64GB on mac silicon?
Just have been curious how well it would do on mac silicon wiht a lot of gpu cores. i haven’t seen a lot of good data there.
That said i get almost exactly a perfect “1 fps” on sl mini with any given 5090
This isn’t a beta - it’s the release version.
12 Likes
Only 0.3fps here on a M2Ultra with 60 GPU Cores 2x upscale of low res SD sources.
Starlight minimum requirement is 36 GB of ram. Do you have that?
I agree and I feel bad for customers who didn’t get the Founders status. I suppose Topaz is moving away from individual customers towards company or professional users in their business model.
I also got that too after updating from v1.0 to v1.0.1. I also have the base 32 GB Studio Max. However, something has changed WITHOUT BEING INFORMED between 1.0 and 1.0.1 in terms of minimum requirements. I’ve just reinstalled 1.0 and Starlight IS available in that version.
Typical of Topaz to make this change without any notification in the release notes.
Andy
I think also you can’t buy perpetual anymore, but people who have already perpetual they can’t them take it away 
and why should they pay every year for the Standard, when Starlight is gone (and if other old models does not get updated) then they can stay on the frozen perpetual as long as it works, that was my point.
that’s the question how they handel new subscriptions, at the moment existing subscriptions without founder, they still can use TVAI7 with Starlight mini …and the “new” Starlight from the Studio is the same, not better ..but I’m not sure how long V7 and SLm will work
But the thing is, the 36 GB is a currently a recommendation for starlight, not a requirement, on their website. IF topaz decide to raise the minimum requirements, then so be it - but they should make it that very clear in release notes and on the website so that we all know where we stand.
Andy
FWIW, I was able to still buy a perpetual license for Gigapixel this morning (already had Video AI) after emailing support, so I don’t think it’s too late for anyone who is still after a Founders’ Studio sub.
3 Likes
Anyone know what this means? I don’t have any issues right now, so I’m wondering if I should update, or just wait until the next patch. Anyone notice any differences in the starlight or starlight sharp models in this release?
I am a founders member with both a Mac and a Windows Nvidia Setup. I also have a Pro License. For me, I have only been able to get my Pro License to show up on my Windows Nvidea machine for the 7.1.4 Topez Video AI release, this new Studio release 1.0.1 will not recognize my Pro License, at all. But the patch does recognize my Founders status, so I can now use SLM and SLS in the new Studio release on my Nvidia PC,
However, the quality and speed of the regular SLM or Starlite as they call it now is not the same as on my 7.1.4 release. On the Studio version, it renders very fast at speeds typical to SLS, and gives bad quality like SLS does as well, and not the tru good quality that SLM is giving me at the slower speed (0.3fps) as it does on the 7.1.4 release. So for me, the Studio version is USELESS as they have dummied does the Starlite model to go faster but at crappy quaulity. So I am going to keep using 7.1.4 for all my Starlite Mini needs, it takes longer but gives the best quality results.
Dakota said that the Studio release for Starlite would be the same as it is in 7.1.4 for Nvidia users, but this is NOT the case at all. When it is rendering in studio, the FPS is all over the place acting much like how SLS acts when it is rendering, which I find very odd. It is almost as if it is rendering using SLS for just a regular Starlite setting even though I have not selected the SLS option.
Do you have any comparison images?
On an M3 Ultra with 80‑Core GPU, 512GB RAM. Only getting 0.2fps for 480p to 1080p with SL. GPU runs at 100% and ffmpeg process peaks at 58GB of RAM. Seems like there’s still a lot of work to be done for it to be properly optimized for Apple silicon, I hope it gets prioritized at some point.
1 Like
It’s nothing related to the actual model. Just cloud-based license management. Reprise License Manager - Wikipedia
yeah it would be nice, but it depends on how much apple develops for the gpu cores as well.
There’s a reason AI companies don’t use AMD — Nvidia caters to the AI market and I"m pretty sure with the advantage of having hte fast shared memory, apple should want to take advantage of the potential. I do think intel is going to make a big push into the AI gpu space though and specifically targeting the “prosumers” and … “university labs” and " small businesses.
Nvidia is too busy making the huge money off of their major ai enterprise clients to develop a good card for
The new intel cards are promising, but hopefully this time they fully commit unlike the first run. I think they will though, nvidia is buying into intel after the US gov did as well. I think intel is going to be the company that makes AI be “more local” instead of in the cloud… i think by the end of this year you’ll see intel gpu machines with 192GB+ of vram for … fairly affordable prices.. They will run something similar to nvlink to make a bank of cards operate as one…
Hopefuully both apple and intel aim to do more in the locally rendered ai “space”
it does look promising at least for intel, i’m just not sure where apple stands tbh.
1 Like
Complete confusion with the output resolution.
No matter what I choose, I get the same result!
I would like to choose 3X for Starlight, as I did in the earliest versions!
2X gives a much worse result than 3X.
And why is “–upscale-factor” always ‘1’ in the “command line”?
4 Likes
Agreed, the whole situation is a complete nightmare…I’ve just wasted a whole morning to find the resolution is different when rendering locally vs cloud, even though the ‘settings’ were identical.
Because I am unable to run Starlight with 36GB RAM, every time I start Topaz Video, my system crashes… this is because Topaz Video tries to resume my project, which consumes much more RAM than I have. I would like to suggest a question being asked “Do you want to resume unfinished projects?”. To resolve the situation, I had to erase the “Topaz Video Projects” folder in the Documents folder. I don’t have a problem doing that, but this is rather unintuitive and cumbersome. Thanks.
2 Likes
MODEL_INFERENCE_ERROR during Starlight Sharp processing
details:
01998ba2-35d5-71df-a2f3-47fa668dc554 {“estimates”:{“cost”:[150,165],“time”:[3585,3612]},“message”:“Processing failed - MODEL_INFERENCE_ERROR”,“progress”:0,“status”:“failed”}
Logs were sent to support.
Please note: When you use the “Copy” button, it no longer just copies the text of the dialog box, but also the underlying HTML source code.
__
For anyone who wants to quickly translate HTML code: https://htmleditor.gitlab.io
I cannot continue testing Starlight. I cannot process a 3s clip (356p) with 36GB of RAM on Mac (Starlight Sharp is cloud only). Hope the memory situation can be addressed in the future.
1 Like
