Not dead, just coming from a long way behind and it is hard to change what you have been doing for 30 years, while the world was screaming at you that you needed to change, but you just made more and more money, until suddenly you didn’t.
Some of the latest processor’s are apparently a real step forward. Just depends on whether they can make it to the next couple of generations and continue that progress. With ARM, RISC V growing rapidly and AMD also doing interesting stuff as well, we will have plenty of options available going forward.
I’m looking at the baseline model of the new Mac Mini M4. Any hope of doing DSD256 with some (if not all) filters and DAC correction for the Holo DACs or Gustard A26?
Or would I be better off with a PC for this? I’m considering either the Mini M4 or a PC that can do double duty for HQP and some casual gaming. Say, a decent CPU and a mid-range Nvidia GPU? Any suggestions on the latter? 13600k and 3080ti seem like a good start, but will I get it done for, I don’t know, under $1k?
Sorry for the slight thread hijack. I, too, am curious how capable the new Minis will be. Thanks.
Memory bandwidth applies to CPU as well. This is also apparent on my performance test with M1Max that has 400 GB/s memory bandwidth. In tests that have largest memory footprint, M1Max reaches 60% of 14900K speed. While tests with smaller memory footprint 14900K performs better and M1Max reaches just 30% of 14900K speed. This is with 14900K using DDR5-6400 CL32 RAM modules.
Intel is more energy efficient than AMD though. Especially since 12th Gen when they added efficiency cores in similar style as Apple M series. Efficiency cores have been in use on mobile phone SoC’s for couple of decades already, but now landed on desktop too. Thus for example the 14900K has 8 performance cores and 16 efficiency cores. And HQPlayer can use those E-cores similar way as GPU for offloading part of the work. This works very nicely (on Windows and Linux where such is possible).
AMD doesn’t yet have anything like efficiency cores on their CPUs yet. Only big banks of performance cores.
I think that you might be mistaken re: Intel and AMD TDP comparison: the i9-14900K has a specified 125W TDP but its peak power consumption is well over 300W:
TDP in Intel’s case only refers to the power consumption at base clock (3.2/2.4GHz) which is a far cry from the max 6 GHz turbo frequency.
Meanwhile the 9950X only draws ~200W at default settings, and is not known to draw more than 300W under any circumstances short of extreme overclocking:
Similarly the AMD chips perform way better than Intel ones when their power target is limited to, say, 65W.
One reviewer (Geekerwan) has benchmarked the M4 (in the iPad Pro) using the SPEC 2017 test suite, and achieved a rate-1 score (single threaded) of 11.72/17.96 respectively for integer and floating point. The same reviewer was only able to surpass that result by overclocking the Ryzen 9950X to 6.7 GHz under LN2.
For further comparison, AnandTech’s SPEC 2017 rate-1 results have the 14900K at 10.94/16.90 (methodology may vary and I found AnandTech to produce slightly higher scores on SPEC compared to those by Geekerwan) Note that the results are not a measurement of IPC but is rather frequency-dependent.
The M4 Pro has 10 performance cores, clocked at >4GHz. The frequency alone is 33% higher than that of the M1 Max, so even when not factoring in IPC uplifts (which are significant, >40% on FP) the M4 Pro should be 40% faster than the M1 Max. I would not be surprised if the M4 Pro is >50% faster than the M1 Max (which has the same CPU cores and configuration as the M1 Pro)
I am not concerned with thermal throttling as it would be stupid for Apple to promise a certain level of performance (especially for M4 Pro models) but fail to deliver on that. However I am indeed concerned about noise levels.
So yay or nay for m4 pro mini with 24gb ram for us normal mortals ?
I don’t want to blow 600-900watt or at something like listening to music… might as well leave a vacuum cleaner running for a few hours every day and at the end of the year be surprised about the electricity bill
We’ll see when someone tests… I will also test at some point.
M1Max has twice the amount of cache and memory bandwidth (512-bit memory interface), compared to M1Pro (256-bit memory interface). This certainly makes a difference.
This is also difference between M1Max with 400 GB/s memory bandwidth and M4Pro with 273 GB/s memory bandwidth. M4Max betters that to 410 GB/s on 10 P-core model and 546 GB/s on 12 P-core model.
For my typical playback case at DSD512 on i9-14900K + RTX A4500 combination, it draws a bit over 100W from the grid. Which is less than my power amp…
This is not what Apple specifies and what I’m seeing with HQPlayer. But please note that the amount of RAM installed has impact on this, especially on M1. Additional RAM is populated on additional memory channels. So for example basic M1 with 8 GB of RAM is notably slower than basic M1 with 16 GB of RAM.
They later tried to alleviate this with additional memory controller on M2, but it actually resulted in degraded performance.
Please send me your benchmark program (or send me a link to it) and I will run it on my M4 Pro Mac mini when it arrives on Friday. Also, please tell me the results you saw with other Apple SoCs and what you see with the 14900K…
Hey, any updates on this one. I’ve the trigger on the finger for either a M4 pro mac mini or M4 max MacBook Pro. I’d rather a mini format, but not sure if M4 pro will give me all the power I’d like for DSD512
I tried those settings with my M3 Pro Macbook Pro and I couldnt get a smooth playback experience. It was very close though so you might edge it with an M4 Pro.
HQPlayer has already benchmarking capability through the null output plugin and printing timing information on the status bar after completion. One just needs to decide which track and settings to use as benchmark.
