I will worry about tomorrow when it comes.
For now, I will enjoy today.
For now, I will enjoy today.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Mac Studio M1 Ultra GPU will be beaten by AMD and Nvidia in just 6-7 months.
- Thread starter exoticSpice
- Start date
- Sort by reaction score
To get the same amount of GPU memory you will need 5 or 6 RTX 4060s.
Or you could go with the quadro equivalent, which, going off the RTX A6000 48GB card, will be $5000 for 48GB. The A6000 card is probably the best equivalent in that it is designed for the same professional workflows with large data sets as M1 Ultra.
For gaming, sure the Nvidia 40xx cards will probably be faster. For workflows that actually take advantage of the card it is less clear cut. The M1 Max comes in spitting distance of the 3090 desktop card when rendering the Moana data set, and beats 2x2080ti. The M1 Ultra should roundly beat the 3090. Although looks like Redshift doesn't scale too well...
at 350W besides $450 you will have to count the energy bill difference between M1 ultra and this
))This chase for performance with higher W must to stop..nvidia and amd must find an different way...in 10 years you will have a mid-high end gpu that draws 800W alone
Nvidia already beats Apple based on raw computational power.
A RTX3090 has already about 36 TFLOPS in computational power. This is 80% more than the M1 Ultra at 20 TFLOPS.
This is comparable to the ratio between M1 Max and GTX 3080 for laptops.
The mobile RTX3080 is at about 18 TFLOPS, also about 80% more than the M1 Max at 10 TFLOPS.
The M1 Ultra Duo or whatever it will be called in the upcoming Mac Pro will likely have 40 TFLOPS, which is faster than all current Nvidia/AMD cards. However, people who use those, probably have 2-3 in their PC. It will be interesting to see if Apple will try to compete with that, by making the chips modular, i.e. allowing the addition of a second M1 Ultra Duo, or just offering th GPU part, so that one could get to 80 or 120 TFLOPS.
I know that this is by far not the only metric, and Apple's benchmarks have some truth to it. Nonetheless, in raw computational power, the picture is clear.
A RTX3090 has already about 36 TFLOPS in computational power. This is 80% more than the M1 Ultra at 20 TFLOPS.
This is comparable to the ratio between M1 Max and GTX 3080 for laptops.
The mobile RTX3080 is at about 18 TFLOPS, also about 80% more than the M1 Max at 10 TFLOPS.
The M1 Ultra Duo or whatever it will be called in the upcoming Mac Pro will likely have 40 TFLOPS, which is faster than all current Nvidia/AMD cards. However, people who use those, probably have 2-3 in their PC. It will be interesting to see if Apple will try to compete with that, by making the chips modular, i.e. allowing the addition of a second M1 Ultra Duo, or just offering th GPU part, so that one could get to 80 or 120 TFLOPS.
I know that this is by far not the only metric, and Apple's benchmarks have some truth to it. Nonetheless, in raw computational power, the picture is clear.
Wait wait wait. Back the horse up. So you're telling me products to be released later in time will be superior to products currently being released? HOW DOES THIS HAPPEN?!?! Should I just keep waiting for them to stop making stuff so I'm sure the stuff I own is the best stuff? I'll never be able to talk about my stuff at this rate..
I read somewhere that the RTX3090 TFLOPS numbers are computed based on the number of ALUs and clock rates to arrive at the 36 TFLOPS. But due to the design of their architecture, some of the ALUs have dual functions, as in they can either do integer or floating point math, but not both at the same time. So the theoretical 36 TFLOPs cannot be achieved.
I read that is why the AMD RT6900XT with only 24 TFLOPS can beat the RTX3090 in some raster tasks.
So it seems that comparing TFLOPS across architecture may not tell the entire picture.
Add in the fact that Apple's SoC uses TBDR compared to IMR for the RTX3090, I'll not be surprised that the M1 Ultra will come out ahead in raster workload.
On top of course the integrated memory of Apple's M1 is an advantage. 64GB or 128GB of GPU memory is quite a statement, especially for this price.
But for example looking at machine learning, raw computational power together with memory are important factors. That's quite different from rendering an image.
What I find amazing is that the M1 architecture is basically Apple’s first shot at creating a desktop-class GPU and they’ve done a pretty awesome job at supplying performant hardware, competing with the best discrete GPUs out there. If your use case requires this kind of power, you can buy it.
