Discussion Cinebench 2026

[Markfw]

My result on a 9755 128 core. Almost twice That ampere at 33862. I was running 3.3 ghz minimum 4 ghz at times. This is on air cooling.

 

[igor_kavinski]

nm

 

[511]

It's reporting up to 60W, quite incredible 🙂

The delta between CPU and total system power is too high 30-35W 🤔🤔.

 

[Nothingness]

The C1 Ultra scores in GB have quite a range (up to 34xx) but cluster around 3200, max clock right around the A19 Pro (4.21GHz vs 4.26GHz).
So we’ll call it 760 pts/GHz.
The X925 scores around 2700 at 3.63GHz. Or around 745 pts/GHz.

Whereas Apple increased from ~866 pts/GHz to ~905.

X925 performance per clock on SPECint 2017 is at M4 level so about 10% behind M5: https://blog.hjc.im/spec-cpu-2017
As others said, PPC is not the issue, clock (and power) is.

 

[jdubs03]

X925 performance per clock on SPECint 2017 is at M4 level so about 10% behind M5: https://blog.hjc.im/spec-cpu-2017
As others said, PPC is not the issue, clock (and power) is.

That result is the only one I’ve seen where it’s even that close, actually more.. unfortunately so far there is not a C1-Ultra result.

And here is a different result, showing a much different picture from the mobile lens.

So I stand by the statement that the reference cores being behind in PPC/IPC; at a minimum the gap is the same in 2025 compared to 2024.

Clearly power is a different story. But clocks are nearly the same for C1U and the A19P.

 

[adroc_thurston]

But clocks are nearly the same for C1U and the A19P.

They are NOT.
Full stop.

 

[jdubs03]

They are NOT.
Full stop.

So what are they getting up to? Fine I’ll amend it, “reported clocks”.

 

[CouncilorIrissa]

That result is the only one I’ve seen where it’s even that close, actually more.. unfortunately so far there is not a C1-Ultra result.

And here is a different result, showing a much different picture from the mobile lens.
View attachment 142443
Clearly power is a different story. But clocks are nearly the same for C1U and the A19P.

They're not.
Here's an example of clock speed behaviour of a ~3.3 gb6 run.

Spoiler: .gb6 dump

"processor_frequency": {
"frequencies": [
3632,
3665,
3659,
3656,
3658,
4066,
3803,
3992,
3854,
3778,
4058,
3884,
4071,
4177,
4162,
4027,
4025,
3852,
3997,
3919,
4074,
3798,
3938,
3876,
3839,
3911,
3870,
3839,
3916,
3773,
3926,
4137,
4001,
4011,
3906,
3766,
3876,
3662,
3709
]
}

The clock speed drops as low as 3.6, and it's not even the worst case scenario.
In this run the clock speed drops as low as 3.38.

Spoiler: .gb6 dump

"frequencies": [
3793,
3377,
3769,
3785,
3540,
3680,
3791,
3467,
3761,
3793,
3544,
4195,
4189,
4115,
3906,
4125,
4038,
4020,
4040,
4178,
4082,
3924,
4181,
4051,
4116,
4099,
4074,
3977,
3842,
3888,
3893,
3975,
4029,
4017,
4013,
4015,
3926,
4097,
3849,
4007
]
}

 

[Nothingness]

And here is a different result, showing a much different picture from the mobile lens.

These are phone devices, constraints are vastly different between various phone makers, both at the HW and the OS levels. We're in the Cinebench thread and I think laptop comparisons make more sense even though admittedly SPECint is a poor proxy for Cinebench 🙂

 

[jdubs03]

They're not.
Here's an example of clock speed behaviour of a ~3.3 gb6 run.

Spoiler: .gb6 dump

"processor_frequency": {
"frequencies": [
3632,
3665,
3659,
3656,
3658,
4066,
3803,
3992,
3854,
3778,
4058,
3884,
4071,
4177,
4162,
4027,
4025,
3852,
3997,
3919,
4074,
3798,
3938,
3876,
3839,
3911,
3870,
3839,
3916,
3773,
3926,
4137,
4001,
4011,
3906,
3766,
3876,
3662,
3709
]
}

The clock speed drops as low as 3.6, and it's not even the worst case scenario.
In this run the clock speed drops as low as 3.38.

Spoiler: .gb6 dump

"frequencies": [
3793,
3377,
3769,
3785,
3540,
3680,
3791,
3467,
3761,
3793,
3544,
4195,
4189,
4115,
3906,
4125,
4038,
4020,
4040,
4178,
4082,
3924,
4181,
4051,
4116,
4099,
4074,
3977,
3842,
3888,
3893,
3975,
4029,
4017,
4013,
4015,
3926,
4097,
3849,
4007
]
}

Interesting, that helps, I’d imagine the A19P is also dropping a bit as well. It’s not just going to pin at 4.26GHz. That’d be a good comparison point.

These are phone devices, constraints are vastly different between various phone makers, both at the HW and the OS levels. We're in the Cinebench thread and I think laptop comparisons make more sense even though admittedly SPECint is a poor proxy for Cinebench 🙂

Maxon should create a mobile version to make it easier on us hah.
But the single-thread tests among each do correlate pretty well it seems.

 

[CouncilorIrissa]

Interesting, that helps, I’d imagine the A19P is also dropping a bit as well. It’s not just going to pin at 4.26GHz. That’d be a good comparison point.

Maxon should create a mobile version to make it easier on us hah.
But the single-thread tests among each do correlate pretty well it seems.

The high-scoring runs (3.8+) are basically locked to 4.1-4.2

https://browser.geekbench.com/v6/cpu/17804393

Spoiler: .gb6 dump

"processor_frequency": {
"frequencies": [
4224,
4235,
4247,
4232,
4251,
4256,
4250,
4241,
4249,
4250,
4250,
4240,
4251,
4256,
4241,
4243,
4250,
4253,
4252,
4224,
4244,
4256,
4233,
4245,
4244,
4254,
4253,
4252,
4253,
4237,
4252,
4255,
4174,
3993,
4253,
4254,
4250,
4251,
4256,
4247,
4251,
4239,
4252
]
},

 

[lightmanek]

The delta between CPU and total system power is too high 30-35W 🤔🤔.

It's possible PSU, as 210W unit is not that efficient at 70-90W, then you have losses in internal power circuit and I think that CPU power reading does not include memory.
The rest is screen at a bright setting, fans at almost full blast and all the storage and wifi peripherals.

 

[511]

It's possible PSU, as 210W unit is not that efficient at 70-90W, then you have losses in internal power circuit and I think that CPU power reading does not include memory.
The rest is screen at a bright setting, fans at almost full blast and all the storage and wifi peripherals.

Most likely I guess it only reads out Core powers Snapdragon X2 has added real telemetry from PMIC which is nice fwiw.