As I recall, the "xMac" - and there were other names for it through the years like a plain old mid range headless desktop - was born out of the quadrant of products laid out from Steve Jobs in 1999, or perhaps in 2000:
Jobs basically retired all the Mac products lines, which included a plethora of boxes, save for the quadrant above. No more mid-range box, no pizza box. Everything was cancelled save for the quadrant. xMac was born as a wished for Mac headless desktop because there were a lot of people who didn't want an affordable but un-expandable iMac AIO but couldn't afford a Power Mac G3 tower. The only thing that changed across 20 years was the iMac and the Mac Pro moved up-market, so, the desire for "xMac" remained and the hole only got bigger.
If Apple puts an M1 Pro in the Mac mini for $1200, an M1 Pro for the iMac 24 for $2000, it will come pretty close to a detente for folks. The M1 Pro has a suitably performant GPU, really performant CPU, and has a 32 GB of RAM. Curious why they haven't done it yet. There's definitely some upsell games there, but they are at the stage of the game where they should offer a good Mac product at every single price tier, as well as offering multiple product lines. Ie, a headless desktop isn't the same product as an AIO, and they should have a full product line for both.
tht said: The Mac Studio with a M1 Ultra uses a copper heatsink while the Mac Studio with the M1 Max uses an aluminum heatsink. That accounts for the vast majority of the weight difference.
Kind of disappointing that they resorted to copper. There is still a bit of designing a product from the outside in. So, they choose a Mac mini footprint of 7.7 x 7.x7 inches and very purposefully choose 3.7 inch height before they knew how hot the Ultra was going to be. 3.7 inch is very deliberate because they wanted it to be able to sit underneath a Studio Display. And 7.7 x 7.7 footprint seems sacrosanct because of the Mac mini and the ecosystem developed around it?
Yeah, I've seen some teardowns now. Quite impressive. Interesting about the dimensions. It looks like they did a good job though. On MaxTech they threw about everything they could at both and the fans don't even ramp up. It will be interesting to hear what the actual sound levels are. Quite/loud mean all sorts of things to different people. My 'silent' Blackmagic eGPU, can be heard if the environment is quiet and you know what you're listening for, at least when it gets pushed to the max. But, compared to nearly anything else with a fan, it's silent. I tell people it sounds like a super-faint version of the Starship Enterprise warp engines (even if you can hear it, it is a very non-annoying sound compared to most fan noise).
tht said: Jobs basically retired all the Mac products lines, which included a plethora of boxes, save for the quadrant above. No more mid-range box, no pizza box. Everything was cancelled save for the quadrant. xMac was born as a wished for Mac headless desktop because there were a lot of people who didn't want an affordable but un-expandable iMac AIO but couldn't afford a Power Mac G3 tower. The only thing that changed across 20 years was the iMac and the Mac Pro moved up-market, so, the desire for "xMac" remained and the hole only got bigger.
If Apple puts an M1 Pro in the Mac mini for $1200, an M1 Pro for the iMac 24 for $2000, it will come pretty close to a detente for folks. The M1 Pro has a suitably performant GPU, really performant CPU, and has a 32 GB of RAM. Curious why they haven't done it yet. There's definitely some upsell games there, but they are at the stage of the game where they should offer a good Mac product at every single price tier, as well as offering multiple product lines. Ie, a headless desktop isn't the same product as an AIO, and they should have a full product line for both.
Yeah, I think the xMac was two camps as well. One camp were the expansion/repair/hobbyist people, who wanted a box with slots and user swappable parts.
The other camp are people like me who want more performance than 'consumer' but don't have a Mac Pro budget. The mini w/eGPU is pretty close, but it still isn't great at cooling itself. The Studio hits it out of the park, with the exception of the GPU. I'm really hoping Apple eventually supports eGPUs again, as that would end the debate. But, until then, or Apple matures the GPU, it's fairly powerful, but missing a lot (it doesn't have raytracing hardware, for example). Some uses look like it will be as good or better than a 3080/90, but for other things, it might fall rather flat. That kind of sucks for a machine that expensive.
Fortunately, for most of the work I do, I'm guessing it will fall into some pretty nice performance. For gaming or other things (ie. 3D software... depends on the package, likely), it probably won't do so well. I really, really want one, but might end up waiting a generation or so to see where this goes (as my i7 mini + eGPU is doing fine). If I do get one, I'll maybe not get an upper end one (as much as I want to), so I can more easily to move to it in a 2nd or 3rd gen. Need to see more real-world performance, though.
