The Zen 4c Cores in the Ryzen 8000G APUs are Clocked Slower than the Zen 4 Cores

[Fouquin]

7540u Was Zen4, it was replaced... with Pheonix 2 mixed Zen4/Zen4c cores as listed.
View attachment 330991

https://www.amd.com/en/product/13886

https://www.amd.com/en/product/13216

The wayback machine shows 7440u being not listed as mixed cores back in August, but it was refreshed as hybrid in November, so worst case we have 2mo without full specs.
Not great but, idk if anyone actually got their hands on a hybrid zen4/zen4c machine yet.

The 7440U was launched in May. It took until third-party developers asking AMD in back channels about why the cores were different for AMD to admit that they were in fact, different. Then fast forward to a little over a week ago TomsHardware goes asking AMD in back channels why they don't have any clock ratings listed for hybrid core processors, and AMD suddenly finds the time to update their specification pages again with the correct specs.

Again, AMD launched the first consumer Phoenix2 parts in May of 2023. They did not have accurate specifications listed until after CES 2024. That is abysmal.

 

[Patriot]

The 7440U was launched in May. It took until third-party developers asking AMD in back channels about why the cores were different for AMD to admit that they were in fact, different. Then fast forward to a little over a week ago TomsHardware goes asking AMD in back channels why they don't have any clock ratings listed for hybrid core processors, and AMD suddenly finds the time to update their specification pages again with the correct specs.

Again, AMD launched the first consumer Phoenix2 parts in May of 2023. They did not have accurate specifications listed until after CES 2024. That is abysmal.

For the life of me, I cannot find me a laptop with a 7440u. I see a benchmark score leaked in Oct '23.

I am for clarity in documentation and transparency... but I honestly think this is more of a paper launch with lacking details than anything else.
As I am a stickler for details... the specifications were not inaccurate so long as they listed base/boost cores/threads cache tdp correctly.

I would love to see a boost clock listing as we used to in the how many threads can hit x frequency... but its fairly well known that boost clock is 1-2 cores. I think they need to list clocks for all cpus based on loading.

For me the far bigger issue is the branding mixing Zen2/3/4 all together. Beyond that so long as the single threaded and multithreaded performance matches advertisement...
I also really don't care about paper launches/announcements. So long as the documentation is correct by the time people can actually buy the product and read reviews on them...
That said, the only hybrid I can find is the Z1 and its specs are not listed in detail still.

 

[Count von Schwalbe]

I still maintain that for the scheduler, AMD's 4+4c configuration is a harder nut to crack than Intel's P+E. For AMD, there are four levels of performance for a thread (4 without HT, 4c without HT, 4 with HT, 4c with HT, probably in that order). For Intel, there are "just" three (P without HT, E, P with HT, probably in that order).

How do you disable HT on a core-by-core basis?

If disabled, you have 2 levels of performance per core. You don't have a mixture of logical and physical cores that may or may not have better performance depending on the workload.

 

[trsttte]

With AVX512 out of the picture, the instruction sets of P and E cores don't differ.

They are still very different cores with completely different architectures (whatever refresh cove and gracemont) , i.e. E cores don't have hyperthreading, different branch predictors, etc etc etc. Zen 4 and 4c, if AMD is to be believed, are exactly the same but one is crammed together (namely by reducing L3 cache) to fit a smaller space.

Intel Architecture Day 2021: Alder Lake, Golden Cove, and Gracemont Detailed

www.anandtech.com

Intel Architecture Day 2021: Alder Lake, Golden Cove, and Gracemont Detailed

www.anandtech.com

 

[Oberon]

Sure, but clock speed is an exact measurable number. It shouldn't be reported as an average of clock speeds across different cores. It's unnecessary confusion.

It is not an exact number and hasn't been ever since the advent of Precision Boost. The CPU has a max clock and a base clock; the actual operating clock varies between those two depending on the workload.

 

[R-T-B]

It is not an exact number and hasn't been ever since the advent of Precision Boost. The CPU has a max clock and a base clock; the actual operating clock varies between those two depending on the workload.

And? The points are still fixed, quantifiable and meaningful stats.

 

[Oberon]

And? The points are still fixed, quantifiable and meaningful stats.

So they give you the fixed points AND some extra information about what you should typically expect. If they didn't give that information, you people would just be in here complaining about them not running at max clocks all the time.

 

[AusWolf]

I still maintain that for the scheduler, AMD's 4+4c configuration is a harder nut to crack than Intel's P+E. For AMD, there are four levels of performance for a thread (4 without HT, 4c without HT, 4 with HT, 4c with HT, probably in that order). For Intel, there are "just" three (P without HT, E, P with HT, probably in that order).

Nah - there's only slower and faster cores which has been a thing on Intel since 11th gen and Turbo Boost 3.0 (or whatever it's called). Even Windows 10's scheduler is well aware of that, directing lightly threaded loads and foreground activity to the two most performant cores.

It is not an exact number and hasn't been ever since the advent of Precision Boost. The CPU has a max clock and a base clock; the actual operating clock varies between those two depending on the workload.

Right... The base clock and maximum boost clock are exact numbers, not an average or something calculated by a formula.

 

[Oberon]

Right... The base clock and maximum boost clock are exact numbers, not an average or something calculated by a formula.

... and? They give you those and more. There's literally nothing to complain about.

 

[AusWolf]

... and? They give you those and more. There's literally nothing to complain about.

I disagree. If my CPU has cores running at 3.5 and 4.5 GHz, then it's not a 4 GHz CPU. A 4 GHz CPU would mean that all cores are capable of running at 4 GHz.

I like calling things on their name. Maybe I'm weird.

 

[R-T-B]

So they give you the fixed points AND some extra information about what you should typically expect. If they didn't give that information, you people would just be in here complaining about them not running at max clocks all the time.

I thought the whole issue was that they weren't really giving all those details, just an "average clock."

Also lol at "you people." My god my main system couldn't bleed more AMD if it tried.

 

[Wirko]

How do you disable HT on a core-by-core basis?
If disabled, you have 2 levels of performance per core. You don't have a mixture of logical and physical cores that may or may not have better performance depending on the workload.

I didn't mean enabling or disabling, I meant scheduling two threads or one threads to run on a core, which of course dynamically changes. HT is supposed to improve performance by a factor of about 1.3x at best, but that's for the two threads combined. This means about 0.65x for each thread, as none of them can be prioritised in the x86 architecture.

Nah - there's only slower and faster cores which has been a thing on Intel since 11th gen and Turbo Boost 3.0 (or whatever it's called). Even Windows 10's scheduler is well aware of that, directing lightly threaded loads and foreground activity to the two most performant cores.

Yes, the scheduler is aware of preferred cores, but even more importantly, it's aware of hyper-theading. Since Windows 2000 SP3, it can handle this (not simple) problem of computing, sometimes better, sometimes worse, as it seems.

They are still very different cores with completely different architectures (whatever refresh cove and gracemont) , i.e. E cores don't have hyperthreading, different branch predictors, etc etc etc. Zen 4 and 4c, if AMD is to be believed, are exactly the same but one is crammed together (namely by reducing L3 cache) to fit a smaller space.

The instruction sets of P and E cores still don't differ. But your point is important, too. The throughput/performance of certain types of instructions, loops etc. can vary significantly between "Cove" and "Mont" architectures, and the compiler can optimise code for one of them, not both. AMD CPUs shouldn't have this same problem (and they have to chew through code compiled for Intel anyway, so they need to be one level smarter).