Anyone else feel the M1 Ultra could be a bit faster?

[sext-scientist]

Don't get me wrong the M1 Ultra is a good chip, but it's not quite the "knock it out of the park" you'd expect in today's competitive computing world.

Just to give an example, the 20-core M1 Ultra with a 3.2ghz turbo scores about a 40,000 for multi-core on passmark's benchmark. Source

For comparison, the Ryzen Threadripper W3960X with 3.8ghz turbo scores 55,000 for multi-core on this benchmark by comparison: Source

Likewise the 20-core Ultra's GPU scores 95,000 on Geekbench's test, while an AMD 6900XT scores 165,000 on the same test: Source

The AMD CPU and GPU can be had for $1450, and $850 respectively. Source [CPU], Source [GPU]

Bottom line is the M1 Ultra seems to be 50% more expensive once you add up the rest of the system at $4999, than alternatives which are 50% faster. Yes it uses 3 times less power for a given performance but it just doesn't seem very competitive for an "Ultra" workstation.

Should Apple have stuck 3 M1 chips together and done a 30-core to make the Ultra moniker hit home? Is the second generation going to be much faster?

 

[AltecX]

Really the big impressiveness is what it does with so little power draw, heat generation and such small size. That said I feel Apple could have clocked it up a bit higher as it's not a chip that's going into a battery powered device.

Personally, I think the Pro chips should have included the Max branded chips as well, and the Ultra should be called M1 Studio as the Desktop variation. I just find Ultra to be a very 2001 sounding term, and Max makes me just think of a bigger phone.

 

[MediaGary]

Superlatives aren't really useful in a world of constant technological improvements. Every marketing team would like to name everything Maximum, Super, Ultra, etc. all the time. The trouble with that pattern is, as so eloquently stated in the 'The Incredibles', "When everyone's super, [muwahaha] no one will be."

That's the trouble we're having with USB naming when the word SuperSpeed became attached to the 5Gbps implementation and then stayed for 10Gbps, and now is applied with 20Gbps. It also happens in the other direction, with so many backward-looking logos stuck onto SD cards to confirm that this 2022 product is faster than a donkey cart. We know, we know...

Let's just watch the benchmarks, consider price, power consumption, likely computing needs, and enjoy what's available to us. If we all limit our emotional bonding to beings that can love us back, and limit our highest pride-of-ownership to things that we've built/designed/modified, then overall tech-life happiness can improve.

Since this is becoming a rant, to give a sort of answer to your original question; well...
I think the Ultra is an absolutely great product, and the base model is priced $500 more than I would have thought as "worth it". I also think that it's too early to see what the full software/firmware exploitation of its capabilities will be. I live in hope that there will be M2 models that cover the chasm between the Max and Ultra while remaining silent.

 

[AltecX]

Superlatives aren't really useful in a world of constant technological improvements. Every marketing team would like to name everything Maximum, Super, Ultra, etc. all the time. The trouble with that pattern is, as so eloquently stated in the 'The Incredibles', "When everyone's super, [muwahaha] no one will be."

That's the trouble we're having with USB naming when the word SuperSpeed became attached to the 5Gbps implementation and then stayed for 10Gbps, and now is applied with 20Gbps. It also happens in the other direction, with so many backward-looking logos stuck onto SD cards to confirm that this 2022 product is faster than a donkey cart. We know, we know...

Let's just watch the benchmarks, consider price, power consumption, likely computing needs, and enjoy what's available to us. If we all limit our emotional bonding to beings that can love us back, and limit our highest pride-of-ownership to things that we've built/designed/modified, then overall tech-life happiness can improve.

Since this is becoming a rant, to give a sort of answer to your original question; well...
I think the Ultra is an absolutely great product, and the base model is priced $500 more than I would have thought as "worth it". I also think that it's too early to see what the full software/firmware exploitation of its capabilities will be. I live in hope that there will be M2 models that cover the chasm between the Max and Ultra while remaining silent.

