For inference, GPUs are still dominant as well, it's just that bandwidth tends to matter more than compute (though depending on the bandwidth the latter can still contribute - i.e. it's been suggested that Apple devices have more than enough bandwidth that on inference tasks the GPUs are compute bound and thus improving compute would dramatically improve inference performance, especially for multi-batch inference).
Do Mac users even have the *option* of GPUs that powerful? Efficiency is great but if we can't even reach that power, that's kind of an issue. External GPUs used to seemingly be a way around that but I don't think they're supported anymore
I agree with everything except the last part - the $9.5k M3 Ultra has 512GB of VRAM, that's its only differentiating factor with the $5.5k M3 Ultra which has 96GB of VRAM, same as an H100 with 80-96GB. So depending on what Apple actually releases in this space, its performance, and what the VRAM requirements of the user are, I could actually see Apple releasing a very price efficient product relative to Nvidia. Of course uhh ... other than world events being a thing. But then that'll affect all the competitors too.
On power consumption, TSMC has been promoting it as an advantage of this next generation, beyond “conventional” 3D IC. See the tiny battery powering the rocket ship below…
AMD uses it for their 3d V-Cache chips. With this go around putting the cache under the die instead of on top. I assume the same could be done with logic. Maybe Apple could put the whole GPU under everything else allowing them to up the core count.
Thanks! Getting a new PC with 40 series so this will be fun.No!
Game developers urge Nvidia RTX 30 and 40 series owners rollback to December 2024 driver after recent RTX 50-centric release issues
Rolling back to older drivers seems to be the only fix for some.www.tomshardware.com
I use Mac and PC and can say the Mac M3 Ultra certainly does not destroy the desktop 5090 in the video testing I've done.
There is a new driver coming out this week, maybe it will have fixed the issues.
Yep. As someone with three Ultra Mac Studios, Apple is really dropping the ball here. But people like to point out how little people buy these systems so it is okay.I use Mac and PC and can say the Mac M3 Ultra certainly does not destroy the desktop 5090 in the video testing I've done.
My 5090/9950X3D can render a sample Davinci resolve project (open to anyone to test on there own system) more than a minute faster than the full 80GPU M3 Ultra which has more Ram than the PC (96GB vs 64GB).
A minute maybe doesn't sound like much but when a PC does it in 2 mins and the M3 Ultra takes 3 mins (give or take a couple of seconds either way), as a percentage, that's pretty big for machines and hardware that launched at almost the same time and are flagship hardware.
As for power usage, those spending 5k upwards aren't really that bothered about a slight increase in electric bills annually if work can be done faster.
I also use my office heating much less in Winter when the machines are running at full load so that offsets things
Correct. And also complicated. I live in a country in Europe and prices here are insane for apples products. the h100 that we bought recently cost us about 30K and a 80 core m3 ultra cost 8K. Then of course the nvidia card comes in an overpriced dual xeon server with 256GB ram and lots of disks for 10k extra... So if you want to promote the apple way, I guess you could go for the baseline m3 ultra at 5K and compare to a 40-50K h100 as it is usually sold
Oh well, I now I really would like a m3 ultra as my daily driver and hope intensely that MLX will mature quickly.A large part of this is NV pushed up their power consumption a lot. They gained performance between generations in part by throwing more wattage at it. The 3080 mobile version drew 80w-115w, and often was closer to the 80w spot on most laptops. The 5090 mobile version alone, not counting the rest of the laptop, draws between 135w and 160w. That's more than the entire M4 Max MBP draws, total, even on the lower end of the power draw for the 5090.
If you already have H100, I don't see how an Ultra will match in the near future. MLX is lot more than what it was a year and half ago, but it is not going to magically fix the hardware limitations. Apple need to provide ability for MLX to leverage NPE/Core ML.Correct. And also complicated. I live in a country in Europe and prices here are insane for apples products. the h100 that we bought recently cost us about 30K and a 80 core m3 ultra cost 8K. Then of course the nvidia card comes in an overpriced dual xeon server with 256GB ram and lots of disks for 10k extra... So if you want to promote the apple way, I guess you could go for the baseline m3 ultra at 5K and compare to a 40-50K h100 as it is usually sold
Depending on the codec and complexity, I’m sure. For h.264/hevc, given Nvidia added extra encoders, the 5090 will be faster. The 4000 series certainly wasn’t. For ProRes, I doubt the 5090 is quicker. Worth remembering also that 4:2:2 10 bit performance was poor previously on Nvidia cards.I use Mac and PC and can say the Mac M3 Ultra certainly does not destroy the desktop 5090 in the video testing I've done.
My 5090/9950X3D can render a sample Davinci resolve project (open to anyone to test on there own system) more than a minute faster than the full 80GPU M3 Ultra which has more Ram than the PC (96GB vs 64GB).
A minute maybe doesn't sound like much but when a PC does it in 2 mins and the M3 Ultra takes 3 mins (give or take a couple of seconds either way), as a percentage, that's pretty big for machines and hardware that launched at almost the same time and are flagship hardware.
As for power usage, those spending 5k upwards aren't really that bothered about a slight increase in electric bills annually if work can be done faster.
I also use my office heating much less in Winter when the machines are running at full load so that offsets things
I’m just going by what I use my own machines for as they pay my bills.
Perhaps I missed it, but if you didn’t buy the M3 Ultra, how do you know the 5090 better for video work?
The new 50 series encoders/decoders suit me well.
It’s to Steam’s benefit to attract more customers, same as with Apple. Apple’s currently selling 20+ million Apple Silicon Macs a year without providing incentives to Steam and, they’ll likely sell 20+ million Macs THIS year without providing incentives to Steam. If Steam’s looking at 20+ million users and going “Nah, we’d rather not try to win those customers” then that’s on them. Games will never drive Mac sales in a huge way as anyone not in the market and looking to play a given game has a wide range of options, almost all less expensive than a Mac, to play that game.
