the TSC. But the manual claims, that it's not just the Intel Core, but also the Intel Atom that returns phantasy values.

I will test the program, thank you.

Here are the results from ThrottelStop.

Regarding the results of your MandelX program: I would say both results are valid. The result from the Atom and the result from the Core are valid. It's neither my fault, nor your fault, nor the fault of the program MandelX that Intel sold a processor with 4 cores @ 1.8 Ghz but in reality only 1 core works at 0.8 GHz. There is no good reason for the phantasy values of the RDTSC instruction nor the incompatibility. So an Intel Core is just a slow and power hungry processor even if it could perform better if Intel delivered a better product.

Start me up wrote on 2025-06-17, 19:01:

Here are the results from ThrottelStop.

Regarding the results of your MandelX program: I would say both results are valid. The result from the Atom and the result from the Core are valid. It's neither my fault, nor your fault, nor the fault of the program MandelX that Intel sold a processor with 4 cores @ 1.8 Ghz but in reality only 1 core works at 0.8 GHz. There is no good reason for the phantasy values of the RDTSC instruction nor the incompatibility. So an Intel Core is just a slow and power hungry processor even if it could perform better if Intel delivered a better product.

Just as I suspected: Your Core I5 currently works at ~800 MHz. But do not worry. With the help of ThrottleStop there is a chance you can increase the clock of your processor to its nominal value:
1. Check the checkbox next to 'Set Multiplier'.
2. Change the multiplier value to 18. (Check if the FID value changes or not.)
3. Press Save.
4. Rerun the benchmark.

Falcosoft wrote on 2025-06-17, 19:08:

Just as I suspected: Your Core I5 currently works at ~800 MHz. But do not worry. With the help of ThrottleStop there is a chance you can increase the clock of your processor to its nominal value:
1. Check the checkbox next to 'Set Multiplier'.
2. Change the multiplier value to 18. (Check if the FID value changes or not.)

Yes it does and after doing so, the bench mark MandelX returns higher results. This is interresting. So this processor needs a driver of some sort to work correctly.

I think Windows XP introduced a processor driver.

Start me up wrote on 2025-06-17, 19:27:

Falcosoft wrote on 2025-06-17, 19:08:

Just as I suspected: Your Core I5 currently works at ~800 MHz. But do not worry. With the help of ThrottleStop there is a chance you can increase the clock of your processor to its nominal value:
1. Check the checkbox next to 'Set Multiplier'.
2. Change the multiplier value to 18. (Check if the FID value changes or not.)

Yes it does and after doing so, the bench mark MandelX returns higher results. This is interresting. So this processor needs a driver of some sort to work correctly.

I think Windows XP introduced a processor driver.

Great news! So overall MandelX helped to find and solve a CPU clock related anomaly 😀
I can say from my own experiences that 3rd gen Ivy Bridge works perfectly with Win XP SP3. Only drawback is that AVX is not available/supported but there is no XP era software that would use it anyway...

Well, I learned quite something from this conversation. I guess that you only got some confirmation for what you already knew. So thank you for sharing your knowledge.

Even though it is tempting to dig deeper and start working on a processor driver for Windows 2000, that's just something I can't do at the moment because my todo list is pretty full at the moment. However, knowing what's going on is at least a start.

Here is some more data for the Kraken.

Why did you spend so much time into programming this bench mark? Are you working on something specific that needs top performance?

Mine from an AMD K6-2+ 500MHz and a Ryzen 5 2400G (Zen 1).

Start me up wrote on Yesterday, 00:22:

...
Why did you spend so much time into programming this bench mark? Are you working on something specific that needs top performance?

1. It was not that much time considering that MandelX is an old program of mine (as a Mandelbrot/Julia fractal generator). The 1st lines of code was written more than 20 years ago. I have just added the benchmark capabilities recently.

2. Nothing special, I'm just interested in the history and evolution of x86 micro-architectures. It's interesting to reveal the weak and strong points of different architectures and notice the trends in CPU generations (what parts they focus on and what parts are neglected).
Traditional modern benchmarks usually follow a different philosophy and so cannot help in such investigations. Even if they implement different code paths for different architectures they usually use a dynamic dispatcher that always selects the best/fastest code path. Of course this approach also makes sense but it does not allow such real apples to apples comparisons that MandelX benchmark does. Also benchmarks that can run on both old and new CPUs as well as work from Win9x to Win11 (and can provide meaningful results) are rare.
E.g. the trend of latest Intel/AMD CPUs to de-emphasize traditional x87 FPU performance , as far as I know, has not been revealed so far explicitly.
It's also interesting to notice how much potential was left unused in K6-2/3 considering its exceptional 3DNow! performance.
And building a public a database about these results maybe can help someone to make further such discoveries.

jtchip wrote on Yesterday, 00:30:

Mine from an AMD K6-2+ 500MHz and a Ryzen 5 2400G (Zen 1).

Thanks, the results are uploaded.
It would be interesting to see the results of older, rarer CPUs such as later Cyrix/VIA chips. Or even IDT WinChip and Transmeta Crusoe 😀