Yes in my mine it just makes sense that M1 would be the layman's chip for mobile, Pro for mobile professionals/creators and layman's desktop and Studio is for heavy lifters. Keeps it simple to me with out try to sound Supermega awesome powersex fast!

 

[spiderman0616]

I have used the M1 MacBook Air, the M1 Mac mini, the M1 Pro 14" MacBook Pro, and the M1 Pro 16" MacBook Pro, all pretty extensively. Not once during using any of those machines at any time have I thought, "I wish this thing ran faster." I haven't thrown a game or project at ANY of them yet that they didn't chew up and spit out.

 

[terminator-jq]

I can’t really speak on the M1 Ultra but as an M1 Max owner, I think it’s very important to remember that:

1. Benchmarks aren’t everything and most benchmark software isn’t fully optimized for Apple Silicon.

2. Real world use of the M1 Max and M1 Ultra show that these chips are absolute screamers in the CPU department. The GPU side is where Apple definitely has some work to do but even for GPU task, the M1 family definitely shows steps in the right direction. I just finished rendering an entire 3D animation project on my M1 Max without an issue.

3. We are still very early in Apple Silicon both for hardware and software. It may take a couple more years for more software companies to really take advantage of these chips. On the hardware side, obviously the M1 family is the first of its kind. We can only go up from here. No doubt we may see some hints of what’s coming next for Apple Silicon when Apple reveals the next Mac Pro.

4. Last point worth mentioning is just because something supports Apple Silicon doesn’t mean it’s fully optimized. This means over time, the programs we are already using may get faster. One example is Blender. Version 3.1 shipped with Apple Silicon GPU rendering. However, an Apple engineer helping with the project said in previous test builds they were getting up to 2X the performance compared to the launch version. It’s all about optimization and more is on the way.

 

[leman]

Two notes:

1. Passmark is a very poor benchmark. Using an industry benchmark like SPEC the Ultra should be around a 32-core threadripper (it’s a shame Anabdtech didn’t do an Ultra review after Andrei F. Left).

2. Geekbench compute underestimate Apple GPUs since these tests are very short and Apple needs a longish warmup period to trigger the max performance mode. The 6900XT should deliver around 23 peak theoretical TFLOPs compared to Ultras 20 TFLOPs.

 

[kschendel]

I think your price comparisons are a bit unfair. Sure, a 3960X is $1450, but have you priced sTR4 motherboards recently? and you don't have any video yet. Similarly, a 6900XT might be $900 (on sale), but you don't have any CPU yet. The Ultra includes both. Put together a real system with a 3960X CPU and a 6900XT GPU, the 1.2kW power supply that you'll need, and the motherboard, memory, CPU cooling, and storage bits, and I bet you aren't actually all that far behind the M1 Ultra, price-wise. And you don't have any software / OS yet.

It's fair to complain about the "ultra" name, if you like. It's not a world beater along any single metric. What it does give you is very competitive performance, in a small package, with low power draw, in the Mac ecosystem. Actually pretty remarkable for a first iteration of the Apple ARM series.

 

[Adult80HD]

I have two Ultra's and one Max, as well as two Ryzen Threadripper Pro workstations (32-core and 64-core), as well a 16" MacBook Pro Max and two Mac Pros (one 28-core and one 16-core). Yes, I have a computer addiction, but these are all for my business and used by various employees. When they first arrived in March, I benchmarked some of the common tasks we use them for here, mostly revolving around digital photography and editing. I've personally found that the synthetic benchmarks are pretty much useless for evaluating real-world performance at tasks that matter to me.