Huh? I wonder if the Max has less power-supply/cooling/heatsink, etc. than they showed, as it doesn't need them? Interesting.
The Mac Studio with a M1 Ultra uses a copper heatsink while the Mac Studio with the M1 Max uses an aluminum heatsink. That accounts for the vast majority of the weight difference.
Kind of disappointing that they resorted to copper. There is still a bit of designing a product from the outside in. So, they choose a Mac mini footprint of 7.7 x 7.x7 inches and very purposefully choose 3.7 inch height before they knew how hot the Ultra was going to be. 3.7 inch is very deliberate because they wanted it to be able to sit underneath a Studio Display. And 7.7 x 7.7 footprint seems sacrosanct because of the Mac mini and the ecosystem developed around it?
My guess is that Apple prototyped hundreds of configurations for physical dimensions and still ended up with 7.7 x 7.7 because the Mac mini has a longtime datacenter presence. There are miscellaneous other accessories like VESA brackets or under-desk trays.
I'm also guessing that Apple tried a bunch of thermal solutions of aluminum or copper heatsinks and different sized fans/blowers. Sticking with one case size was probably a requirement for Apple. After all >85% of Macs are notebook models, the Mac mini is pretty niche (I'm guessing about 3% of total Mac unit sales) and the higher priced Mac Studio will be even more niche.
For Apple to hit their gross margin targets, they knew they could only have a limited number of physical component differences between the Mac Studio (Max) and Mac Studio (Ultra).
Huh? I wonder if the Max has less power-supply/cooling/heatsink, etc. than they showed, as it doesn't need them? Interesting.
The Mac Studio with a M1 Ultra uses a copper heatsink while the Mac Studio with the M1 Max uses an aluminum heatsink. That accounts for the vast majority of the weight difference.
Kind of disappointing that they resorted to copper. There is still a bit of designing a product from the outside in. So, they choose a Mac mini footprint of 7.7 x 7.x7 inches and very purposefully choose 3.7 inch height before they knew how hot the Ultra was going to be. 3.7 inch is very deliberate because they wanted it to be able to sit underneath a Studio Display. And 7.7 x 7.7 footprint seems sacrosanct because of the Mac mini and the ecosystem developed around it?
Why is it disappointing that they chose copper? Copper is a very effective material for cooling.
The weight of a desktop machine is not of great import.
Huh? I wonder if the Max has less power-supply/cooling/heatsink, etc. than they showed, as it doesn't need them? Interesting.
The Mac Studio with a M1 Ultra uses a copper heatsink while the Mac Studio with the M1 Max uses an aluminum heatsink. That accounts for the vast majority of the weight difference.
Kind of disappointing that they resorted to copper. There is still a bit of designing a product from the outside in. So, they choose a Mac mini footprint of 7.7 x 7.x7 inches and very purposefully choose 3.7 inch height before they knew how hot the Ultra was going to be. 3.7 inch is very deliberate because they wanted it to be able to sit underneath a Studio Display. And 7.7 x 7.7 footprint seems sacrosanct because of the Mac mini and the ecosystem developed around it?
Why is it disappointing that they chose copper? Copper is a very effective material for cooling.
The weight of a desktop machine is not of great import.
Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
mpantone said: For Apple to hit their gross margin targets, they knew they could only have a limited number of physical component differences between the Mac Studio (Max) and Mac Studio (Ultra).
I actually wonder about this. It could end up being pretty popular, and until the iMac gets a 'pro' treatment, it's the only game in town for performance computing before going to the Pro (well, OK, the 16" MBP counts too, now).
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Can I see your data, Mike?
Probably not, unless you have a security clearance. There's a reason I spoke about it in the review in a vague fashion, and why I don't have specific fan-blade speed numbers.
I use the hardware pre-release day at a federal testing lab that I do contract work for. We do that, so our reviews aren't based on two or three days of use, like the pre-seeded by Apple reviews that many YouTubers do. That also gets me some insight on what the DOD and contractors are evaluating the machines for.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Can I see your data, Mike?
Probably not, unless you have a security clearance. There's a reason I spoke about it in the review in a vague fashion, and why I don't have specific fan-blade speed numbers.
I use the hardware pre-release day at a federal testing lab that I do contract work for. We do that, so our reviews aren't based on two or three days of use, like the pre-seeded by Apple reviews that many YouTubers do. That also gets me some insight on what the DOD and contractors are evaluating the machines for.
