Apple M1 Ultra's 64-Core GPU Fails to Dethrone the Mighty RTX 3090
tomshardware.com
tomshardware.com
But on the GPU side NVIDIA and AMD have been battling it out for a while, competition is tight, and thinking Apple could come, compete and actually beat the fastest GPU currently available for consumers is a bit of a pipe dream, in hindsight. The GPU market hasn't been sitting on laurels for as long as Intel has been. That said, I am still positively surprised at how close they managed to reach the 3090, but alas, no cigar.
What evidence do you have for that?
Intel really wanted to release (in my opinion) very interesting processors, but had serious problems over years with getting their 10 nm process "stable" (i.e. sufficiently high and reproducible yield rates). They even back-ported some of their planned CPU architectures to their working 14 nm process "just to get something out". Since these processors were not developed with 14 nm in mind, they of course were not as good as they could have been.
Only after AMD released Ryzen, Intel had to respond at their 8th gen by cranking up core counts. And IPC did not have any increases, until 11th gen(the backported arch you mentioned). In my opinion, the performance delta between 28nm Sandy Bridge and 14nm Kaby Lake is ridiculously small.
Two of my Linux workstations are Sandy Bridge and Kaby Lake, and my real world experience bears this out. I can't distinguish between the two for everyday use cases; only synthetic benchmarks show any real advantage in the newer system. I can't speak for Windows performance differences, as my only Windows system is my gaming rig with a Ryzen 5 3600, which of course trounces both of the workstations no matter the OS.
I'm not in the market for a new gaming PC quite yet, but it's also going to be a personal workstation, so I don't want to deal with anything like above if it can be helped.
ECC -> Xeon/C
OC -> K/Z
Virtualization -> Xeon&non-K/C&H
https://www.tomshardware.com/news/intel-enables-ecc-on-12th-...
By the time I get around to actually building it, there may be some similarly-styled mini-ITX AM4 motherboards, so not all hope is lost.
Regardless of the reasons I hope they recover; multiple providers of cutting-edge fabrication technology will be essential.
You can run Windows on a Intel cpu that is 10 years old and notice hardly any difference in performance.
And on the other hand, they changed everything. Their cpus are actually innovative and really fast, and brought the entire multi core thing into consumer hands in a real way.
Intel was just doing what they have been doing for the last 40 years - building faster x86 CPU's.
They weren't even considering something as grand for desktop/laptops as Apple was with the M1 (i.e. a fully integrated SOC).
I don't think they were twiddling their thumbs to be fair - they were probably pushing hard in the same direction they have been pushing in for the last 40 years, but failed to see the industry change under their feet.
Perhaps I misinterpret your argument, but it is my impression that Intel (and also AMD!) did huge steps into that direction, just in a more incremental way than what Apple did:
- integrated GPU: check
- integrated memory controller: check
- integrated northbridge: check: https://en.wikipedia.org/w/index.php?title=Northbridge_(comp...
- developing a smartphone SoC: Intel did invest serious money into it and developed SoFIA and Broxton. Intel even strongly subsidized smartphone producers to use them. This all turned out to be a huge commercial failure and thus Intel left the smartphone SoC business.
It is also not the case that a fully-integrated SoC is "better". Rather having not everything in one SoC enabled much more flexibility for OEMs. Fully integrated SoC versus more chips is rather a trade-off between various goals.
Similarly the M1 chip came out of a smartphone SOC - it was just that Apple saw the potential for laptop/desktop adoption while Intel clearly didn't (maybe because they failed to get into the smartphone business - but their failure in that market is yet another example of 'too slow, too little, too late').
The fact that Microsoft has released several ARM laptops but selected Qualcomm to provide the processors suggests that Intel at least had an opportunity to play the game, they just never came to the table. It’s hardly Microsoft slowing them down when Microsoft is ahead of Intel on this, but their reluctance to push forwards means that they are now behind.
It’s not Microsoft’s job to push intel anyway, it’s intels job to create a product so compelling that their partners adopt it. If they want the support of major Linux distros, they just have to write it themselves rather than wait for volunteers to do their work for them.
Yes Intel has been doing some minimum improvement to CPUs each year, but the reality of "Intel was just doing what they have been doing for the last 40 years - " is....