What I found is that the Ultra outperformed all of the other systems for most of the tasks I do, with only one exception--the 64-core AMD had a slight edge in exporting a huge number of files (this is a CPU-bound task that is also optimized for multiple cores). It was an impressive showing, and I retired my 28-core Mac Pro with a W6800X Duo to one of my staff and replaced with a loaded Ultra. What's great is that's also just crazy fast at all of the regular day-to-day tasks I throw at it. All of this in a tiny package that is nearly silent. Compared to the AMDs which have loud fans and heat up an entire room quickly--or the Mac Pros which are nearly silent but also heat up a room. Fabulous performance per watt, and that's got real-world benefits.

 

[spiderman0616]

I have two Ultra's and one Max, as well as two Ryzen Threadripper Pro workstations (32-core and 64-core), as well a 16" MacBook Pro Max and two Mac Pros (one 28-core and one 16-core). Yes, I have a computer addiction, but these are all for my business and used by various employees. When they first arrived in March, I benchmarked some of the common tasks we use them for here, mostly revolving around digital photography and editing. I've personally found that the synthetic benchmarks are pretty much useless for evaluating real-world performance at tasks that matter to me.

What I found is that the Ultra outperformed all of the other systems for most of the tasks I do, with only one exception--the 64-core AMD had a slight edge in exporting a huge number of files (this is a CPU-bound task that is also optimized for multiple cores). It was an impressive showing, and I retired my 28-core Mac Pro with a W6800X Duo to one of my staff and replaced with a loaded Ultra. What's great is that's also just crazy fast at all of the regular day-to-day tasks I throw at it. All of this in a tiny package that is nearly silent. Compared to the AMDs which have loud fans and heat up an entire room quickly--or the Mac Pros which are nearly silent but also heat up a room. Fabulous performance per watt, and that's got real-world benefits.

Really cool that you're in a business where you can put things through their paces like that!!! It's nice to hear that real world workloads are giving evidence that these chips are the real deal, not that I didn't already believe it. I feel like I'm carrying around a portable desktop computer rather than a laptop these days. The only two apps I've used so far that have even gotten the fans to kick on were Civ VI and Psychonauts 2, both at near maximum or maximum settings. (The games themselves both run great, despite the fact that they're pushing the system pretty hard.)

I haven't gotten to do any photo/video/audio editing on my M1 Pros yet, but I'm assuming it's amazing based on what I saw with my M1 Air.

 

[darngooddesign]

Don't get me wrong the M1 Ultra is a good chip, but it's not quite the "knock it out of the park" you'd expect in today's competitive computing world.

Just to give an example, the 20-core M1 Ultra with a 3.2ghz turbo scores about a 40,000 for multi-core on passmark's benchmark. Source

For comparison, the Ryzen Threadripper W3960X with 3.8ghz turbo scores 55,000 for multi-core on this benchmark by comparison: Source

Likewise the 20-core Ultra's GPU scores 95,000 on Geekbench's test, while an AMD 6900XT scores 165,000 on the same test: Source

The AMD CPU and GPU can be had for $1450, and $850 respectively. Source [CPU], Source [GPU]

Bottom line is the M1 Ultra seems to be 50% more expensive once you add up the rest of the system at $4999, than alternatives which are 50% faster. Yes it uses 3 times less power for a given performance but it just doesn't seem very competitive for an "Ultra" workstation.

Should Apple have stuck 3 M1 chips together and done a 30-core to make the Ultra moniker hit home? Is the second generation going to be much faster?

Apple is still going to drop an UltraDuo or UltraQuad in the upcoming MacPro; let's wait for those numbers.

 

[Kelly Jones]

Don't get me wrong the M1 Ultra is a good chip, but it's not quite the "knock it out of the park" you'd expect in today's competitive computing world.