Did it take milliseconds, seconds, tens of seconds, thousands of seconds?
Was the task similar to one of the following GB5 compute sub benches?
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Can I see your data, Mike?
Probably not, unless you have a security clearance. There's a reason I spoke about it in the review in a vague fashion, and why I don't have specific fan-blade speed numbers.
I use the hardware pre-release day at a federal testing lab that I do contract work for. We do that, so our reviews aren't based on two or three days of use, like the pre-seeded by Apple reviews that many YouTubers do. That also gets me some insight on what the DOD and contractors are evaluating the machines for.
Did it take milliseconds, seconds, tens of seconds, thousands of seconds?
Was the task similar to one of the following GB5 compute sub benches?
Were you able to test on M1 cores with 8 core (M1), 16 cores (M1 Pro), 32 cores (M1 Max), 64 cores (M1 Ultra)?
Hundreds to thousands of seconds, repeated ad infinitum with about a 40-second break between runs; closest to depth of field and horizon detection feature matching; and on M1 Ultra absolutely maxed out in processing regards with a 2TB SSD.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Can I see your data, Mike?
Probably not, unless you have a security clearance. There's a reason I spoke about it in the review in a vague fashion, and why I don't have specific fan-blade speed numbers.
I use the hardware pre-release day at a federal testing lab that I do contract work for. We do that, so our reviews aren't based on two or three days of use, like the pre-seeded by Apple reviews that many YouTubers do. That also gets me some insight on what the DOD and contractors are evaluating the machines for.
Did it take milliseconds, seconds, tens of seconds, thousands of seconds?
Was the task similar to one of the following GB5 compute sub benches?
Were you able to test on M1 cores with 8 core (M1), 16 cores (M1 Pro), 32 cores (M1 Max), 64 cores (M1 Ultra)?
Hundreds to thousands of seconds, repeated ad infinitum with about a 40-second break between runs; closest to depth of field and horizon detection feature matching; and on M1 Ultra absolutely maxed out in processing regards with a 2TB SSD.
We’re you able to test a M1 Max and did performance scale linearly with GPU cores?
An I assume, Metal optimized with Apple’s GPU arch in mind?
Mike Wuerthele said: They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Yeah, I heard you mention that on Space Javelin (I think, or I read it in your review, or both). I'm puzzled by that disparity. I'd wonder if they somehow turned off the fan speed auto setting or something with those utilities, but if they saw it across 3 machines, that probably isn't the case. They claim theirs just stayed at the default fan speeds all the time.
tht said: We’re you able to test a M1 Max and did performance scale linearly with GPU cores?
An I assume, Metal optimized with Apple’s GPU arch in mind?
One thing I was thinking, as I watched the videos and read articles, is how early on we are yet. I wonder how optimized the Metal-optimized stuff really is?
Fortunately, I don't have to make a buying decision yet. As much as I want one, I'm probably best off waiting for the M2 or M3 if I can, so I really see where they are headed with the GPU stuff. Otherwise, this is absolutely the perfect *type* of machine for me.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Can I see your data, Mike?
Probably not, unless you have a security clearance. There's a reason I spoke about it in the review in a vague fashion, and why I don't have specific fan-blade speed numbers.
I use the hardware pre-release day at a federal testing lab that I do contract work for. We do that, so our reviews aren't based on two or three days of use, like the pre-seeded by Apple reviews that many YouTubers do. That also gets me some insight on what the DOD and contractors are evaluating the machines for.
Did it take milliseconds, seconds, tens of seconds, thousands of seconds?
Was the task similar to one of the following GB5 compute sub benches?
Were you able to test on M1 cores with 8 core (M1), 16 cores (M1 Pro), 32 cores (M1 Max), 64 cores (M1 Ultra)?
Hundreds to thousands of seconds, repeated ad infinitum with about a 40-second break between runs; closest to depth of field and horizon detection feature matching; and on M1 Ultra absolutely maxed out in processing regards with a 2TB SSD.
We’re you able to test a M1 Max and did performance scale linearly with GPU cores?
An I assume, Metal optimized with Apple’s GPU arch in mind?
Sorry about the long time between responses. This load was not tested on the Mac Studio with M1 Max, but was on the 16-inch MBP with M1 Max. It's not quite a linear scale, but it's more linear, than, say, a second whole Intel processor versus one.