"Milking their Monopoly"
The AMD lawsuit with Intel back in the Netburst days showed how Intel was just as bad if not worse than Microsoft at anticompetitive behavior to lock out competitors in the PC market. I'll throw Intel a bone in that for decades with this monopoly power they still continued to push the process and design envelope (probably because they were afraid of becoming Motorola, and when they still had engineering leadership left over from the earlier days).
But Intel is a badly overfed Jabba the Hut. Gelsinger has his work cut out for him.
No. Intel just wanted to show some little improvements to keep performance gap constant, hide their neat tricks until competition catch up, and use them when clients or the market really demanded it.
The only notable efforts I've seen were reducing performance penalty of SpeedStep performance switching, making better memory controllers to "Catch" AMD, and other power-gating and independent throttling capabilities to address density issues in systems.
When fab/power/thermal issues became apparent, they started to hide AVX/AVX2 frequencies, created frankenprocessors for some applications, etc.
However, I've seen no real effort to make groundbreaking innovations in x86 space rather than protecting what they already had.
Performance counters, and other underlying piping to make processor observable was nice though.
As a result, I can still use a 3rd generation i7 as a daily driver for almost all tasks at hand, including development. The only definitive performance difference shows itself when I run my scientific code after compiling it with platform platform specific optimizations on newer systems. On that regards, an M1 MacBook air can be 25% faster than a 7th gen 7700K processor, and I find it ironic.
I consider what AVX-512 has to offer to be highly innovative.
Unluckily, just when they planned to introduce AVX-512 into most desktop/laptop CPUs (not just server CPUs or special-purpose accelerators), the problems with 10 nm occured. So this was delayed a lot and even today, many desktop/laptop CPUs of Intel have no support for this feature.
Also Intel TSX was in my opinion really innovative (even though this feature was to my knowledge mostly used in (business) databases; what a pity).
AVX-512 is an example of a feature that improves special cases that show up in faux-workloads (eg: fancy benchmarks and HPC) but does not manifest higher performance for the vast majority of workloads, including things that ostensibly should be embarrassingly parallel and reap gains from SIMD.
As an HPC sysadmin and scientific software developer/researcher, I can confidently say that SIMD can provide real performance gains, however there are trade-offs and decisions to be made.
- First of all, SIMD is very data-hungry. You either need to constantly push data into it, or modify the data you've pushed a lot. Otherwise you just sit.
- Then there comes power and frequency penalty. In Intel's case, it needs humongous amount of power in CPU budget terms, and it creates heat and slowdowns. So you have to test your code with SIMD or without it (-mtune, -march, etc.). If your code is as speedy or faster, use SIMD.
- Moreover, you can't just compile an extremely optimized binary and fan it out. Older processors will just throw "illegal instruction" and halt. You either will provide multiple binaries with specific optimizations for each, or lowest common denominator for a vendor (AMD binary and Intel binary), or just throw all out. The best way is giving the source out and providing a simple makefile to let the researcher/user compile it, but not all code is open, one may guess. Creating a universal with multiple code paths is also possible, yet needs a lot of elbow grease, and may not be always optimal.
- Lastly, your code don't have to be embarrassingly parallel to be able to use SIMD. Matrix/linear algebra libraries like Eigen can almost abuse the processor's all units when compiled with correct flags (-O3, -mtune=native, -march=native). However, if you want to accelerate small data with SIMD, you need to create a parallel loop which needs to saturate SIMD pipelines. Which OpenMP can easily do with parallel_for.
All of this doesn't change that SIMD is a special horse which can't run in all courses, however its not useless.
I think you're missing the point of my post, I agree with all your points in specificity (except one, but not the forum to discuss FMV in modern compilers) but they miss the grander point that Intel hasn't made computers faster via more SIMD. The amount of expertise required to make use of it is just more evidence of that.
I feel Intel failed hard at diffusion of the ISA (why not put it everywhere, with half-perf, it'll improve later, no change in code) and also at not pushing more energy/dollars into ispc. Yeah yeah your compiler engineers are clever, but you've been doing this for 20 years and autovectorization is still ways off. Let me write code in a way that can be easily vectorized. A subset of C. Less awkward than cuda.
Now it seems AVX512 and large vector units is dying and still is too niche. Sad.
