As you can see in this list [1], there are several systems which exceed 21k. The 19k scores are mostly stock systems with no tuning (for example, RAM at 3200Mhz, stock fan etc). The Macs would have been carefully tuned, so it's only fair that you'd do so on assembled kits.
I haven't tested Eco mode myself, but from what i've read performance drops by 10-20% depending on the configuration (105w vs 65w). Still in the ballpark.
If we use an overlocked 7590X – ignored the additional power draw – in 65W eco mode, using a rounded up 22K GB6 score, and assume a 20% performance loss (~17600 GB6), the 7950X, at 40% higher TDP than the M3 Max, is still 10% slower than the M3 Max.
Which is especially insane considering the 7950X is a desktop-class CPU, and the M3 Max is sitting in a laptop.
Since we agree on the non-linear relationship between TDP and benchmark scores, can we also agree that further reduction in power (to say 45w) would only cause a small drop in performance? For argument's sake, let's triple the 10% to 30% and also assume that TDP = avg power draw during the benchmark. The 5nm 7950x is then 30% slower than 3nm M3 Max.
My original argument was that Apple's advantage is mostly just TSMC.
> Which is especially insane considering the 7950X is a desktop-class CPU, and the M3 Max is sitting in a laptop.
7945HX3D is a 5nm 55W laptop part, and it scores 15-16k on GB6.
> Since we agree on the non-linear relationship between TDP and benchmark scores, can we also agree that further reduction in power (to say 45w) would only cause a small drop in performance?
You're confusing actual power with the system's configured power limit. When the default power limit is higher than the actual power draw during a given workload, then you obviously have headroom to lower that power limit quite a bit without severely reducing performance. That doesn't mean further reductions in the power limit will have similar impact, once you're working in the range where the power limit actually starts to kick in. And as you get to even lower power limits, a smaller fraction of that power budget is available for doing useful work as subsystems like the memory controller cannot reduce their power consumption as readily as the CPU cores.
The 7945HX3D is a mobile version of the 7950X3D (not 7900X3D), but the only way to consider it a 64W CPU is to ignore the power used by the IO die—and a CPU isn't much use without a memory controller.
Don't mistake a long-term sustained power limit that OEMs can freely adjust for an actual power consumption measurement, especially when discussing a benchmark that only does short bursts of work.
Yes, but not a m2 pro, m2 max, or m2 ultra. Or a M3 pro or M3 max.
Sometime in early 2024 AMD is supposed to release a new APU (CPU+iGPU) called strix. Doesn't seem particularly noteworthy, but mid to late 2024 AMD is going to bring out a chip called strix halo that FINALLY brings more than 128 bit memory system to a APU.
It baffles me that despite a huge GPU shortage that lasted years and shipping a huge number of XboxX and PS5 with nice memory systems that they didn't bother to ship a decent APU and a decent memory system for the desktop.
At least the halo strix should give the M3 pro a run for it's money, still half the M3 max and 1/4th of the M2 ultra.
If they were going to announce Zen 5 high-end mobile parts at CES in January 2024, they would have launched Zen 5 desktop parts by now (because the high-end -HX mobile parts are literally the desktop silicon put into a BGA package instead of LGA). A successor to the 7945HX3D can't be much less than a year away, meaning the 7945HX3D is less than halfway through its product lifecycle.
Actually their standard mobile lineup comes first. Then desktop or if the gains are just a small step they skip and go to premium laptop. So it goes standard laptop -> desktop -> premium laptop.
That is how AMD has released chips for the past six years.
Don't look at the model numbers, look at the architectures.
Zen 1 desktop processors (branded Ryzen 1000 series) were released spring 2017; Zen 1 mobile processors (branded Ryzen 2000 series) were released starting in fall 2017. Zen+ desktop processors (branded Ryzen 2000 series) were released spring 2018; Zen+ mobile processors (branded Ryzen 3000 series) were released at the beginning of 2019. Zen 2 desktop (branded Ryzen 3000 series) were released mid 2019, followed by Zen 2 mobile (branded Ryzen 4000 series) in spring 2020.
For Zen 3 desktop, they skipped 4000 series branding to catch up with the mobile branding: Zen 3 destkop (branded Ryzen 5000 series) launched late 2020, followed by Zen 3 mobile (branded Ryzen 5000 series) at the beginning of 2021. Zen 3+ (branded Ryzen 6000 series) was a mobile-only update to Zen3 (same CPU microarchitecture, minor die shrink, new memory controller) launched at the beginning of 2022. Zen 4 desktop (branded Ryzen 7000 series) launched fall 2022, followed by Zen 4 mobile (branded Ryzen 7000 series) at the beginning of 2023.
Their new architectures launch on desktop and server first, using the same CPU chiplets in both segments. The monolithic mobile processors come later. But every year, they increment the model numbers of their mobile parts whether or not they have a new architecture, and the mobile parts are almost always announced at CES in January; that's simply how the laptop market functions.
Zen 5 desktop and server parts aren't here yet, so whatever 8000 series mobile parts they introduce at CES in January 2024 either won't be using Zen 5, or they'll be announcing at the beginning of the year but not shipping until fall at the earliest. Recent rumors suggest that their high-end monolithic mobile chip (a new product segment for them) has been delayed from late 2024 to early 2025.
Sometimes they do launches at and during CES so definitely keep an eye. The biggest benefit is that it prevents companies from course correcting their design if they announce and release at the same time. Intel gets rug pulled. Which seems to be the strategy for the last few years.
Also if the 8k is launched it is usually a small limited run of their basic chips. Desktop chips will deifnitely be later in the year.
https://browser.geekbench.com/processors/amd-ryzen-9-7950x
I'm also not sure what you mean by 7950x having a lower TDP than the M3 Max.
Even in eco mode, the 7950x has a TDP of 65W, vs the M3 Max's 45W:
https://www.anandtech.com/show/17585/amd-zen-4-ryzen-9-7950x...