Mike Wuerthele said: They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Yeah, I heard you mention that on Space Javelin (I think, or I read it in your review, or both). I'm puzzled by that disparity. I'd wonder if they somehow turned off the fan speed auto setting or something with those utilities, but if they saw it across 3 machines, that probably isn't the case. They claim theirs just stayed at the default fan speeds all the time.
tht said: We’re you able to test a M1 Max and did performance scale linearly with GPU cores?
An I assume, Metal optimized with Apple’s GPU arch in mind?
One thing I was thinking, as I watched the videos and read articles, is how early on we are yet. I wonder how optimized the Metal-optimized stuff really is?
Fortunately, I don't have to make a buying decision yet. As much as I want one, I'm probably best off waiting for the M2 or M3 if I can, so I really see where they are headed with the GPU stuff. Otherwise, this is absolutely the perfect *type* of machine for me.
There were some tests here showing the Mac Studio power usage:
During the gaming test with Shadow of the Tomb Raider, which uses Metal, it uses 130W. During the Blender test, it used 70W on just the GPU but 120-130W with both CPU/GPU and oddly around 90W with just CPU, although it is drawing into a framebuffer so the GPU is still being used a little.
The graphs at the end of the gaming video show a small difference between Max and Ultra at 1080p but close to double at 60FPS vs 33FPS when running at 4k. The same can be seen comparing a 3060 laptop against a 3090 desktop:
GTA V at medium ~720p is around the same result on 3060 and 3090 but over 2x faster on the 3090 at 4k.
Some tests will be CPU bound and not able to send to the GPU often enough to max it out. Some Mac apps are still Rosetta.
The M1 was roughly 10W each for CPU/GPU (4 performance cores, 8 GPU cores) and the Ultra is 4x CPU, 8x GPU (16 performance cores, 64 GPU cores). It would be expected that at full load, the Mac Studio would be 40W CPU and 80W GPU or 120W for the package plus some overhead and the above tests show this.
Apple says that the lower Ultra has a peak power usage of 215W, that may include powering devices over Thunderbolt:
Given that some tests show a 2x speedup between Ultra and Max suggests it is scaling ok but some software will be held back being CPU bound. The main thing holding back some of the 3D tests is lack of hardware raytracing but it is being compared to computers using as much as 5x the power.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Can I see your data, Mike?
Probably not, unless you have a security clearance. There's a reason I spoke about it in the review in a vague fashion, and why I don't have specific fan-blade speed numbers.
I use the hardware pre-release day at a federal testing lab that I do contract work for. We do that, so our reviews aren't based on two or three days of use, like the pre-seeded by Apple reviews that many YouTubers do. That also gets me some insight on what the DOD and contractors are evaluating the machines for.
Did it take milliseconds, seconds, tens of seconds, thousands of seconds?
Was the task similar to one of the following GB5 compute sub benches?
Were you able to test on M1 cores with 8 core (M1), 16 cores (M1 Pro), 32 cores (M1 Max), 64 cores (M1 Ultra)?
Hundreds to thousands of seconds, repeated ad infinitum with about a 40-second break between runs; closest to depth of field and horizon detection feature matching; and on M1 Ultra absolutely maxed out in processing regards with a 2TB SSD.
We’re you able to test a M1 Max and did performance scale linearly with GPU cores?
An I assume, Metal optimized with Apple’s GPU arch in mind?
Sorry about the long time between responses. This load was not tested on the Mac Studio with M1 Max, but was on the 16-inch MBP with M1 Max. It's not quite a linear scale, but it's more linear, than, say, a second whole Intel processor versus one.
Yes, Metal optimized. Bespoke code, as it were.
No, no. No apologies needed. Thanks to you as you are the one who's providing actual data and doing us a favor here.
Can you give number on how much the fall off with GPU cores is?
During the gaming test with Shadow of the Tomb Raider, which uses Metal, it uses 130W. During the Blender test, it used 70W on just the GPU but 120-130W with both CPU/GPU and oddly around 90W with just CPU, although it is drawing into a framebuffer so the GPU is still being used a little. ... The M1 was roughly 10W each for CPU/GPU (4 performance cores, 8 GPU cores) and the Ultra is 4x CPU, 8x GPU (16 performance cores, 64 GPU cores). It would be expected that at full load, the Mac Studio would be 40W CPU and 80W GPU or 120W for the package plus some overhead and the above tests show this.
