Thoughts on Apple M chips?

[Demmers]

I've been running a Ryzen 3400G CPU for some years now (no GPU), and am looking to upgrade over the coming months.
I was originally looking to upgrade to the latest AM5 Ryzen chips (8700G), that is until Apple recently released the latest Mac Mini with M4 for just £600, and now i'm torn!
This will be my daily driver PC, and yes i'm fine going from Windows to Mac OS. But i'm just curious if there are any folders using Apple silicon chips, and their thoughts on them, especially if any fold in the background while using the Mac for anything else. I've already looked on here, and all i've really seen are remarks that FAH needs to be set to the Performance cores count for the M chips, so in the M4's case, that will be 4 (10 in total, 6 Efficiency). That seems small compared to the maximum of 16 for the 8700G! But realistically I would probably set that to 12 or 14 so I can comfortably use the machine at the same time.
I've had a look on here https://folding.lar.systems/cpu_ppd/overall_ranks but neither chip are recorded yet.
According to these benchmarks https://nanoreview.net/en/cpu-compare/a ... en-7-8700g the M4 appears to be the better chip overall, but the folding performance could sway it for me.
And before anyone asks, yes I know I can get more points with a dedicated GPU, but I like the idea of helping smashing through the CPU jobs while everyone else does GPU. The difference between the amounts of them as shown here is staggering! https://apps.foldingathome.org/serverstats

[Joe_H]

I have a Mac Studio with the M1 Max chip. I set it up to use the 6 performance cores for folding, and have not noticed any impact at all while gaming in WoW or other games so far. I see a bit higher PPD than shown on the Lars site, but that includes an average over all cores. The economy cores do slow things down. The Mac Studio barely warms up, and I notice very little fan noise.

The configuration I have has 32 GB RAM and the 512 GB SSD. For full performance you will want to use the v8 client. The v7.6.21 client will work, but does not properly identify the CPU to the servers. That results in the v7 client downloading the Intel version of the folding core. It works fine under the Rosetta emulation, but there is some loss of performance. Currently there is also a slow memory leak in the client code, I barely notice on my system with the 32 GB, but it may show up more on a smaller memory configuration. Restarting every so often will reset usage.

[calxalot]

My M1 mini 16GB has been fine with memory, but I’m not trying to run much else. It is a secondary machine. Virtual memory has presumably handled the slow memory leak.

[Demmers]

I have a Mac Studio with the M1 Max chip. I set it up to use the 6 performance cores for folding, and have not noticed any impact at all while gaming in WoW or other games so far. I see a bit higher PPD than shown on the Lars site, but that includes an average over all cores. The economy cores do slow things down. The Mac Studio barely warms up, and I notice very little fan noise.

The configuration I have has 32 GB RAM and the 512 GB SSD. For full performance you will want to use the v8 client. The v7.6.21 client will work, but does not properly identify the CPU to the servers. That results in the v7 client downloading the Intel version of the folding core. It works fine under the Rosetta emulation, but there is some loss of performance. Currently there is also a slow memory leak in the client code, I barely notice on my system with the 32 GB, but it may show up more on a smaller memory configuration. Restarting every so often will reset usage.

Interesting, thanks Joe! I just saw your numbers in folding.extremeoverclocking.com, averaging over 300,000 PPD on just 6 P cores?! From what I can see on the Lars site (when compared to previous G chip), the 8700G should be well into 6 figures as well, but the M4 only has 4 P cores. But then there's the difference in Performance Per Watt, where the M chips just destroy everything else. Decisions decisions lol!

[calxalot]

In the US at least, there is an extended return period.

The education and military discounts are available to family members.

[calxalot]

My M1 mini gets 40-240k ppd with bonus. Usually 150k. I expect the base M4 to get twice the ppd.

[Joe_H]

Interesting, thanks Joe! I just saw your numbers in folding.extremeoverclocking.com, averaging over 300,000 PPD on just 6 P cores?!

The EOC stats are for two systems. The other is an i7-770K equipped Hackintosh I put together half a dozen years ago. Folding on 4 cores that puts out about 140K PPD, though it looks like the projects are a bit more variable running on it. Current project is reporting about 90K PPD.

[Demmers]

My M1 mini gets 40-240k ppd with bonus. Usually 150k. I expect the base M4 to get twice the ppd.

Well anything released in 2024 is going to smash my current chip, which averages 40k! I just have a hunch though the 8700G will still beat the M4, albeit using slightly more power, just due to the FAH "CPU" count available. Just so I get a better idea, how many cores do you run with?

[calxalot]

The M1 has 4 p-cores. So I use 4.

The geekbench multi core scores will be distorted by the e-cores, which you will not use for folding. Makes comparison with x86 processors dodgy.

Other people have measured 24W at the wall for the M1 mini folding.

[calxalot]

Seems like there are a lot of enthusiastic people with M4 mini on reddit. Maybe you can get someone there to try fah.

[toTOW]

I was originally looking to upgrade to the latest AM5 Ryzen chips (8700G), that is until Apple recently released the latest Mac Mini with M4 for just £600, and now i'm torn!

Stick to AMD CPUs, this will avoid you the pain of dealing with big.LITTLE architectures that FAH doesn't like ...

[calxalot]

Big.little is really not a problem on macOS, as long as you don’t exceed p-core count.

[klasseng]

I’m running an M2 Mac Mini (6 out of 8 cores) & an M1 Max Mac Studio (8 out of 10 cores), 24hrs per day for the last 4+ weeks. Averaging 433,000 ppd (both machines). https://folding.extremeoverclocking.com ... =&u=114950

Was easy to setup, runs without impacting normal operations. Highly recommended!

[Demmers]

To be honest, after just securing myself a new job which pays better than the last (I work in IT funnily enough), i'm thinking to hell with it, and save up for a custom-build 16 core/32 thread Ryzen machine! Set FAH to 28 CPU's and see if the integrated graphics will be strong enough for normal usage. If not, i'll just plumb in a low-end 40 series Nvidia card and fold on that as well.
I've asked in Reddit regarding the M4 Mini, so who knows. Thanks for all the advice everyone!

[daiko]

I just picked up an M4 Pro Mac Mini, and have to say I’m impressed after just a couple of WUs processed. I’m getting over 600 KPPD on these, compared to ~300 KPPD on my M1 Max Mac Studio. Both have 8 Performance Cores, but the increase in clock speed and efficiency really shows. Typical multi-core CPU benchmarks I’ve seen online show a ~60% increase in performance over the M1 Max, but this folding result is pretty striking. Of course, different WUs and projects will vary widely on PPD—I’m hoping to see both machines running the same project at some point for a (hopefully) more equitable comparison. But this Pro Mini certainly gets my endorsement.