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.
You're Apple. You just built a 40-core SoC for Mac Pro. Now what?
- Thread starter senttoschool
- Start date
- Sort by reaction score
Um, Apple isn't the only company looking to move beyond x86. Microsoft is dipping a toe in the water. Amazon offers ARM instances in AWS. You can get Chromebooks with ARM processors as well. When it comes to performance for desktop and mobile devices, Apple has a pretty strong lead. They're now bringing this performance to the Mac. They can leverage success in iOS because they are vertically integrated. Microsoft doesn't have this advantage. Google is now also moving toward custom silicon. Hell, even AMD is partnering with Samsung to bring their GPUs to ARM processors.
If desktop/laptop gaming is your primary use case, then yes, x64 and Windows is currently your best option. That will likely be the case for some time. These sorts of transitions take time. We're not even a full year with M1.
If desktop/laptop gaming is your primary use case, then yes, x64 and Windows is currently your best option. That will likely be the case for some time. These sorts of transitions take time. We're not even a full year with M1.
1. In your prior post you asked if those games ran on the Mac, not if they were M1 native. This is an example of you shifting the goalposts every time your position is proven wrong. You also shifted the goal posts by now declaring the games to be "old", even though titles such as Cyberpunk 2077 are on the list as well. Also, you're the one complaining about iOS apps on MacOS, not me. Not sure why you felt the need to project your feelings on me, but I guess that was the best argument (in your mind) you had.
2. TikTok is just one app. That does not change anything regarding my points regarding issues with using mobile apps on a desktop OS, regardless of whether they were written for iOS or Android. That was also a very weak attempt at an argument.
3. We have an established track record from Apple of supporting older architecture throughout a lineup change and for several years after. While I only used the PPC-Intel switch and the original Rosetta as a guideline of what to expect, there was a similar timeframe in place (without a tool such as Rosetta) when Apple switched from 68k to PPC, so there is an established track record, with nothing from Apple indicating a dissimilar timeframe for this transition.
4. By "no reason to explain anything in detail", I take that to mean you have nothing of substance to add to your arguments because there is no evidence to support your claims. I brought numerous examples to disprove your claims (none of which are based on any evidence, just personal bias and conjecture) which were supported by evidence, yet you keep replying with little short blurbs then basically say "I don't have to provide proof if I don't want to". I'm done with this conversation because you've been proven wrong time and time again by multiple people in this thread, yet still continue to repeat the same unfounded claims and opinions as irrefutable fact.
Last edited:
As someone who has worked on a (large) project that runs on both iOS and Mac, this is pretty much spot on. Our graphics code was nicely shared between the two platforms. Much of the lower level stack was the same too. Where things got a bit weird was having to deal with memory pressure (especially early on with the iPad 2 and 3), and input devices, along with the UIKit vs AppKit disparity. Or features that we couldn’t enable on iOS due to App Store rules/etc.
What if they kicked it off low-key, using the existing MacCoLo sites (and investing in that infrastructure?) before setting up their own facilities...?
I know that at least one company has started supplementing their existing Intel Mac Mini server farm with an M1-based server farm. It's great for companies who make such a switch since other than swapping out hardware, everything else is the same. I think you won't see a big shift from Intel to Apple Silicon until the 2nd or even 3rd generation SOCs hit the market though.
The fact that there are companies out there creating these Mac server farms should tell Apple that there is already an immediate market for Apple Silicon Cloud. Heck, Apple even sells a Mac Pro in a server chasis.
I think it's only a matter of time before Apple creates Apple Silicon Cloud.
To be fair, I would say the rackmount Mac Pro is aimed more at audio & video folks who keep them in remote machine rooms, where the noise & heat can be handled; and especially for the audio folks, it also keeps the noise out of the room(s) with the sensitive microphones.
Mac Stadium has been doing this for years already. If Apple does do something, it won’t replace having a powerful Mac at home/work but be a supplement like Mac stadium.
Having Apple Silicon Cloud solves the problem of having to build your own server room. You can just have a cheap Mac Mini connected to a big monitor and still have the power of a 40 core SoC.