Apple says that the lower Ultra has a peak power usage of 215W, that may include powering devices over Thunderbolt: ... Given that some tests show a 2x speedup between Ultra and Max suggests it is scaling ok but some software will be held back being CPU bound. The main thing holding back some of the 3D tests is lack of hardware raytracing but it is being compared to computers using as much as 5x the power.
The Tally Ho Tech videos (he probably shouldn't be proud that his logo branding is illegible) measured power from the wall. The CPU itself wasn't using 90 W. It was probably 60 W or less. The machine idles at 15 W: so logic board, fans, whatever was being used for the monitor, so on and so forth. It looked like he was running his tests from an external SSD, so 5 to 15 Watts for that while actively used, and he had something else connected to the front TB/USBC port. RAM will also take a nontrivial amount of power, 5 to 10 W. The 370 W power supply probably loses 5 to 15 W just converting AC to DC. So, that 90 W at the wall can rapidly go down to 45 W at the CPU cores.
Similar story for the GPU testing. It wasn't using 130 W. That was a measurement at the wall. Some amount of CPU is used for the test, in addition to all the aforementioned users of power in the box. So that 130 W probably meant 60 to 80 W were being used by the GPU cores.
It's good that some tests are showing that there is a near-linear performance improvement with increasing GPU core counts, out to 64 cores. Some GPU compute process are likely bottlenecked by non GPU related areas. Some GPU compute processes aren't really optimized for Apple's GPU architecture. The are compiled and run on Apple Silicon with Metal APIs, but aren't designed for Apple GPUs so there is a lot of performance left off the table. Apple GPUs with more than 8 cores is basically fresh off the market for 5 months now. It probably takes 1 to 2 years for software to actually be optimized for the architecture. Software is where Apple is really behind.
Apple's document says:
Power consumption data (Watts) is measured from the wall power source and includes all power supply and system losses. Additional correction is not needed.
"Max" is defined as the maximum possible power draw based on the computer's power supply rating.
"Idle" reflects the power used with only Finder open, using the default power management settings.
The 215 W "Max" number for a Mac Studio with "M1 Ultra 20-Core CPU & 48-Core GPU, 64GB unified memory, 1TB SSD". If it included power from all the ports, this M1 Ultra package is basically sipping power at max around 140 W or so. 10 W per USB port is 60 W, plus SD Card and ethernet, gets to a total of 70 W? Though Apple's definition of "Max" is strange.
Well, at least it looks like they have some room to grow in terms of power consumption. Another 16 GPU cores, 64 GB RAM and 7 TB of NAND can't be using that much more power, and it has a 370 W PSU.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Can I see your data, Mike?
Probably not, unless you have a security clearance. There's a reason I spoke about it in the review in a vague fashion, and why I don't have specific fan-blade speed numbers.
I use the hardware pre-release day at a federal testing lab that I do contract work for. We do that, so our reviews aren't based on two or three days of use, like the pre-seeded by Apple reviews that many YouTubers do. That also gets me some insight on what the DOD and contractors are evaluating the machines for.
Did it take milliseconds, seconds, tens of seconds, thousands of seconds?
Was the task similar to one of the following GB5 compute sub benches?
Were you able to test on M1 cores with 8 core (M1), 16 cores (M1 Pro), 32 cores (M1 Max), 64 cores (M1 Ultra)?
Hundreds to thousands of seconds, repeated ad infinitum with about a 40-second break between runs; closest to depth of field and horizon detection feature matching; and on M1 Ultra absolutely maxed out in processing regards with a 2TB SSD.
We’re you able to test a M1 Max and did performance scale linearly with GPU cores?
An I assume, Metal optimized with Apple’s GPU arch in mind?
Sorry about the long time between responses. This load was not tested on the Mac Studio with M1 Max, but was on the 16-inch MBP with M1 Max. It's not quite a linear scale, but it's more linear, than, say, a second whole Intel processor versus one.
Yes, Metal optimized. Bespoke code, as it were.
No, no. No apologies needed. Thanks to you as you are the one who's providing actual data and doing us a favor here.
Can you give number on how much the fall off with GPU cores is?
