High End GPUs for Stable Diffusion - Nvidia RTX and Apple M3 - Best Graphics Cards in Local Installs
HTML-код
- Опубликовано: 11 ноя 2024
- Following on from the guide for low to mid-range graphics cards for Stable Diffusion, this video looks at these 9 options amidst an ongoing shortage of RTX 4090s. The situation is moving fast and some of the data will be out of data in weeks or days, subscribe for more updates. Join the channel membership for more exclusives.
Stable Diffusion Courses - use Limited Time Discount Code - SHORTAGE
cutt.ly/MwIZrXS2 - expires 25 December
🖥️Nvidia RTX 4000 SFF - 20 GB
geni.us/B2TDd
🖥️RTX A6000 Ada Lovelace Generation 48 GB
geni.us/xGGu
🖥️Nvidia RTX A6000 48 GB
geni.us/oFLooon
💻Apple M3 MacBook Pro - 64 GB 128 GB 48 GB
geni.us/e2gd
geni.us/Lj5N
🖥️Nvidia RTX A4500 20 GB
geni.us/FNGMR
Nvidia RTX A5000
geni.us/FBMJZ
🖥️EVGA RTX 3090 FTW3 Ultra Gaming, 24GB
geni.us/A1Um
🇺🇸 RTX 4090s Occasionally in stock at list price
geni.us/0kzwWu1
geni.us/AV0MRT
🇺🇸 Professional Shop
geni.us/SNqDCS3
🇺🇸 Nvidia H100 Hopper PCIe 80GB HBM2
geni.us/VQBmUG4
🇺🇸 Nvidia A100 80Gb HBM2
geni.us/HPdvc
🇬🇧 RTX 4090s
geni.us/5ABcB5K
geni.us/GfnpdGe
🇬🇧 RTX 4080s
geni.us/GQ0C
geni.us/QwNpAX6
Prior videos
The End of the RTX 4090 - • RTX 4090 Running Out o...
Mid range GPUs - • Best Entry / Mid Level...
Join 🏆 this channel to get access to exclusive content and perks:
bit.ly/XOVjoin
As an Amazon Associate Pixovert earns from qualifying purchases, as well as from associations and partnerships with other retailers, advertisers and sponsors.
Following on from the guide for low to mid-range graphics cards for Stable Diffusion, this video looks at these 9 options amidst an ongoing shortage of RTX 4090s. The situation is moving fast and some of the data will be out of data in weeks or days, subscribe for more updates. Join the channel membership for more exclusives.
Stable Diffusion Courses - use Limited Time Discount Code - SHORTAGE
cutt.ly/MwIZrXS2 - expires 25 December
🇺🇸 RTX 4090s Occasionally in stock at list price
geni.us/0kzwWu1
geni.us/AV0MRT
🇺🇸 Nvidia H100 Hopper PCIe 80GB HBM2
geni.us/VQBmUG4
🇺🇸 Nvidia A100 80Gb HBM2
geni.us/HPdvc
🖥Nvidia RTX 4000 SFF - 20 GB
geni.us/B2TDd
🖥RTX A6000 Ada Lovelace Generation 48 GB
geni.us/xGGu
🖥Nvidia RTX A6000 48 GB
geni.us/oFLooon
💻Apple M3 MacBook Pro - 48 GB 64 GB 128 GB
geni.us/e2gd
geni.us/Lj5N
🖥Nvidia RTX A4500 20 GB
geni.us/FNGMR
Nvidia RTX A5000
geni.us/FBMJZ
🖥EVGA RTX 3090 FTW3 Ultra Gaming, 24GB
geni.us/A1Um
🇺🇸 Professional Shop
geni.us/SNqDCS3
🇬🇧 RTX 4090s
geni.us/5ABcB5K
geni.us/GfnpdGe
🇬🇧 RTX 4080s
geni.us/GQ0C
geni.us/QwNpAX6
🇺🇸 Professional Shop
geni.us/SNqDCS3
Prior videos
The End of the RTX 4090 - ruclips.net/video/Oi7OfnagzAs/видео.html
Mid range GPUs - ruclips.net/video/4f14_izBPII/видео.html
Join 🏆 this channel to get access to exclusive content and perks:
bit.ly/XOVjoin
As an Amazon Associate Pixovert earns from qualifying purchases, as well as from associations and partnerships with other retailers, advertisers and sponsors.
Unified memory is amazing. I get bigger images on M1 16GB and cannot have them on RTX 2060 on my gaming laptop, besides the RTX 2060 having faster processors than the 2020 M1 GPU.
Those 64 or 128GB are insane, considering that all that memory can be used on the GPU as well.