SVE is SIMD actually done right – on the Arm side in the near future, everything from smartphones to massive HPC boxes will be covered by the same clean SIMD ISA.
Mask registers offset the size problem a bit. I just wish we'd rebuild a language or clean libraries to take full advantage of this programming model. Is ispc still maintained? Does anyone use it in prod? Genuinely curious.
I feel SVE is 'too late' as most CPU makers seem to go back to smaller vector units (leaving the vectorized stuff to gpus - I know they're not the same thing, but if you're investing in heavy perf hardware, for repetitive computing...) and even Intel doesn't seem very serious about AVX512 except in the Xeon world. But then if you pay 8000EUR for a platinum thing, you might be able to pay for top talent to handcraft some intrinsics.
They likely knew what apple were up to with efficiency cores several years ago, and only decided to accelerate manufacturing of Alder lake once they realised the market was cool with that form or architecture.
What baffles me about this 'innovate on something else than peak perf' is... What did they innovate on massively instead then, if not that? Apart from AVX512?
It can indeed be a sensible business strategy, but it will much less likely lead to remarkable innovations.
Meanwhile AMD gave the market what customers clamored for: a 64-bit extension to the IA32 platform. Then AMD gave us massively performant APUs. Then AMD gave us multi-die packaging and left the IO die on a more sensible process for that function while using smaller processes where frequencies matter more.
Then Apple with some help from ARM gave us a SOC for laptop and desktop use that's frankly kind of embarrassing AMD and Intel, not so much for the core design really as its integration with memory.
Intel isn't just unlucky here. They've made a series of serious missteps going back a couple of decades now.
Intel rested on their laurels because AMD almost wiped themselves out, which led to them getting complacent with the process. The actual designs were quite good, it took AMD a few years of better process and newer designs to properly dethrone intel (which intel have now reclaimed in the mid range, this will now yo-yo as it should always do)
You mean, Intel almost illegally wiped AMD out using its monopoly position in the x86 market:
Did Bulldozer really lose or was the competition cheating?
Spectre mitigations shouldn't effect quite a lot of programs because there's nothing to mitigate within the same address space.
Worst case I see is approximately 33% performance penalty in aggregate for Xeon parts, with much worse performance in specific scenarios. Comparing the original Bulldozer benchmarks, this does close the gap like GP suggested.
Also, refresh your memory on how the mitigations work because there’s definitely an impact for most programs.
Also bulldozer is almost definitely vulnerable to spectre.
Bulldozer and it’s competition were both designed at a time where leaking information to another process on the same physical core was outside the threat profile CPU memory protection was designed to mitigate against. Intel was optimizing aggressively within the envelope of expectations at the time, which were upended by the rise of cloud computing.
Here's the terms: https://www.sec.gov/Archives/edgar/data/2488/000119312509236...
I don't think any money actively changes hands on an ongoing basis for this, but when this was signed Intel did pay $1.25 billion and AMD agreed to drop anti-trust complaints against Intel.
- building on the ARM architecture, which (1) saved them a lot of design costs and (2) being a simpler architecture already has a "built-in" performance advantage over x86.
- having years of experience already with the "A-series" of chips used in iPhones and iPads since 2010. Everyone is talking about "Apple silicon" now, but these predecessors are often forgotten.
- having a privileged partnership with TSMC, where they have access to the latest and best processes and priority over all other TSMC customers.
- This is not heir first rodeo. They've migrated from PowerPC to x86 and developed first Rosetta, UniversalBinary format and related machinery.
They've tapped into that experience, and it went way way smoother.
They had to research how to make it completely seamless that time, for the first time.
What I like about the M1 is that I get 60-70% of the performance with a chip that is much smaller and cooler than the GPUs it beats. I think for a long time we were thinking about computers as big bulky noise heat generators and Apple just shows the way that we can achieve roughly the same in a much smaller equipment that is much cooler (in terms of temperature).
This is why M1 is a breakthrough for me.
I assume you mean "enclosure" that's smaller?
M1 Ultra is 114 billion transistors, while the RTX 3090 is 28 billion. (Not sure of physical size of the chips, though - do you have a reference for that?)
Original point stands, of course, on efficiency/heat!
