OSDev.org

I read on the internet the other day that there is a difference in calculation of the speed of each core of the processor in AMD and Intel.
In Intel, each core runs at the speed that is shown in the system properties. i.e. Each core in an Intel Core 2 Duo 2.93GHz will run at 2.93GHz.
In AMD, the total speed is achieved by multiplying the speed of each core with the number of cores. i.e. Total Speed = Number of cores * Speed of each core.
In other words, the cores of AMD processor runs slower than what is displayed in system properties.

My question is, is this saying about AMD CPUs true?

Hi,

trinopoty wrote:My question is, is this saying about AMD CPUs true?

No. If Windows says "2.80 GHz" then each core should be running at a nominal clock speed of 2.8 GHz.

Note: "Nominal" means that, due to power management the CPU may run slower than the stated speed, and due to things like TurboBoost the CPU may actually run faster than the stated speed.

Also note that (as far as I can tell) Windows is a stupid piece of poo that only displays whatever the CPU's "CPUID" tells it to and doesn't actually measure anything. Once upon a time AMD (and some other CPU manufacturers) did use a non-standard definition of "speed" in their model numbers, etc. For example, a (factitious) "AMD K5 123" might be running at 100 MHz but might be rated at the equivalent performance (for some unknown definition of "performance") of an Intel Pentium running at 123 MHz. You might see Windows report this as part of the brand name instead of the (mostly irrelevant) clock speed.

Finally, for some AMD CPU's there's a set of MSRs that can be used to set the brand string reported by CPUID. This means that (e.g. with some boot code that sets these MSRs then chainloads Windows) it's easy to trick Windows into displaying anything you like; and Windows might show "58.6 GHz Z80" or "Intel Xeon" or "4 Hz ARMv8" or anything else you feel like.


Cheers,

Brendan

So what you are saying is that in now if an AMD processor says 2.6GHz, each core will run at 2.6GHz(ignoring power saving and speed boosting functions).

As Brendan suggested, You're probably mixing the brand name with the actual CPU specification.

AMD used to list the clock frequency an older processor needed to run to match its speed. An Athlon XP 2200+ would for instance run at 1.7GHz, but would need a 2.2GHz Thunderbird to break even on processing power. And of course, the moment you try that with multicores, you will need to double the clock on a single core for the same instructions per second. I think the latest practice includes the core count as an actual part of the model number.


There's also the problem of putting a processor on an inferior motherboard, you'll have an underclocked bus from the CPU's perspective. And since the CPU typically runs at a multiple of the bus, you get a lower CPU frequency even if it was rated higher.


At any rate, A tool like CPU-Z is a good way around the M$-induced reporting bugs. Use that to find out what you truly have.

[edit]Fixed as per Owen[/edit]

I do not have an AMD machine but I plan on switching to AMD from Intel as it provides better performance.
What I have gathered by searching on the web is that I can continue to use the same definition for clock speed, "the number of clock ticks produced per second in a single core".

Doing more research on google, I found that the performance provided by AMD processors is decreasing and its best to stick with Intel. Is that really the case?

Hi,

Clock speed has very little to do with performance. For example, is a 2 GHz CPU that executes an average of one instruction per cycle better or worse than a 1 GHz CPU that executes an average of 2 instructions per cycle? Is a 2 GHz CPU with small/slow caches and a crappy bus interface better or worse than a 1 GHz CPU with large/fast caches and a very fast bus interface? Is a 2 GHz CPU that constantly overheats and frequently drops back to 250 MHz better or worse than a 1 GHz CPU that never overheats and frequently gets "turbo boosted" up to 1.5 GHz?

This is actually the reason why AMD decided to use "performance ratings" in their marketing instead of raw clock speeds - people were comparing clock speeds alone (which are misleading at best), so AMD (and Cryix?) tried to give people something less misleading to compare.

trinopoty wrote:Doing more research on google, I found that the performance provided by AMD processors is decreasing and its best to stick with Intel. Is that really the case?

Sounds extremely misguided.

In a *fair* comparison (e.g. not comparing a new low power mobile CPU with an older "150 watt" server CPU or something) I seriously doubt AMDs recent CPUs have worse performance than AMD's previous CPUs.

To correctly compare CPUs, forget about clock speeds, and also don't compare a $2000 Intel CPU with a $800 AMD CPU. Instead you want to compare "MIPS per dollar" (and "MIPS per watt" too if it's for a laptop or something).

Of course I don't do this. I normally look at what features the CPU supports (e.g. AVX, FMA4, etc) that my existing systems don't already have; then I buy the cheapest (slowest) CPU that has those features. Despite this "lack of emphasis on performance", I still have no problem regretting the time I've wasted playing computer games.


Cheers,

Brendan

Combuster wrote:AMD used to list the clock frequency an Intel processor needed to run to match its speed. An Athlon XP 2200+ would for instance run at 1.7GHz, but would at least need a 2.2GHz pentium to break even on processing power. And of course, the moment you try that with multicores, you will need to double the clock on a single core for the same instructions per second. I think the latest practice includes the core count as an actual part of the model number

AMD's old PR rating system (Used from the Athlon XP up to the Athlon 64s) was based upon a performance comparison with a 1GHz Athlon Thunderbird (which was defined as 1000 on that scale) on a specific benchmark suite