CPU-Z Multi Thread Ratio greater than number of threads

[Shrek]

I have a Core 2 Quad (no multi-threading), so was wondering why CPU-Z reports

Multi Thread Ratio: 4.21

 

[BlaezaLite]

Punching above its weight?

 

[GerKNG]

maybe yorkfield scales better in multicore than single core (it has a lot of cache)

 

[Shrek]

Maybe it first tests the single core, which loads the cache, and that runs in favor of the multi-core test? Pure speculation on my part.

 

[eidairaman1]

Ask CPUID about their measuring algorithms?

 

[Shrek]

Great idea: done

 

[ratirt]

I'm not sure what seems to be the problem here. Single thread equals one core. (I think these quad cores don't have SM?). Multi-core utilizes all 4 cores this higher score. Or is that not what you are asking?

 

[Shrek]

I did not expect 4 cores to be more than 4 times as fast as 1 core, but

4.21 > 4

(The Core 2 Quad has no multi-threading)

 

[mx62]

maybe some problem in awakening the cpu?
or in single core its using one that was doing something?

 

[Sabotaged_Enigma]

I've seen some old Intel CPUs did the same thing. One was a 2c/2t G3258 and it did 2.01 just slightly greater than 2 threads. I reckon it has something to do with mainboard BIOS settings, such as muti-core enhancement or similar features.
Or the BIOS is not driving the CPU correctly. I once got an A10-7890K but when it was published it was nearly the very end of FM2+ platform, so BIOS wasn't carefully made, and the CPU thus performed badly, single-thread just sucked, but again something it was normal.
But your 4.21... is really confusing...

 

[Nike_486DX]

must be just issue with cpuz, its testing algorithm is crooked (so to speak lol), on my 7900X i got 6500-ish on the first pass, then on the second pass it was down to 5150 (wtf), then came back later and it did 6200. That is @4.3ghz 1.1v, all cores staying below 60C, and no background activities, all other benchmarks (cinebench, geekbench, passmark) show consistent results. I only see this in cpuz v17 bench, those that are marked with v19 actually dont show that issue.

Try the 19.x (beta) benchmark version. And also cinebench. To see if its just cpuz being weird.

 

[Vya Domus]

A CPU core can execute multiple instructions at once, usually having one thread is a waste of execution resources, that's why basically all desktop CPUs these days allow the execution of 2 streams of instructions simultaneously per core (SMT/HT).

Core 2 didn't have HT, it's possible that having the OS switch between more than 1 thread allows for more instruction parallelism hence improving the scaling. But who knows, CPU-Z is a horrible benchmark, it's way too weak and I suspect it measure integer performance mostly, thus making it a terrible benchmark to run on just 1 thread. So it probably is a case of simply not utilizing the hardware enough.

Another explanation could be memory, often you can get more bandwidth from using multiple threads, so if the tests are memory bound it may cause the same kind of strange scaling.

BTW I've just tired to run only 4 threads on my 1700X and it also scales more than 4X but in this case it makes somewhat more sense since this CPU does have SMT.

 

[Dr. Coonan]

Works like this. Turbo boost allows for a single core to raise it's frequency higher than normal until it hits a thermal or voltage wall. All 4 cores of your Core2Quad can do this in succession as the bench proceeds. This is how it can perform better than the base frequency gives you for all core.

 

[R-T-B]

Works like this. Turbo boost allows for a single core to raise it's frequency higher than normal until it hits a thermal or voltage wall. All 4 cores of your Core2Quad can do this in succession as the bench proceeds. This is how it can perform better than the base frequency gives you for all core.

That doesn't make any sense. You don't run a multithreaded bench as "single process in succession" you run them all at once on every core.

 

[Bones]

While I agree with the above there is at least one (Futuremark CPU Max) that kinda does that - But it gradually decreases threads used during the run to obtain a result.
Goes from max (16) then to 8, 4, 2 and finally 1 at the end.

I can't think of one that gradually increases threads used during a run but it's possible.

Works like this. Turbo boost allows for a single core to raise it's frequency higher than normal until it hits a thermal or voltage wall. All 4 cores of your Core2Quad can do this in succession as the bench proceeds. This is how it can perform better than the base frequency gives you for all core.

No, that's not how it works.

The results in CPU-Z are based on a given result/standard and the results you get are compared to it.
The result 4.21 means their chip/setup performed .21 better than the "Sample" Q9550 chip used as a standard for the comparison in CPU-Z.

 

[phanbuey]

While I agree with the above there is at least one (Futuremark CPU Max) that kinda does that - But it gradually decreases threads used during the run to obtain a result.
Goes from max (16) then to 8, 4, 2 and finally 1 at the end.

I can't think of one that gradually increases threads used during a run but it's possible.



No, that's not how it works.

The results in CPU-Z are based on a given result/standard and the results you get are compared to it.
The result 4.21 means their chip/setup performed .21 better than the "Sample" Q9550 chip used as a standard for the comparison in CPU-Z.

955.7/227.1 = 4.21 - it's just a MT straight ratio, there is no comparing to anything. The reference processor score is a separate thing.

 

[Bones]

Actually that is the comparison as you put it.
It's simply comparing a result vs another chip used as a baseline and expressing the results as a vs ratio.