Tony359 wrote:A PC will give you expandability a Mac will never have...And a Mac with the same "rendering power" of a PC will cost you a fortune...
That was the old viewpoint, and I think it was valid back then. E.g, the fastest 2019 iMac 27 had an i9-9900K Coffee Lake CPU and an 8GB Radeon Pro Vega 48. A top-spec configuration in 2019 cost about $5,000. You could obviously build a faster PC back then for $5k, esp. GPU performance.
OTOH today your RTX 3070 costs more than an entire M4 Mac Mini. You could upgrade the GPU to a 16GB RTX 5080, but that costs more than an entire M4 Pro Mac Mini.
That M4 Pro Mac Mini has about 1.8x the single-core CPU performance as your Ryzen 5700X, and 2.3x the multi-core performance.
The GeekBench 6 OpenCL GPU score for the RTX 3070 might only be about 9% faster than the M4 Pro Mac Mini GPU's Metal score. That's not a great test using two different APIs, but they are supposed to represent equivalent units of work.
There's an easy way to test that on a real-world workload, which is using NeatBench 6. I suspect even your current RTX 3070 is considerably faster on NeatBench 6 than the M4 Pro Mac Mini, but why not try it?
https://www.neatvideo.com/download/neatbenchI have an M4 Pro Mac Mini, and I got these results on NeatBench 6 at the default 4k settings:
CPU (13 cores) and GPU (Apple M4 Pro): 15.74 frames/sec
Despite the progress with Apple Silicon, the NeatBench 6 scores for the RTX 5090 are vastly faster than an M4 Pro Mac Mini -- about 3x faster than my scores.
OTOH it's only about 2x faster than my now-obsolete M1 Ultra, and the M4 Ultra will probably be about 2x faster -- maybe rough parity with the 5090 on some tests. It will be interesting to see.
The M2 Ultra Mac Studio is about $5k, and if the M4 Ultra Mac Studio is about the same price, I'm not sure that would cost much more than an equivalent PC with similar CPU and GPU performance. The 5090 alone is about $2k.
Another key issue is the increasing use of AI neural network code paths in local apps. The new GeekBench AI test shows the quantized Neural Engine score on my M4 Pro Mac Mini is about 3x faster on that test than my M1 Ultra, 49289 vs 15984.
https://www.geekbench.com/ai/