Anandtech has M1 Max at 432 mm2, Wikipedia has 3090 at 628 mm2.
So, since M1 Ultra is 2x M1 Max, it is physically larger than the 3090.
When Intel has been failing at discrete graphics competition with ATI/NVidia for... two decades? Yet Apple cranks out a functional competitor a year after their first desktop chip. Yes they probably have lots of experience with integrated graphics on SOC for the iPhones so this isn't totally out of the blue. But still.
What has Intel been doing all these years? Is it THAT dysfunctional? Add to the failures of VLIW/Itanium, loss of the SSD crown, underwhelming XPoint, no ARM competitors, no real position in mobile, twice getting passed up by AMD in the performance race, a company with 1/10th, possibly 1/100th the resources.
That said, am I the only one getting annoyed with shallow comparisons of the M1 products? Or is SEO BS hiding the good stuff? It seems like the rule most sites are following is 50% or more of your benchmark graphs should be Geekbench which is quite eye-roll inducing when you've been spoiled by comprehensive benchmarks in the PC market for decades.
Could "Many tech 'journalists' are Apple fanboys" be a reasonable explanation?
Some of the world's best actual journalists, like Peter Jennings, were high school dropouts. At the same time, some of the world's most prominent people who often present themselves as journalists are absolutely not.
The Medill School of Journalism at Northwestern University was one of the most highly-regarded journalism schools in the nation.
Then, its focus changed, and it is now the Medill School of Journalism, Media, and Integrated Marketing Communications with the motto "The future belongs to those who understand the art and science of marketing communications."
https://www.medill.northwestern.edu/imc/
I used to work at a place where we would get Medill students as interns. I was told that they were always high-quality. But then we started getting "journalism" majors who couldn't spell, and didn't know the first thing about grammar, fact-checking, interviewing, storytelling, or the difference between fact and opinion; but were experts at putting together videos for YouTube.
I generally wait till somewhere like Anandtech publishes benchmarks (which tends to take a while longer, given the more detailed testing).
Also they just handwave "2019 Mac Pro" for the comparison, though the 2019 Mac Pro has _5_ GPU options. Going toe to toe with the intel gpu is embarassing, going toe to toe with the W6800 is quite good. Though I suspect since they've had the device for two years it's the W5700 model, but would be good to have the specified.
Geekbench's Compute is useless for M1s with more than 32 Cores, according to Anandtech. Shadow of the Tomb Raider runs as an x86 program under Rosetta 2.
Isn't there a single GPU benchmark that actually does the same work so that comparisons can actually be made?
But benchmarks where one hardware has to do much more work than the other is obviously not fit to determine the actual performance differences between hardware; and that is exactly what the article we're commenting on is doing.
People aren't concerned about the hardware itself. They're concerned about what it can do. If circumstances mean that your powerful hardware can't be utilized, then the powerful hardware is just extra cost and no benefit.
Apple does not support OpenGL or Vulkan, only Metal, and most app devs have better things to do than rewriting code for the Mac.
The recommended way of gaming on a mac is to use emulation by emulating whatever the game is used with a metal wrapper.
IMO the claim that these benchmarks aren't fair is naive. I don't care about how good the hardware is, but about what performance I get. If I get poor performance because the software, drivers, etc. are poor, I want to know that.
Of course, I'm not at all saying that you're wrong to want that, or that these benchmarks don't show what you're interested in.
However this and many other articles are using these benchmarks to derive comparative hardware performance, which is simply wrong to do. That's what I'm criticizing.
What one can compare, is the performance of the "Apple platform" and the "PC platform" at similar price points, power budgets, features. This is more meaningful for most people, which mostly care about how fast can the computer do X (whether it uses the CPU, GPU, or some other chip, most people don't care).
QEMU simply translates calls from one hardware API to another hardware API, so it isn't an emulator either right?
But from a technological perspective, the logical way to test is to eliminate as much variable as possible and really compare the hardware compatibility. Using software that’s not optimized for the hardware is not an objective test of the hardware.
There are many ways to try to conduct as objective a test as possible. In that criteria, I found that only those review from Anandtech is up to standard. Article like this is more like click bait.