This is actually one of the main reasons why cloud computing took off. It’s almost always way better to rent a big server from the cloud than to build your own. The cost savings, maintenance cost, and productivity are just much better with the cloud. And you don’t have to upgrade your $20k Mac Pro just to get the latest SoC.
Not even Apple can escape this trend. This trend does not decrease revenue or profit for Apple. It will simply increase the market size for powerful Apple Silicon computers since it’s easier and more affordable. IE, perhaps big studios will switch from cloud Xeons to Apple Silicon Cloud.
The market for renting a cloud computer for personal productivity/gaming is growing fast. Cloud computing used to be mostly for hosting your app backend or doing expensive calculations. Now, consumer and professional workflows are increasingly moving to the cloud as well. For example, GitHub Spaces creates a dev environment for you in a few seconds. A new Y Combinator company is putting Chrome in the cloud to speed up web applications for professionals. More and more applications with a UI are getting accelerated in the cloud.
This doesn’t make prosumer computers like the MacBook Pro less important. They still need to be powerful since many applications that users use are latency sensitive.
Last edited:
We are not terribly far enough from professionals renting their personal computers from the cloud being as common as leasing iPhones.
If gaming can move to the cloud, anything can.
Been using GeForceNow from my iMac a lot more lately than my gaming PC. Despite very expensive rig, I still get better FPS in games via GeForceNow, and the room doesn't heat up by 10° either..
Hence, it’s not an “if” but “when” Apple will offer Apple Silicon in the cloud.
When Google/Microsoft/Amazon can start clearly provide more via the cloud, Apple will need to as well.
Hell, if you have signed up for the Steam Beta, they support a feature called Steam Link. For people that have both a Windows and Mac, you can play any game installed on either computer via Steam Link. The host computer does the heavy lifting, but it really opens up what you can play on the M1 Mac (including games that are Windows-only or never were updated to 64-bit on the Mac side of things).
Meet Windows 365 | Windows 365
Deploy and manage Cloud PCs with Windows 365, a Microsoft virtualization solution designed for hybrid work. Find the right Windows 365 plan for your employees.
www.microsoft.com
Microsoft is doing this already.
There is simply no escaping this trend.
I think Apple will follow within three years or so. But unlike Microsoft which puts Windows on a browser, Apple will likely do a deep integration with iPad/Macs so that you won't really notice if you're using a local machine or using a cloud instance.
One day, we can all chill at a coffee shop with our M5 Macbook Air and be able to access the power of an Apple Silicon with 256 CPU cores and 512 GPU cores.
It’s possible. But let’s also not forget that Apples and MS businesses are very different. MS sells services, Apple sells product experience.
Lion share of Microsoft's revenue comes from office and corporate/business users (over 70%). They want you to use MS Office and Active Directory, and offering Windows in the cloud is a smart way of making sure you stay in their revenue-generating ecosystem no matter which device you prefer. In addition, office/business work is well suited to remote computing, since display updates are rare and the latency of data transmission over internet can be tolerated.
Appel however wants you to use their physical devices. They care less about you running their OS itself, they want you to be part of the experience they set up by matching hardware and software features, as that is where their profits come from. Services like iCloud, the new Xcode cloud make perfect sense, since they add value to the seamless "Apple magic" and allow you to use your device more comfortably. But the experience is still focused on you and the device in front of you. I am not sure how macOS in the cloud fits in here as it won't be able to deliver that low-latency user experience Apple wants to be known for. Take video editing for example. Current M1 Macs are excellent video editing machines, because their memory architecture allows them to say super responsive while working with high-res content that brings nominally faster machines to their knees. You just won't get the same experience from the cloud computing: bandwidth and latency are simply not where you want them to be.
I can imagine some sort of "compute server" in the cloud, where you app runs locally on your Mac but will offload the data to the Apple server to have it processed there, but again I am not sure how useful that would be. Data transmission would probably take longer than the time you'd save from using a more powerful cloud machine, so that leaves you with only a few very niche workloads.
