For the lucky ones that have the Ultra, are there any benchmarks yet on how it performs in TVAI?
I’m a 5090 desktop user but I also bought a 16" MBP Max (but returned it) as the results were very mixed in TVAI. I know Macs have always been worse generally in TVAI but has the Ultra improved things?
Getting one at the end of October. My expectation would be to be close to 5070-5080 speed, but not 5090.
I also don’t have a sense for how well Topaz has optimized the code for apple M-series vs. CUDA.
Not sure about the latest Ultra, but I have M2 Ultra 128Gb it is slower (0.1 fps, 8 min) than M5 Max 128Gb (0.2fps, 6min) for 0.5sec ProRes video (2560x1920) output.
M5 Ultra is two M5 Max connected together - hence its speed cannot be significantly more than twice the speed of M5 Max. Based on smdow’s test above M5 Ultra may be able to output the 0.5 sec video in 3 min - that is about 2.66 times faster than M2 Ultra at 8min
My experience with M2 Ultra 60 GPU cores is 6+ times slower than 5090 and 4+ times slower than 4090. 5080 is slightly more than 50% the speed of 5090
My estimate is M5 Ultra 80 cores may be slightly slower than 5080 performance, possibly faster than 5070, May be about the same as 5070ti but running at much lower power
The M5 Ultra also has twice the memory bandwidth of the M5 Max in addition to more processing power. I don’t know how much the extra bandwidth will help improve Topaz performance. It will be interesting to finally get some real performance results. It might be a moot point if Topaz doesn’t figure out how to fix the problems their software appears to be having with macOS Golden Gate.
The doubled memory bandwidth are providing data to up to twice the amount of GPUs and CPUs
If comparing M5 Max 18 CPU 40 GPU with 30 CPU 64 GPU M5 Ultra, the per CPU/GPU memory bandwidth is increased. If comparing with 36 CPU 80 GPU M5 Ultra, there is no memory bandwidth increase at full load
Posted M5 Ultra 96GB 80 GPU core in benchmark section…
