yes “recoverable error” is a well known “feature” 
Speed also pends on the resolution gap between source and target. 240p x4 to 960p is slower than 480p to 960p, or doing 1080p to 2160p is also faster than 240p x4
2 Likes
I have seen this and submitted some logs that seemed helpful, hoping for good news soon!
Since I restarted the mac, I haven’t (yet) ran into this issue again. As a precaution, I only generate TIFF files, so I am not losing anything in case of a crash.
2 Likes
Will try the same.. thanks
New “benchmark” with Starlight Precise 2.5, enhancing video 2x:
Mac Studio M4 Max 14 CPU 32 GPU 36GB RAM
- 4.8s per frame for a 360p video
- 6s per frame for a 480p video
Mac Mini M4 10 CPU 10 GPU 16GB RAM
- 14.4s per frame for a 360p video
- 18s per frame for a 480p video
Would love to see SP in future benchmarks 
I have no experience with Macs, but hypothetically speaking, what Mac hardware would be best for running SLP and avoiding jittering issues? 
An M2 Ultra still is about the fastest - maybe you can get one of those used?
1 Like
Wouldn’t an M3 Ultra be faster?
Yes, that’s the only M chip that is really faster than the M2 Ultra, but quite a bit less than what Apple wants you to believe.
In benchmarks I’ve seen here from M3 Ultra Computers they were only marginally faster than my M2 Ultra and due to its higher age maybe the M2 can be gotten cheaper.
Besides topaz had quite some issues on M3 based computers in the past.
(Of course the M4 and M5 are faster than the M2 or M3- but you don’t get those in an „ultra"/ dual-chip variant).
I agree, GPU count is more important than process generation. So Ultra >> Max >> Pro, and M5 > M4 > M3…
Note the M5 Max is getting closer with less but better GPU cores. However there is no desktop with M5 Max at this moment, only laptop.
If you need something now, the Mac Studio M3 Ultra is the fastest.
Edit: a year ago, I chose the M4 Max over the M3 Ultra, I thought M4 would be more future proof. There could be considerations other than processing speed in your decision.
M2 Ultra and M3 Ultra are hard to get from vendors here in my country. Sure, they’re available on eBay and similar marketplaces, but I’d rather not buy from those sources. The M4 Max is much easier to get, i have a very good offer for the 64 GB M4 Max…but I’m unsure
Nothing wrong with Ebay vendors. Pay with paypal and ur always safe..
M4 Max is a great machine but for this kind of work you will really benefit from the Ultra.
What is the price of the M4 Max 64?
1 Like
I noticed I get the recoverable error after the process has been going on for almost 24h. Its the 3rd time it reaches 23h45m, 23h56m and 23h38m.
As if the app somehow has a limit of how long it can process a certain export
Be aware that Starlight models will be terribly slow on Apple Silicon, even MUCH more than on midrange NVidia cards. 10+ seconds per frame aren’t unusual with 2x or higher upscales.
So you need LOTS of patience.
Also, you need a rig that runs 24/7 and has sufficient cooling (no problem with the studio, but I don’t know for the mini and definitely a laptop isn’t good for this task).
2 Likes
I’ve been testing it on multiple mac rigs.
Two Studios; M2 Ultra, M4 Max, M4 Mini, M4 MBP. Studio’s on 10% fan @ 50c. Mini on 24% fan @ 62c. MBP on 60% fan @ 80c
so it’s fine.. the MBP not ideal to run it so hot for long periods of time but I don’t think its a no go..
Somehow the only mac that has given me “recoverable” errors is the Mac Mini M4 with the exact same config as the MBP M4. Bare install of macOS.. Really curious what can cause these errors..
Also tried my old Windows rig with a Titan RTX 24GB. It did pass the initial test but then when I tried an export it failed, saying “out of memory”
1 Like
Theres really not much that the M2,3,4 offer against a M1 other than (sometimes only little) more speed.
The M5, with its quite improved Neural Engine could be a different case. However, Topaz doesn’t really use the NE (even not at all for all newer models) so the M5 wouldn’t benefit.
1 Like
BTW you said midrange Nvidia. You talking 5070 or something? Don’t want to spend a fortune again as I plan to get an M5 Ultra when it comes out..
Yes, 4060 - 5070TI range. For the Starlight models you better get 40/5070 TI and up.
But this is getting quite off topic.
1 Like
im crying
About 2,200 CHF, which is around 2,700 US$ for the 64GB M3 Max. Ultras are not in stock in my country at the moment. The best offer I’ve found is the M3 Ultra for $4,400, but that one is also sold out. More units may become available later, but there’s no guarantee.
My thinking is that if I sell my RTX 5090, I could buy a Mac with that money But well, I’ll wait and see, I use SLM too and the 5090 is great for that.
1 Like
Just like when Neuroserver was first released, it went to Photo first. They need more time to address issues and get it working within the Video app architecture.
5 Likes