Just to give an example, the 20-core M1 Ultra with a 3.2ghz turbo scores about a 40,000 for multi-core on passmark's benchmark. Source

For comparison, the Ryzen Threadripper W3960X with 3.8ghz turbo scores 55,000 for multi-core on this benchmark by comparison: Source

Likewise the 20-core Ultra's GPU scores 95,000 on Geekbench's test, while an AMD 6900XT scores 165,000 on the same test: Source

The AMD CPU and GPU can be had for $1450, and $850 respectively. Source [CPU], Source [GPU]

Bottom line is the M1 Ultra seems to be 50% more expensive once you add up the rest of the system at $4999, than alternatives which are 50% faster. Yes it uses 3 times less power for a given performance but it just doesn't seem very competitive for an "Ultra" workstation.

Should Apple have stuck 3 M1 chips together and done a 30-core to make the Ultra moniker hit home? Is the second generation going to be much faster?

I think for certain workloads and environments, you are correct. Plus you have the ability to swap in a more powerful GPU at a later date and you can upgrade the RAM and storage. However, I think Apple's target market is one where a small form factor and quiet operation are also important. Maybe an ad agency or creative department or a video production studio where the main emphasis is not 3D rendering or complicated AI models. For my purposes, I would probably forego the extra GPUs and save $1000 to get the $3900 model. Also note that the cost of the other components of the AMD system are not trivial.

In my opinion, if you need the things that the AMD system offers, then just get it; it is more cost effective.

 

[Jumpthesnark]

I have two Ultra's and one Max, as well as two Ryzen Threadripper Pro workstations (32-core and 64-core), as well a 16" MacBook Pro Max and two Mac Pros (one 28-core and one 16-core). Yes, I have a computer addiction, but these are all for my business and used by various employees. When they first arrived in March, I benchmarked some of the common tasks we use them for here, mostly revolving around digital photography and editing. I've personally found that the synthetic benchmarks are pretty much useless for evaluating real-world performance at tasks that matter to me.

What I found is that the Ultra outperformed all of the other systems for most of the tasks I do, with only one exception--the 64-core AMD had a slight edge in exporting a huge number of files (this is a CPU-bound task that is also optimized for multiple cores). It was an impressive showing, and I retired my 28-core Mac Pro with a W6800X Duo to one of my staff and replaced with a loaded Ultra. What's great is that's also just crazy fast at all of the regular day-to-day tasks I throw at it. All of this in a tiny package that is nearly silent. Compared to the AMDs which have loud fans and heat up an entire room quickly--or the Mac Pros which are nearly silent but also heat up a room. Fabulous performance per watt, and that's got real-world benefits.

That's great to hear. I'm a photographer and photo editor, using Photo Mechanic, Lightroom and Photoshop for my work. Pretty basic stuff. I'm looking forward to upgrading from my Intel-based MacBook Pro and iMac 5K Retina machines to something in the Apple Silicon ecosystem. Are those the applications you're using (especially Lightroom for large exports)? Thanks!

 

[mi7chy]

TR 3960X and 6900xt are overkill. $550 5950x (46193 CPU Mark) and $579 6800 (129800 Metal) are still faster while being virtually silent, low power (~300W total system power consumption from wall under full load) and ~$2K total system cost.

 

[tmoerel]

TR 3960X and 6900xt are overkill. $550 5950x (46193 CPU Mark) and $579 6800 (129800 Metal) are still faster while being virtually silent, low power (~300W total system power consumption from wall under full load) and ~$2K total system cost.

These benchmark results say only little. Real life usage is what matters, and that is where the M1 series excel as they have all kinds of off-CPU accelerators.

 

[Adult80HD]

That's great to hear. I'm a photographer and photo editor, using Photo Mechanic, Lightroom and Photoshop for my work. Pretty basic stuff. I'm looking forward to upgrading from my Intel-based MacBook Pro and iMac 5K Retina machines to something in the Apple Silicon ecosystem. Are those the applications you're using (especially Lightroom for large exports)? Thanks!

Lightroom for almost everything, but also Topaz Sharpen and DeNoise. Those products are crazy fast on Apple Silicon. I do a lot of retouching with Lightroom, so memory helps. I'd suggest 64GB of RAM as the minimum if you're working with a modern high-megapixel camera, or if you can spring for it, 128GB. Not really needed now, but when I run a lot of apps simultaneously, it's definitely snappier.