Now I’m not an expert, but if I were to compare the performance somewhat objectively, I might start with TensorFlow where Apple has releases a metal backend for it to run. Then may be write some naive kernel in Julia using CUDA and the pre release metal library (It might not be fair, but that’s where I would start given what I know.)
There's IndigoBench, but some people will complain about it using OpenCL (which still works great in spite of Apple's efforts to kill their own creation to the detriment of the whole GPU compute industry) instead of Metal.
Would be interested in benchmarks for this using the new M1 Ultra: https://www.indigorenderer.com/indigobench
Disclosure: worked on Indigo as 1/2 engine devs for 9 years.
What's interesting here is that is that we're haggling over comparing something that is an SOC design with fairly modest energy usage and cooling with a video card and cpu (presumably) that need a lot of power and cooling. Imagine Apple putting a few of these CPUs in the mac pro next year (or whenever they release that).
Of course these things are pricey. 5K for the 64 core model, ouch. I think they are going to be making lots of money selling these.
As for intel, amd, and nvidia. They each need to get their act together on the SOC front. They are not for crippled laptops anymore but for high end work stations as well. Nvidia should re-assess what they want now that their arm acquisition is off the table. IMHO they should get rid of amd/intel as the go-to cpu for their gpu and just build their own SOC and target it at gaming machines, workstations, etc. There's no good reason to stick with x86 anymore either. License ARM, or maybe hop on the Risv V bandwagon.
Octane Render runs on Metal and CUDA so can at least compare compute aspects of GPU.
First off, I don't doubt for a second that Apple has over-egged the performance of the M1 Ultra's GPU – or has, at least, selected performance indicators which aren't totally representative of common workflows. It's important that we have reviews that make it clear that this is the case and look at more useful numbers.
But this article is a terrible example of doing that. It's just a republishing of minimal benchmark results from another outlet, but without even the tiny amount of analysis that was conducted there. Even the original article gives us close to no information about the scope or setup of this testing, and proceeds to draw conclusions that are suspect at best – for instance, there's a <20% gap in 1440p "Shadow of the Tomb Raider" framerate between the Ultra and the RTX3090 – not mentioning that this is a game that's running under Rosetta – and no 4K comparison, which is the only one likely to be GPU-bound.
More importantly, the actual benchmark results are clearly and obviously wrong, with the numbers for 1080p and 4k performance being switched (unless graphics cards magically now render faster at higher resolution). The Verge has fixed the original, but how can you realistically quote these numbers and make a judgement on them without noticing this?
Just an absolutely embarrassing indictment of the state of technology journalism.
Because most consumers dont search for these articles, so they blindly believe all the fake comparisons that apple makes on their webpage
Also, we forget that these new "techs" come every year and its boring to track each of these. Companies are using this as advantage.
More like, "inexperienced". Tech aside (contexts where complexities offer controversial space in some comparisons, out of benchmarks that may not measure the actual general goal), many are convinced that the spread of false information is severely limited through social and legal devices: loss of reputation, sanctions (e.g. "false advertisement") etc.
A week later, most of the press has already forgotten about Apple's new stuff already, and noone cares about the actual benchmarks. Except us techies, who didn't believe Apple's claims in the first place.
To be honest I'm much more disappointed that they shipped yet another crappy webcam in the Studio display after claiming it's the best webcam ever. (The webcams in all the M1 Macs suck ass, despite what Apple and the early press reviews claim)
That's very debatable. Even individuals on HN (which is rare atmosphere) defended Apple's claims.
The other side of this is that people have been exclusively using Apple products for so long, that they don't have the experience to compare what Apple is doing with the rest of the industry. Apple only has to outdo themselves.
This isn't an accident, it happens due to Apple's pricing/marketing strategy. If you can convince people to pay a premium for your product (when compared to competitors), people psychologically _need_ that product to be better than the cheaper alternatives, or else they feel ripped off. Best to ignore anecdotes from customers and instead rely only on hard benchmarks/data and make purchasing decisions based on what is most suited to you.
As far as I understand it, their laptops have also been within a pretty reasonable margin of being competitively priced (at least for base models) for many years. Sure you can find cheaper alternatives, but those usually come with significant sacrifices (build quality, missing features or lower-quality versions of those features), etc. Obviously different people can vary widely in how much they value feature sets, but competitors' laptop models that line up closely in those feature sets have tended to be around the same price for awhile now.