😲 80GB GPU's! That's some very serious computational hardware!
You're mistaken about linking not possible with the end of NVlink... Nvidia now does this over the pcie bus at speeds comparable to NVlink, so the connectors/ports are just not necessary anymore.
Does this allow combined memory operations though
It would have been far better if you could have acutally run sdxl and tested on these GPUs because some videos on the RUclips have show be that high end prop GPUs are quite worse than the consumer ones like 4090
The high end gaming GPUs usually top on raw performance benchmarks, the advantages for the Pro cards are lower energy requirements which matter more for that segment. They have potentially better stability too, but that can be hard to measure.
Please run test Stable diffusion and ComfyUi on new M4 Max laptops from Apple.
I think it would be interesting to see up to date benchmarks of the A770 16gb, and the M3 not to mention the M1 or M2 series. If the software has caught up to the point that AMD, Intel and Apple are able to do Stable Diffusion far better than they were it would also apply to their earlier models. The problem with the Macbook Pro is you are paying for an entirely useless screen, keyboard, trackpad and terrible cooling. If you can wait, you are going to get more power for less money from an M3 Mac Studio or an M3 Mac Mini. And of course, depending on how the other chips now perform that might not even be worth it.
I know even Marquess Brownlee said he wasn't going to move on from his M1 Mac to the M3 as he just doesn't need it for on the move editing. That's kind of bonkers to me, but it's an admirable level of frugality from a man whose time is clearly worth a lot. Point being, I've not seen even the massive fans of Macs suggest it's worth upgrading from one to the other yet.
Given the leaps with MS Olive and AMD/Intel I do wonder if the hardware is more capable than the dominance of Nvidia would suggest and part of the problem has been that SD etc just weren't coded for the AMD/Intel hardware since it doesn't exist on any meaningful level yet (I think AMD is something like 11% of the Steam survey?). It may be no different to my ZX Spectrum in the 80s - by the end of it there were games with voice samples, the colour clashing was solved and graphics were greatly improved, all in the same 48/64K computers just because they learned how to write efficient code on the same silicon.
I'm not suggesting you do a tonne of benchmarking by the way. I suppose you and several of the other Stable Diffusion channels might be able to make some progress on that together. Tom's Hardware did some benchmarking but unhelpfully ignored the existence of the 1080 series so you can't see the leap in their charts. I suppose technically the benchmarks are better now than they were a year ago on the same cards due to the software improvements and they just didn't have the energy to test more cards. Problem is you still need the full spectrum of cards it works on to get a complete picture of performance, what upgrades are worth it and so on.
What should I get: Rtx a4000 or Rtx 4070 TI Super? (For stable diffusion)
I have a m3 Max with 36gb of memory, do you think that it would perform on par with a 4060 ti? Or a 3060? I am considering building a dedicated pc with one of those GPUS
hey man, how is it running? i'm just curious whether I should buy mac m3. How much it takes to generate 512x728 image for example, and upscaling as well?
@@yss7557
just don't
it's alight. But I went out and bought a used gaming PC with a 3060 that annihilates my max. If you need something in a pinch it's not bad but there's a lot of differences
@@yss7557 oh it’s alright
I got a RTX 4070s and it’s night and day
Go nvidia when you can but m3 isn’t bad for in painting and upscaling
SD on Apple M3 on 64GB or 128GB RAM would be great
RTX A5000 is this a good option? Saw good one on ebay
xformers downloading speed is 40kb/s for me and it is so irritating. any other way to install it?
0:21 what software are you using here? tia
Stable Diffusion for Apple please
nahh bro ,RTX6000 ada is way better than apple and is affordable , no kidneys missing 🤣🤣🤣
8:06
Oh?? Why banned in China?
The US hopes to damage the Chinese economy in favour of their own by restricting their ability to access AI processing. But national security is the 'reason' given. Of course, China is part way to developing their own high end graphics card industry, and FAB technology and this will not, in fact, slow them down. It might also make them more keen to invade Taiwan which obviously has a lot of the manufacturing capacity for chips. Also why the US is willing to brib... incentivise companies to build FABs in the US so that they have a secure supply of chips in the event that China gets it's Empire on.
Because 4090 is actually massively purchased/used by small to mid sized companies in China, doing all sorts of smaller scale research and AI computing through Nvidia NCCL multi gpu interface (an workaround for lack of NVlink in 4090). The gaming cards are rebuilt to turbo fan cards and massively put into multi GPU servers.
I bought four 4090s in Australia and sold them all to such companies when I visit my parents in China and made some comfortable profit.
👋