Anyway, that's just my thoughts. I don't claim to be any kind of authority on these matters.
That is aimed at corporations, not individual and home users. The documentation and press kits regarding Windows 365 spell that out in detail and with a level of clarity somewhat unusual for the team in Redmond.
Apple is going to have to be very careful if they do this. They're already publicly against cloud gaming and have in fact made it difficult for other providers to do so on the iPhone, so if they suddenly reverse course at the same time they release a cloud gaming service, there's president for an anti-trust complaint.
and when you hit your cell plan slow down point you will notice or you may get an $10,000 roaming bill in some cases.
Every major cloud services vendor has a some ARM effort going. The ARM neoverse platform means anyone can license and jump into the game.
Cloud Computing and Network Infrastructure
The rapid advancement of AI capabilities has accelerated an unprecedented need to deliver maximum compute performance as efficiently as possible, not just in the datacenter, but the entire compute topography from cloud to device. Custom built-for-AI silicon, built on the Arm Neoverse platform...
www.arm.com
N1 could easily hit 64 cores and the follow ones go up from there. Apple coming with 40 cores doesn't automatically make them "big time" player here. Apple has chucked PCI-e aggregate bandwidth to chase single threaded performance using low power RAM ( far more lanes allocated to memory than for external I/O). That makes sense for a single user workstation. That doesn't make much sense at all for a multitenant , cloud services server.
There is zero "first mover ' advantage here for Apple. The M-series being native boot only macOS makes it even less so. ( Linux support dependent upon the jailbreak hack to get going isn't something that critical 24/7/365 business critical workloads are generally going to bet the farm on. It is an amusing hack that works for some. It isn't primary time though. If Apple closed the jailbreak vector in 3rd or 4th generation update that is a dead end. )
The big cloud services folks can just go to Ampere Computing and just buy an Altra series solution. Ampere is getting into semi-custom designs. So not really like folks like Microsoft have to do it all in-house if they don't want to. ( one reason Ampere is adapting is because some many folks are doing "roll your own". )
It isn't just vertical integration. It is the scale that the iOS and iPadOS devices bring. Apple reusing the same baseline microarchitecture core design on both Mac and iOS/iPadOS means that can share the R&D costs over a larger set of devices. The uncore parts of the M-series and A-series are different, but ther are huge overlaps also.
That was highlighted when the T2 took over several jobs in Macs even when still on the x86_64. Imagine processing subsystem. Audio. TouchID. The SSD controller.
Which points to why a server processor that is increasingly detached from the iPad and laptop focus of Apple is increasingly a problem. At its core, Apple doesn't like chasing after low volume niches. Apple shifted from having a narrow XServe product to replacing with with repurposed Mini and Mac Pro and actually grew their macOS Servers sales. They more mainstreamed macOS server and in the big picture it worked better. For the narrow server only focused folks it did not.
The vast bulk of Apple's cloud services run on top of Linux not macOS. It would be pretty straightforward for Apple just to buy some Ampere computing power boards and replace the x86_64 non-Macs that the service largely runs on now.
Apple's apparent focus on iGPU only is going to detach them from a large chunk of the gamer focused crowd. The permanently attached GPU will run into a buzzsaw with the bulk of the Windows Desktop gamer crowd.
It isn't just the M1 hardware. macOS on M-series has zero 3rd party GPU drivers ( which pragmatically means zero focus or interest in discrete GPUs. ) Much of the emphasis at WWDC 2020 and 2021 was on unified memory semantics for M-series graphics. ( higher synergies with native iOS graphics stacks ). On the very similar shared inertia track with iOS/iPadOS as the microarchitectural cores are. If WWDC 2022 goes by and still deep radio silience on dGPUs there is a big gap there.
Gaming on iOS/iPadOS/tvOS is a profitable business for Apple so not disconnected from all gamers , but tracking the same things the hardcore desktop Windows ones tracking is not the case.