 

[Jumpthesnark]

Lightroom for almost everything, but also Topaz Sharpen and DeNoise. Those products are crazy fast on Apple Silicon. I do a lot of retouching with Lightroom, so memory helps. I'd suggest 64GB of RAM as the minimum if you're working with a modern high-megapixel camera, or if you can spring for it, 128GB. Not really needed now, but when I run a lot of apps simultaneously, it's definitely snappier.

Thanks! That's very helpful, I appreciate it. I'll add that info to my shopping wish list for when it's time to spring for new hardware.

 

[l0stl0rd]

Don't get me wrong the M1 Ultra is a good chip, but it's not quite the "knock it out of the park" you'd expect in today's competitive computing world.

Just to give an example, the 20-core M1 Ultra with a 3.2ghz turbo scores about a 40,000 for multi-core on passmark's benchmark. Source

For comparison, the Ryzen Threadripper W3960X with 3.8ghz turbo scores 55,000 for multi-core on this benchmark by comparison: Source

Likewise the 20-core Ultra's GPU scores 95,000 on Geekbench's test, while an AMD 6900XT scores 165,000 on the same test: Source

The AMD CPU and GPU can be had for $1450, and $850 respectively. Source [CPU], Source [GPU]

Bottom line is the M1 Ultra seems to be 50% more expensive once you add up the rest of the system at $4999, than alternatives which are 50% faster. Yes it uses 3 times less power for a given performance but it just doesn't seem very competitive for an "Ultra" workstation.

Should Apple have stuck 3 M1 chips together and done a 30-core to make the Ultra moniker hit home? Is the second generation going to be much faster?

I did not buy if for that reason, perhaps if it was like 1k cheaper.

CPU seems not bad but the GPU scales really bad.

 

[kpluck]

A CPU(GPU) is a collection of specifications that all affect the chip's suitability for a particular purpose. In your case, if that purpose allows you to focus solely on raw number crunching performance per dollar while ignoring power consumption and OS compatibility, Apple Silicon is clearly not the best choice.

I doubt Apple has a goal of making a computer that beats all comers on a simple price/performance comparison. That is simply not one of their goals. And while we still haven't seen their take on a workstation class AS machine, I don't see the AS Mac Pro changing that.

-kp

 

[joema2]

...What I found is that the Ultra outperformed all of the other systems for most of the tasks I do...also just crazy fast at all of the regular day-to-day tasks I throw at it. All of this in a tiny package that is nearly silent....

My documentary team has three Ultras and in general that matches our experience. For those doing real work with real applications the user response is usually "I can't believe how fast it is".

In my own tests the Ultra is about 3x faster on Neat Video 5.5.2(SR) noise reduction than my 10-core Vega 64 iMac Pro. On DaVinci Resolve Studio 17.4.6 face refinement and magic mask rotoscoping, it's about 2x faster. Transcoding in FCP 10.6.3 from 4k ProRes 422 to 50% ProRes Proxy, it's about 3.8x faster.

That said, there are tasks where the Max/Ultra could be faster. It is apparently not using the multiple decode/encode engines in parallel, which would be especially useful on 4k/8k H264/HEVC. My guess is software updates are needed at the application layer, framework layer and system layer to handle multiple accelerators in a thread-safe manner. But even without this it's pretty fast.

 

[eicca]

Use a classic Mac Pro as your daily for a while. Then even the base M1 will feel like a Formula 1 car.

 

[alien3dx]

I can’t really speak on the M1 Ultra but as an M1 Max owner, I think it’s very important to remember that:

1. Benchmarks aren’t everything and most benchmark software isn’t fully optimized for Apple Silicon.

2. Real world use of the M1 Max and M1 Ultra show that these chips are absolute screamers in the CPU department. The GPU side is where Apple definitely has some work to do but even for GPU task, the M1 family definitely shows steps in the right direction. I just finished rendering an entire 3D animation project on my M1 Max without an issue.