Apple’s GPUs are still gaining in performance significantly with each generation, and they’ve done a new deal with Imagination for a very nice set of raytracing technology, which I am sure we will see incorporated soon.
Apple’s GPUs are still gaining in performance significantly with each generation, and they’ve done a new deal with Imagination for a very nice set of raytracing technology, which I am sure we will see incorporated soon.
I think more importantly is that software are being optimised for the current generation of AS Macs to make them more performant. I think there are a lot of hidden potential even in the M1 SoC (e.g. efficiently using the TBDR architect of the GPU) that has yet to be unlocked by developers.
It basically means nothing nowadays (not much in older days either). Below are the TOPs in INT8, from Xilinx (now AMD) VCK5000 press release. Real-world performance differs largely from that theoretical number.
yes, and think about if the M2 pro max and ultra gain the gpu counts and the gpu core improvement from the A15
Just from that we can get over 40% improved at around the same power consumption
Doesn't matter.
Apple is still on generation 1 of their "desktop" GPUs. All M1 GPUs are basically the same but with different core counts.
The interesting part is to see how Apple will be able to follow the performance with generation 2, 3 etc and if they will be able to leap frog with those in the same way as they did on the iPhone side.
This might lead to apple either just keeping up with Nvidia and AMD (at a much lower power level since they already beat both in perf/W) or they will utterly crush them. But for that only time will tell.
Biggest hurdle right now for Apple is poor implementation from a lot of software houses of their graphics/compute API. If the software houses can sort this Nvidia and AMD are going to get a run for their money.
Apple is still on generation 1 of their "desktop" GPUs. All M1 GPUs are basically the same but with different core counts.
The interesting part is to see how Apple will be able to follow the performance with generation 2, 3 etc and if they will be able to leap frog with those in the same way as they did on the iPhone side.
This might lead to apple either just keeping up with Nvidia and AMD (at a much lower power level since they already beat both in perf/W) or they will utterly crush them. But for that only time will tell.
Biggest hurdle right now for Apple is poor implementation from a lot of software houses of their graphics/compute API. If the software houses can sort this Nvidia and AMD are going to get a run for their money.
Newsflash! In 7 years computers will be way faster!
Without hardware acceleration for ray tracing, any M1 Mac in Blender will be clearly dominated by any RTX GPU.
Thing is you can upgrade the Mac Studio, a PC and Mac Pro you can. The Studio will be paper weight in 3-4 years.
Of course they will...it's a tit for tat race where the net result is the customer benefits. Competition raises standards. At least in this field. Technology moves so fast that anything state of the art this year isn't next year, at the absolute latest. This isn't new.
It won't. It will be within 5-7 years of initial launch like all other non-upgradable Macs, in line with when the new MacOS can no longer be downloaded to it legitimately without some DosDude patch. But in any case, certainly the ultra version is aimed at professional customers who can justify replacing it with whatever the latest thing is every three to four years.
Powerful at what power consumption? I believe Apple is still banking on having more performance per watt.
Power consumption does not matter nearly as much on the desktop side of things. Users who want/need more performance won't mind paying for it with more watts.
I’ll agree that perf/watt matters the least on desktops/workstations as compared to laptops and servers buuut it still affects things like chassis size, cooling, noise, (electricity costs), etc ….
The M1 Ultra CPU is a faster i9/mid-level threadripper with a 3060-3090 GPU depending on circumstances (which graphics option, what workload) which if it were those things it would have to liquid cooled to fit in that sized chassis if it could be at all. Instead it’s air cooled and I’m going to guess pretty quiet.
Further perf/W means you can keep scaling a chip design and still get predictable performance (ie not dropping base clocks) without blowing thermals/power.
It does as electricity is not free. There's also cost in heat management of the room.
The DIY/user replacement market isnt tiny but it’s still much smaller than OEM. And even DIY computers often do nearly full replacements in that time frame as new memory, new sockets, new PCIe are required if you’re trying to keep up with the latest and greatest. It may be amortized a little over time, but if anything that’s not the latest and greatest is a paperweight to you, that’s still going to cost. And OEM is always more expensive than DIY.
Now that’s a hot take. Newer products released in the future will be faster than current available products.
Shocker, who could have seen it coming, pesky future and its faster products.
Shocker, who could have seen it coming, pesky future and its faster products.