I'm going to try and get some clarity on what I can and cannot speak about, as it pertains to this workload. I'm toeing up to a line right now, and I'd rather a forum post get misconstrued or get any of the folks I work with in trouble.
tht said: Using copper implies they don't have any headroom left in the Mac Studio to cool future, hotter SoCs. If the Mac Studio was longer (deeper), they could have managed with aluminum, leaving copper as solution for components that are hotter in the future. I suppose they still have the noisier higher RPM fans as an option still. As I was saying before, a square footprint is not optimal for a cooling design.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
Yes, the performance of the Apple Silicon GPUs have been unexpected. Max Tech is also showing that the GPUs aren't using the advertised power levels either. Not by a little, but by a lot. Maxing out the M1 Ultra GPU should hit about 110 W based on Apple's charts, but they were never able to get close. About 80 W.
They didn't try hard enough, then. The audio processing that I spoke about in my review hit 104W, with an clear increase in fan speeds.
Can I see your data, Mike?
Probably not, unless you have a security clearance. There's a reason I spoke about it in the review in a vague fashion, and why I don't have specific fan-blade speed numbers.
I use the hardware pre-release day at a federal testing lab that I do contract work for. We do that, so our reviews aren't based on two or three days of use, like the pre-seeded by Apple reviews that many YouTubers do. That also gets me some insight on what the DOD and contractors are evaluating the machines for.
Did it take milliseconds, seconds, tens of seconds, thousands of seconds?
Was the task similar to one of the following GB5 compute sub benches?
Were you able to test on M1 cores with 8 core (M1), 16 cores (M1 Pro), 32 cores (M1 Max), 64 cores (M1 Ultra)?
Hundreds to thousands of seconds, repeated ad infinitum with about a 40-second break between runs; closest to depth of field and horizon detection feature matching; and on M1 Ultra absolutely maxed out in processing regards with a 2TB SSD.
We’re you able to test a M1 Max and did performance scale linearly with GPU cores?
An I assume, Metal optimized with Apple’s GPU arch in mind?
Sorry about the long time between responses. This load was not tested on the Mac Studio with M1 Max, but was on the 16-inch MBP with M1 Max. It's not quite a linear scale, but it's more linear, than, say, a second whole Intel processor versus one.
Yes, Metal optimized. Bespoke code, as it were.
No, no. No apologies needed. Thanks to you as you are the one who's providing actual data and doing us a favor here.
Can you give number on how much the fall off with GPU cores is?
I'm going to try and get some clarity on what I can and cannot speak about, as it pertains to this workload. I'm toeing up to a line right now, and I'd rather a forum post get misconstrued or get any of the folks I work with in trouble.
Comments
Jobs basically retired all the Mac products lines, which included a plethora of boxes, save for the quadrant above. No more mid-range box, no pizza box. Everything was cancelled save for the quadrant. xMac was born as a wished for Mac headless desktop because there were a lot of people who didn't want an affordable but un-expandable iMac AIO but couldn't afford a Power Mac G3 tower. The only thing that changed across 20 years was the iMac and the Mac Pro moved up-market, so, the desire for "xMac" remained and the hole only got bigger.
If Apple puts an M1 Pro in the Mac mini for $1200, an M1 Pro for the iMac 24 for $2000, it will come pretty close to a detente for folks. The M1 Pro has a suitably performant GPU, really performant CPU, and has a 32 GB of RAM. Curious why they haven't done it yet. There's definitely some upsell games there, but they are at the stage of the game where they should offer a good Mac product at every single price tier, as well as offering multiple product lines. Ie, a headless desktop isn't the same product as an AIO, and they should have a full product line for both.
Yeah, I think the xMac was two camps as well. One camp were the expansion/repair/hobbyist people, who wanted a box with slots and user swappable parts.
The other camp are people like me who want more performance than 'consumer' but don't have a Mac Pro budget. The mini w/eGPU is pretty close, but it still isn't great at cooling itself. The Studio hits it out of the park, with the exception of the GPU. I'm really hoping Apple eventually supports eGPUs again, as that would end the debate. But, until then, or Apple matures the GPU, it's fairly powerful, but missing a lot (it doesn't have raytracing hardware, for example). Some uses look like it will be as good or better than a 3080/90, but for other things, it might fall rather flat. That kind of sucks for a machine that expensive.
Fortunately, for most of the work I do, I'm guessing it will fall into some pretty nice performance. For gaming or other things (ie. 3D software... depends on the package, likely), it probably won't do so well. I really, really want one, but might end up waiting a generation or so to see where this goes (as my i7 mini + eGPU is doing fine). If I do get one, I'll maybe not get an upper end one (as much as I want to), so I can more easily to move to it in a 2nd or 3rd gen. Need to see more real-world performance, though.