3. We are still very early in Apple Silicon both for hardware and software. It may take a couple more years for more software companies to really take advantage of these chips. On the hardware side, obviously the M1 family is the first of its kind. We can only go up from here. No doubt we may see some hints of what’s coming next for Apple Silicon when Apple reveals the next Mac Pro.

4. Last point worth mentioning is just because something supports Apple Silicon doesn’t mean it’s fully optimized. This means over time, the programs we are already using may get faster. One example is Blender. Version 3.1 shipped with Apple Silicon GPU rendering. However, an Apple engineer helping with the project said in previous test builds they were getting up to 2X the performance compared to the launch version. It’s all about optimization and more is on the way.

we love benchmark but in real problem . Sometimes i would said some youtuber doing the wrong way

E.g

Dave2d show how much speed compiling this x project. He just do the compile press but not real workflow in xcode, using storyboard, switch ui and so on .

Sometimes video also i see some odd on youtube, try to add some after effect ,scrub it and put some more more. You just put how much render /scrub for few second , how this meaning-full to real user ? 2x speed compare ryzen , 1x speed compare intel xxx . Oh i don't want that..

 

[leman]

Regarding the M1 Ultra GPU performance: Apple has published a very interesting WWDC session about scaling compute workloads. They discuss in details how even subtle bottlenecks in application code can have dramatic effects on performance. I would recommend anyone interested in the topic to watch this session, it's in a format that should be accessible to any casual tech enthusiast, you don't have to be a GPU expert or even a developer. Also a good primer on how GPUs work.

Scale compute workloads across Apple GPUs - WWDC22 - Videos - Apple Developer

Discover how you can create compute workloads that scale efficiently across Apple GPUs. Learn how to saturate the GPU by improving your...

developer.apple.com

 

[joema2]

...Apple has published a very interesting WWDC session about scaling compute workloads...

Thanks for pointing that out, it is very informative. The last slide mentioned that XCode has several new performance counters related to the MMU (Memory Mapping Unit), which is the hardware block on a CPU that translates logical to physical addresses. Those were MMU Limiter, MMU Utilization Counter, and MMU TLB Miss Rate. TLB = Translation Lookaside Buffer, which is the MMU's cache for recent logical-to-physical translations.

The speaker threw that in at the last moment of his talk, with no other explanation. However it implies that besides the normal GPU-related bottlenecks, there are possible MMU-related bottlenecks the programmer must be aware of. In some of his videos Max Yuryev has speculated that MMU TLB bottlenecks may be limiting GPU scalability on the M1 Max and Ultra.

Unfortunately there was nothing in WWDC22 which addressed scalability across multiple video encode/decode units. That may imply they don't yet have the synchronization and thread pooling support at the OS and framework layers for that, much less published APIs.

When I bypass intra-app thread-level scaling by using Compressor's ability to segment transcoding on a ProRes file by splitting it across multiple worker processes, it still doesn't scale. That greatly diminishes the value of multiple encode/decode units. I hope Apple improves that by the time the Mac Pro ships, otherwise the Mac Pro will contain a lot of dark silicon.

 

[Unregistered 4U]

Don't get me wrong the M1 Ultra is a good chip, but it's not quite the "knock it out of the park" you'd expect in today's competitive computing world.

It’ll never be that and anyone that expects the “Ultra” to ever be a “knock it out of the park” kinda thing will be disappointed. However, each successive “Ultra” processor is guaranteed to be faster than the last “Ultra” processor that shipped. Anyone that needs something faster and their use case specifically does NOT require macOS or macOS applications will generally always find more performant non-Apple Silicon solutions. (This is understanding that there are some edge cases where, due to the architecture, Apple Silicon of any speed will beat Intel)