New Mac M2 max etc. Worth it?

[golden bough]

Just saw the announcement of the new macmini. Mac people would like to hear your feedback on using Invoke on a mac m1 etc now, specifically rendering while multitasking. Thanks.

[stiff quail]

InvokeAI isn't yet the BEST on apple silicon (but it works and is reasonably fast), but the bigger neural processor in the new m2 stuff would be nice in the stuff based on Apple's CoreML implementation of SD, which we hope will make its way out to InvokeAI in the future. rendering while multitasking is always going to be tricky; you'd want as much RAM as possible. on 16gb i get graphical lag and glitching audio while rendering because i'm basically maxing things out

[stiff quail]

just as a bit of an example ... in invokeai usually it takes me somewhere close to a minute to generate an image, but on the coreml projects it's more like 10-15 seconds ... most of this is probably because of FP16 vs FP32 and my 'low' memory

[golden bough]

Ok cool so the best solution to a fast render time would be to invest in a PC with the right GPU for Stable? Or would the new M2's with lots of RAM be a significant upgrade. Right now I am running on a 2.9 GHz Quad-Core Intel Core i7 Macbookpro 2017 and it takes between 5-10 mins to render a 512x512 high rez optimised file. I would assume an M2 now would improved times signifcantly. . . looking at the specs. The alternative is to build a custom PC set up and run that seperate. What do you think?

[stiff quail]

It depends upon what software you plan to primarily use. For me, I use invoke primarily despite the speed boost in the coreml alternatives

[hidden marsh]

Don't use high res optimization with 512x512 images. You're just making it take longer, and not getting extra quality. That setting is only useful for larger images.

As for the M2, the extra gpu cores will make a difference, and the more powerful neural engine will help with the coreml stuff, as already mentioned.

If you're going to buy one, be sure to get at least 32gb of unified memory, and 64 would be much better if you want to generate images larger than 512x512.

nvidia rtx cards are still going to offer the best image generation performance, though, if that's the only thing you're buying it for.

[golden bough]

Seems a waste to just buy a PC to run SD in a corner and nothing else at this stage. My primary business runs through my mac so if I can render while working on other things it would definitely be a plus. I’ll do some more research thanks for the advice. This ol intel machine of mine will have to upgrade eventually.

[stiff quail]

i think if you plan to get a mac anyhow for other reasons, yeah ... these are cool now with the CoreML stuff, are competitive somewhat when using InvokeAI, and the future is quite probably bright. biggest thing as has been said is, don't be like me and get only 16gb of RAM 😄

16gb means ~8gb available for image gen, which at FP32 is more or less bare minimum, and very easy to go over

[golden bough]

I'm a Graphic Artist, so Adobe (Photoshop, Illustrator Indesign) all running at the same time usually, tons of fonts. Web browser and Word/Excel. Those I have open most of the time. I am actually considering moving from Adobe to Affinity, cause of cost really, but eventually as I said I'll need a new machine so the "secret sauce" is more RAM. Apple's disgusting with its upgrade prices though. . . the base machine is bad enough. . . start adding RAM and drive space and you're bankrupt before you even begin lol. My intention is really to start creating "stock images" to use in my layouts and work. . . but still a "hobby" right now.

[royal veldt]

have you considered taking some time to learn how to run remotely on a vast.ai or runpod instance? vast.ai has uninterruptible 3090s for around 10c/hr including disk space rental. i don’t have evidence but i’d advise against trying to multitask while doing stable diffusion stuff - if you hit RAM limits the swapping performance is awful on my m1 mini (tho i only have 16gb)

[golden bough]

interesting, will check it out. thanks.

[hidden marsh]

Yeah, I try not to have invoke open at the same time as photoshop/lightroom, competing for RAM. Illustrator seems to be ok for my purposes, but the stuff I use that for isn't very demanding.

[acoustic vector]

Hey! I'll put in my 5 cents. I've got the 14" MacBook Pro Max 64gb - for me and my preference the flagship of all flagships, in terms of portability and performance. 512x512 fly by pretty quickly, it does take a couple of seconds to generate higher resolutions though. I always have to many applications open (from video editing to music apps to boring stuff) without a problem at the same time as i'm rendering. The only time my machine starts making noise is when creating with Invoke. Not a complaint on invoke, it just means it's the only time I put the machine to full capacity. Instead of having to hate my life with a beefed up Windows machine, I'll opt for one of the options @royal veldt suggested, or equivalent platform. Would I absolutely love a beefed up Mac? Yes. Have I considered a beefed up PC? Very much so. But, then I start the only windows machine I have at my office, try to play nice for just about 15 minutes before the constant pain of Windows makes it to damn impossible to enjoy anything and I shut it down and say "let's go for mac and SaaS". M2 will be better or might be much better. But I do think RAM is important. Hope it helps a bit!

[royal veldt]

yeah windows is bad

[acoustic vector]

Yeah. It's what it is. But for me, it's opposite to fun. And fun is a big part of being creative = reason to try new things.

[golden bough]

Same! SAME! Not the specs but that I can’t stand windows. BUT if I intend to Mass generate images up and above 512x512 and upscale then it’s best to just build my own windows box that works all day. BUT the new M2’s announced seem to be making me think it’s better to stick with the Mac https://youtu.be/swVTHwgW7G4

[acoustic vector]

Yeah it looks like a great update if you don’t have a high end M1 (max/ultimate) already. For us who do I think it’s better to wait in the new architecture where there probably will be a real jump in performance. And when we want real speed, learn to use SaaS

[stiff quail]

https://appleinsider.com/articles/23/01/20/m2-pro-max-gpus-are-fast----but-not-faster-than-m1-ultra some M2 performance comparisons

for those that don't wanna click, hehe

[woven sapphire]

Holy crap... I have an M1 Max now and I wasn't expecting the M2 Max to be THAT large of jump. Hmmmm.... 🤔

[stiff quail]

yeah it's pretty impressive, haha

[hidden marsh]

I have the 24-core M1 max, so the M2 pro is probably faster...

[stiff quail]

if apple can keep this up ... well, one way or another we're going to see some cool stuff

[golden bough]

How is it with rendering? Whats your timing?

[hidden marsh]

512x512 in euler runs at about 1.69it/s

That's in Invoke. The coreml stuff is faster, of course.

My custom diffusers script (not coreml, but it does do fp16 mode) gets up to almost 2it/s with the same settings. RAM requirement is lower though, so I can use it to generate larger images.

[golden bough]

Hows the upscaling? Or do you that elsewhere?

Reason I ask is I plan to render at least 10-15 images in a session and upscale at the same time. I’m planning on making prints

[hidden marsh]

My script doesn't include any upscaling (yet). I like the details better when the last pass over the image is SD instead of an upscaler. I've been working on getting img2img working so that I can do something like embiggen without tiling.

[devout rover]

I am noob and don't see an instal link for mac )))