I'm also guessing that Apple tried a bunch of thermal solutions of aluminum or copper heatsinks and different sized fans/blowers. Sticking with one case size was probably a requirement for Apple. After all >85% of Macs are notebook models, the Mac mini is pretty niche (I'm guessing about 3% of total Mac unit sales) and the higher priced Mac Studio will be even more niche.
For Apple to hit their gross margin targets, they knew they could only have a limited number of physical component differences between the Mac Studio (Max) and Mac Studio (Ultra).
The weight of a desktop machine is not of great import.
MaxTech had a video there they tried their best to stress both the Mac and Ultra versions of the Studio, and neither even moved the fan speeds up. Maybe there is more cooling capacity there than we realize.
I use the hardware pre-release day at a federal testing lab that I do contract work for. We do that, so our reviews aren't based on two or three days of use, like the pre-seeded by Apple reviews that many YouTubers do. That also gets me some insight on what the DOD and contractors are evaluating the machines for.
Was the task similar to one of the following GB5 compute sub benches?
Were you able to test on M1 cores with 8 core (M1), 16 cores (M1 Pro), 32 cores (M1 Max), 64 cores (M1 Ultra)?
An I assume, Metal optimized with Apple’s GPU arch in mind?
One thing I was thinking, as I watched the videos and read articles, is how early on we are yet. I wonder how optimized the Metal-optimized stuff really is?
Fortunately, I don't have to make a buying decision yet. As much as I want one, I'm probably best off waiting for the M2 or M3 if I can, so I really see where they are headed with the GPU stuff. Otherwise, this is absolutely the perfect *type* of machine for me.
Yes, Metal optimized. Bespoke code, as it were.
During the gaming test with Shadow of the Tomb Raider, which uses Metal, it uses 130W.
During the Blender test, it used 70W on just the GPU but 120-130W with both CPU/GPU and oddly around 90W with just CPU, although it is drawing into a framebuffer so the GPU is still being used a little.
The graphs at the end of the gaming video show a small difference between Max and Ultra at 1080p but close to double at 60FPS vs 33FPS when running at 4k. The same can be seen comparing a 3060 laptop against a 3090 desktop:
https://www.notebookcheck.net/GeForce-RTX-3060-Laptop-GPU-vs-NVIDIA-GeForce-RTX-3090_10478_10485.247598.0.html
GTA V at medium ~720p is around the same result on 3060 and 3090 but over 2x faster on the 3090 at 4k.
Some tests will be CPU bound and not able to send to the GPU often enough to max it out. Some Mac apps are still Rosetta.
The M1 was roughly 10W each for CPU/GPU (4 performance cores, 8 GPU cores) and the Ultra is 4x CPU, 8x GPU (16 performance cores, 64 GPU cores). It would be expected that at full load, the Mac Studio would be 40W CPU and 80W GPU or 120W for the package plus some overhead and the above tests show this.
Apple says that the lower Ultra has a peak power usage of 215W, that may include powering devices over Thunderbolt:
https://support.apple.com/en-us/HT213100
Given that some tests show a 2x speedup between Ultra and Max suggests it is scaling ok but some software will be held back being CPU bound. The main thing holding back some of the 3D tests is lack of hardware raytracing but it is being compared to computers using as much as 5x the power.
Can you give number on how much the fall off with GPU cores is?
Similar story for the GPU testing. It wasn't using 130 W. That was a measurement at the wall. Some amount of CPU is used for the test, in addition to all the aforementioned users of power in the box. So that 130 W probably meant 60 to 80 W were being used by the GPU cores.
It's good that some tests are showing that there is a near-linear performance improvement with increasing GPU core counts, out to 64 cores. Some GPU compute process are likely bottlenecked by non GPU related areas. Some GPU compute processes aren't really optimized for Apple's GPU architecture. The are compiled and run on Apple Silicon with Metal APIs, but aren't designed for Apple GPUs so there is a lot of performance left off the table. Apple GPUs with more than 8 cores is basically fresh off the market for 5 months now. It probably takes 1 to 2 years for software to actually be optimized for the architecture. Software is where Apple is really behind.
Apple's document says:
Well, at least it looks like they have some room to grow in terms of power consumption. Another 16 GPU cores, 64 GB RAM and 7 TB of NAND can't be using that much more power, and it has a 370 W PSU.