If their billions in server costs is to run Linux instances their own silicon doesn't buy much of anything significant. Their own silicon doesn't natively support booting Linux ( folks have done it through a jailbreak hack but that is quite illustrative of Apple's intended interest for deployment. If they had intending to go wide on the silicon they wouldn't have closed the ability at lower levels. )
What Apple is largely paying for is "scale out" costs not particularly "scale up". Apple's silicon isn't going to help them necessarily "scale out" any better than the folks they are buying services from. It is probably not a "horsepower" issue. It is a probalby more so deployed at multiple physical locations issue ( number of machines, switches , and disks )
For example, caching video for AppleTV+ isn't gong to run any better on Apple Silcon at all. They need to move the data closer to the consumers to be more effective. That's more duplication. The bigger problem on those costs is really too many folk on super extended long free trials than some "magic button" that Apple Silicon is going to invoke.
The more that Apple gets bigger in the "we store large data in the cloud for you" business the closer they will get to the point where outsourcing pragmatically the whole job doesn't make sense.
" ... Amazon Web Services-designed Graviton2 processors are the “gift that keeps on giving” with “big savings over Intel,” according to Don MacAskill, CEO at Flickr and SmugMug, the online photo management and sharing platforms. .."
AWS Graviton2 Chips A ‘Big Savings Over Intel’: Flickr CEO | CRN
AWS’ new Amazon Elastic Cloud Compute (EC2) X2gd instances are powered by its Graviton2 processors and deliver up to 55 percent better price/performance for memory-intensive workloads than current-generation x86-based X1 instances
[ SmugMug / Flickr have lots of storage usage on AWS storage but they also aren't giving away long free trails either. If have $1B in AWS fees and $1.4B in income then that is OK. ]
If Apple was really having a x86_64 compute cost problem with Amazon then shifting over to Graviton 2 would have cut that 20-40% with putting the Apple Silicon design team off on some tangent at all.
Apple already has cloud services APIs and they are running on top of Linux. Apple Silicon SoC aren't going to break much substantive of anything better than buying a rack rull of Ampere computing nodes would in terms of time to market effectiveness , energy cost effectiveness , or R&D cost return on investment.
Running some macOS specific services like XCode Cloud then sure. Racks of Mini provisioning a cloud is already a proven business model. There is little new there that Apple would bring other than shifting revenues from "partners' into their own Scrooge McDuck money pit.
Apple has probably been running their own XCode Cloud internally anyway. Each WWDC beta period they probably get a spike in doing about "twice as many" CI/CB test runs as have two pairs of macOS, iOS , iPadOS , tvOS , etc in very active flight at the same time. So pretty good chance would need to deploy more nodes into that internal "cloud" to handle that workload. But what do those 'extra' peaker load systems do the other about 3/4 of the year? With this external XCode cloud service they can rent those "extra" nodes off to other folks and have them soak up the acquisition costs. If Apple can rent off enough new ones of those per year then their build cloud could get hardware refreshes every year at even lower costs then it incurs now ( if folks rent them long enough then pretty close to completely free is transfer all of the hardware costs to the external renters minus some internal triage "finance cost" to delay the start of payments. )
The ungrounded assumption is that some server class SoC would have built anyway. Deploying already developed Minis and Mac Pro as essentially single user systems as cloud services is cheap because Apple was going to make single user minis and Mac Pros anyway for non cloud deployment. In fact, that would be the vast majority of the deployments. They would be just be piggybacking on single user desktop deployments similar to how they piggybacked on Mini's and Mac Pro after the discontinued XServe. The "savings" was primarily in not doing XServe in the first place.
As far as I know, the M1's memory bus do not use PCIe at all. That would be disastrous for bandwidth performance. I do not think any CPU's memory busses go thru PCIe. CPUs have direct access to their DRAMs via dedicated memory bus lanes.
There is no “jailbreak” hack needed to boot Linux on an M1 Mac. The M1 boot loader is designed to load alternate kernels. See the kmutil man page
Also some details on how Corellium booted into Ubuntu here
Last edited: