Geekbench Blog
www.geekbench.com
Weird choice of baseline CPU and even weird is that the baseline score is 2500.
i7-12700 does hardly 2000 in GB5 with the fastest DDR5.
What about the HT variable? This clearly shows: "1 Processor, 8 Cores, 12 Threads"
Of course it does, it's a 3 year old chip. The question is why doesn't it get trounced in the mt as well, since the freaking m4 is like what, 4-5 times the transistor count?What about the HT variable? This clearly shows: "1 Processor, 8 Cores, 12 Threads"
Anyway this gets trounced by M4 in ST even ignoring the SME subtests.
Perhaps because it behaves as a 12-core CPU vs a 10-core CPU?Of course it does, it's a 3 year old chip. The question is why doesn't it get trounced in the mt as well, since the freaking m4 is like what, 4-5 times the transistor count?
Why do the number of cores matter? Honestly that's just silly. Is that how you compare gpus?Perhaps because it behaves as a 12-core CPU vs a 10-core CPU?
BTW where do you get 4-5x the transistor count? Don't tell me you pick the whole SoC transistor count?
Do we have some calculations on the size of the cores?
So 1 m4 p core should be 40+% faster than a zen 4 core? Am I getting this right?View attachment 102431
One P-Core is about 2.97mm2
One L2 Cache block is 2.24mm2
If we use 1 P-Core + 1/2 L2-Cache as reference size that is around 4.08mm2 (just for reference!)
It's using old 6.2.2 version.
Sure once I'm home I'll check it out. But why HT off?It's using old 6.2.2 version.
Could you CMD into the Geekbench 6.3.0 folder and run all three in the below screenshot with 4+4 (HT disabled) and share the three different result URLs?
View attachment 102433
Just type their names in CMD and run. Don't double-click them otherwise it will run and then the CMD window will disappear.
Because Apple doesn't have SMT or HT. And also, with HT off, you will get even lower power usage, with a slight hit in the MT score.But why HT off?
Sure. But tbf personally I'd like to know how many transistors are the 4+6 and then compare the equivalent amount of transistors enabled.Because Apple doesn't have SMT or HT. And also, with HT off, you will get even lower power usage, with a slight hit in the MT score.
Or if you want to go the distance and share six URLs (3 with HT and 3 without HT), even better!
HT on or off doesn't matter to me much but turning it off can reveal a power efficiency improvement for the 12900K, if that's what you are really interested in. Otherwise, up to you.Turning HT off is silly cause it takes space on the core, apple not having it allows them to have a bigger core or more cores into the same space.
Why do the number of cores matter? Honestly that's just silly. Is that how you compare gpus?
If one core is made up of 10 times the transistors then it should be 10 times faster. If it isn't then that's an issue.
Yes, I'm comparing the whole soc for both. The m4 is 2.5 times the transistor count of a 12900k if I remember correctly. Since I'm basically using half the cpu, the difference is bigger.
If you have some actual data (die shot) of the m4 then we could sort this out. If not then this is all meaningless, at the end of the day a 3 year old cpu made on 14nm and drawing an average of 19w is around 40% faster than the m4.
So unless the cores of the m4 are super tiny in terms of transistor count, the results look disappointing to me.
I know all that bug peak power isn't indicative either, since it peaks at 150w for 1 or 2 workloads and for the rest it's sitting at or below 60 watts. Even for the mt test. It's really not particularly threadedYour average power measurement isn’t very useful as each GB sub test completes very quickly on modern CPUs and has a relatively long rest period between tests. In other words, your average measurement is being greatly influenced by idle power. You need to compare peak power if you want to compare to the power that has been reported for M3/4.
I know all that bug peak power isn't indicative either, since it peaks at 150w for 1 or 2 workloads and for the rest it's sitting at or below 60 watts. Even for the mt test. It's really not particularly threaded
But you can't use peak either like you just suggested, right?That’s because only 1 sub test can load many cores, but that doesn’t change the fact that you can’t use average power across the test to compare to the M3/M4 power numbers as you are comparing very different things.
But you can't use peak either like you just suggested, right?
*Butcher's voice* "My core is bigger than yours, mate".
But you just said - and I agree - that it rarely hits the peak powers cause most of the run isn't very threaded.If you want to compare to most of the M3/M4 power numbers that get reported, then you should use peak power.
But you just said - and I agree - that it rarely hits the peak powers cause most of the run isn't very threaded.
What do you mean and?And. . .?
What do you mean and?
On what planet do you compare efficiency by looking at peak power instead of averages?
Are you comparing gaming framerate by comparing max fps too?
https://browser.geekbench.com/v6/cpu/6793874 The AVX oneIt's using old 6.2.2 version.
Could you CMD into the Geekbench 6.3.0 folder and run all three in the below screenshot with 4+4 (HT disabled) and share the three different result URLs?
View attachment 102433
Just type their names in CMD and run. Don't double-click them otherwise it will run and then the CMD window will disappear.