Unable to load image

New mac studios, up to 512gb of unified memory :letsfuckinggofast:

https://www.apple.com/shop/buy-mac/mac-studio

local AI chads just won big

33
Jump in the discussion.

No email address required.

Wow 512GB

That's what my $130 laptop had in 2005!

Jump in the discussion.

No email address required.

Jump in the discussion.

No email address required.

https://i.rdrama.net/images/1741202868dmGMVOKMwZICFg.webp

Jump in the discussion.

No email address required.

:marseygold:

Jump in the discussion.

No email address required.

https://media.tenor.com/Xc1nvxg4f_0AAAAx/bait-fish.webp

Jump in the discussion.

No email address required.

:marseybigbrain: the poster is recalling that, by configuring swap, persistent storage can function as extra volatile memory. Such storage will much slower than normal RAM will be, negating much of the advantage. The poster is therefore reminding us that, without citing the speed of memory, is is useless to know its amount.

Jump in the discussion.

No email address required.

https://media.tenor.com/Nxo9h7sUnRIAAAAx/spongebob-patrick.webp

Jump in the discussion.

No email address required.

He's so smart :tayadmire2:

Jump in the discussion.

No email address required.

How does Intel integrated graphics work? Is that a GPU? How is it different than Apple silicon? @Bussy-boy explain

Jump in the discussion.

No email address required.

512GB is just like 512GB but on a Mac Studio instead of an HP laptop.

So yea, pretty different.

Jump in the discussion.

No email address required.

local AI chads just won big

My understanding is that unified memory is really large but slow for AI workloads. !codecels confirm?

Jump in the discussion.

No email address required.

Slower than vram, faster than normal; ram. If you're too poor to afford 512gb of gpu it's a good option

Jump in the discussion.

No email address required.

You can buy 192gb of DDR5 for like $600, but that's the limit for x86 at the moment without some AI Tesla chain. It's slow, but pretty big.

Unified memory this large is an ARM toy, and even if it's a bit slow in practice, it's still very deep. Being slow doesn't matter much when your context window and model size are huge, and this thing can practically run a Large model without a GPU chain.

Now granted, this thing is $10k with that memory configuration and is running on OSX with an ARM structure, so you might actually get better price per dollar building a Tesla chain depending on actual performance. It's just novel as heck that it can run an LLM comfortably in a box.

Jump in the discussion.

No email address required.

If you're just trying to max out memory, you can get Xeon hardware and take your RAM out to 1TB although you're still looking at a minimum of $5000 unless you buy used shit. I don't know enough about AI hardware to know what else you need to add to that to have it be useful though.

Jump in the discussion.

No email address required.

You're better off stringing together Teslas, VRAM is significantly faster than even DDR5, let alone the DDR4 banks you pull off of eBay servers. You can buy a box full of broken shitty Teslas and have a few hundred gigs of VRAM strung up for a few hundred bucks (provided you're willing to pop them open and refurbish them).

NVIDIAs main shipments are corporate grade cards, most of which are compute cards that don't even output. The RTX platform is practically an afterthought.

Jump in the discussion.

No email address required.

When you're running fat butt models with billions of parameters you kinda just need butt much ram as possible before anything else

Jump in the discussion.

No email address required.

That's better achieved with pipeline parallelism, not huge, uniform, slow memory.

Jump in the discussion.

No email address required.

The new Ryzen AI max chips mog this

Jump in the discussion.

No email address required.

My guy the newest AI Ryzen isn't even as good as the M3 Max. This thing is twice as powerful.

Jump in the discussion.

No email address required.

it says 96GB unified memory? That is a lot but not 512 levels of crazy

mediocre for 4k as well compared to stacking GPUs

Jump in the discussion.

No email address required.

The 512 is there if you spec it.

The top spec of this machine is $14k USD lol

Jump in the discussion.

No email address required.

Relatively resonable. Some manufacturers charge slightly more for topend AI workstations these days https://boxx.com/systems/workstations/ai-workstations/raxx-ai-t3p-7995wx-4-x-nvidia-rtx-6000-ada-1024gb--2tb-m-2

Jump in the discussion.

No email address required.

You need to configure it. If you want 512GB you need to max out all specs and its ~$10k

Jump in the discussion.

No email address required.

The most expensive config (M3 Ultra + 512GB RAM + 16TB SSD) is actually a pretty good deal

This last gen from Apple has been surprisingly affordable

Edit : a similar Windows/Linux configs runs you like $10k, it's a 1500W behemoth and you have to build it yourself

Prebuilt is probably 12-13k

Jump in the discussion.

No email address required.

Mac mini base model is a crazy good deal for 600 bux


:ma#rseyduck2:

Jump in the discussion.

No email address required.

It's the best hardware value since the loss leading base model PS3 (which the glowies bought thousands of to make supercomputer clusters)

Jump in the discussion.

No email address required.

M4 up to Max but a weird M3 Ultra :marseyhmm:

Seems like a weird choice along with the M3 iPad Air, a lot of rumors were saying the sudden jump to M4 was due to the TSMC node being used by M3 was more expensive than M4.

I'm rocking an M2 Studio rn and it is a beast, great machine.

Jump in the discussion.

No email address required.

36GB unified memory

512GB SSD storage¹

$1,999.00

96GB unified memory

1TB SSD storage¹

$3,999.00

So it's for r-slurs that think 60GB of RAM + 512GB SSD costs $2000? :marseyxd:

Jump in the discussion.

No email address required.

Lol no they didn't. Model performance is driven by memory transfer speed and CPU to GPU transfer speed. It's gay that Nvidia caps consumer cards at 24gb but latest gen models are all nano, you run agent network's of several which is higher accuracy and performance than one big one.

On memory LPDDR5 so single 32bit channel vs dual for DDR5 vs 16 64bit channels for HBM3e.

PCIe also isn't fast enough, 63 GB/s vs 4 TB/s possible for HBM3e. That's why server GPUs use NVLink for CPU interconnect, 450 GB/s for 4 and next gen is suspected to be multiple TB/s.

Jump in the discussion.

No email address required.

>good thing I just bought my Mac Studio three months ago

Jump in the discussion.

No email address required.

Oh hey! Some hardware that's useful for AI! It'd be a shame if we drove up the actual price to 4x MSRP :marseytroublemaker:

Jump in the discussion.

No email address required.

>128gb starting at 3.7k

>256gb starting at 5.6k

>512gb starting at 9.5k

lmao

Jump in the discussion.

No email address required.

:marseypoggers:

Jump in the discussion.

No email address required.

Jump in the discussion.

No email address required.



Link copied to clipboard
Action successful!
Error, please refresh the page and try again.