If you list all dimensions of Apple's computing products as a radar chart and as a consumer that chart "shape" meets your needs, you will not find a competitor that meets those same needs on all of those dimensions but is also meaningfully cheaper. You can find comparable devices that are comparatively priced, and you can find cheaper devices that come with a slew of compromises, but Apple is incredibly hard to beat at their own game.
Instead of taking Apple's devices as the benchmark, and trying to find a device that's cheaper and better on all axes, you have to start with the users requirements, and see where you get from there.
For example, if your requirement is "I want a silent machine for programming with lots of RAM for running VMs" then it's very unlikely that the cheapest option is going to be a Mac.
If your requirement is "I want a machine for checking my emails and watching Netflix with a big screen" then it's also very unlikely that the cheapest option is a Mac.
The only scenario where a Mac is reasonably priced is if your requirement is "I want a machine that is exactly like one of the few machines that Apple sells and I don't mind its shortcomings".
Apple products have a very distinct and precise market. It's "good enough shit" for "I'm not willing to bother" price. That has _extreme_ value in the modern market. If you're asking anything serious of their hardware then you are, either, ignorant about what else is on offer (having only used Apple hardware for the past <many> years), or have workloads that so-happen to fall into that extremely narrow niche.
> Similar-performing hardware for years
In the past year I have <been forced to> worked with Apple hardware as a daily driver. When did you last you've use one of the alternatives as a daily driver?
I mean, fuck, imagine for a second that Windows updates took 45min, because Apple updates sure as fuck do. That's a hard pill to swallow when my NixOS updates take 0 seconds.
I have yet to see any computer ship with a half-decent camera to be fair, neither in the Windows nor in the Apple world. People want thin screens, about half as thin as your average phone. There's simply no physical space to fit good optics in there - there's a reason why external high quality webcams are so large!
The review in the verge hits this point. The verge previously bought a $16,599 model from apple, which performs worse than the current $4,000 model.
https://www.theverge.com/22981815/apple-mac-studio-m1-ultra-...
They're not that fake to begin with, at least not in "experienced performance" aka how performant the system feels for the user. Apple's stuff is expensive but in terms of performance even the Intel-based Mac lineup just blows the direct competition away:
- Apple's solid-aluminium body provides vastly better thermals than the Windows world which is usually plastic crap, which allows them to run the same Intel processor on boost mode for a longer time, and the fans are at least in my experience a lot better (especially in loudness and "annoyance" factor - even at full power, the fans of a MacBook are way better to listen to than those of your average Windows laptop). For the Mac Pro / iMac lineup, same thing. Most of the Windows world ships with barely adequate coolers and no thinking of airflow, whereas Apple... well, open any tower Mac Pro and see for yourself. That's what a cooler is called...
- Apple's OS has way less crap and clutter going on that gobbles performance (there's just at this moment a submission on the front page that details among other things how Windows got transmuted into a horrible mess of advertising and half a dozen competing frameworks), and they have a lot more control over the hardware and driver stack which means that Apple does not depend on the shoddy work that vendors for the Windows world can pass as "quality tested drivers". The result is better battery performance (because as Apple can control the entire stack they can make full use of sleep modes and other energy saving measures) and less bugs.
- related, but not by something that Apple has done: There isn't much malware going on on macOS and never has been when compared to the Windows world - you can get by without a virus scanner just fine. In contrast, most of the Windows world ships some crap "free for three months" virus scanner bundled which just destroys performance.
- Apple's own chipset designs are, while still based on standards such as the ARM ISA, completely fine-tuned to their needs - which means that, unlike the Windows world which carries about three decades worth of old garbage in their CPUs, Apple can devote more of the same chip area to the needs of modern computing.
- Here's a bit of a controversial take: Unlike the Windows world where most vendors ship systems with highly modular components (i.e. CPU, RAM and storage can be taken and upgraded), Apple solders the chips to the mainboards, very near together. That allows not just for higher performance because the latencies are lower but also because the chips can be cooled together and so run at a higher frequency.
The results are nuts.
https://devtalk.blender.org/t/cycles-apple-metal-device-feed...
I’ve already seen a significant CPU improvement on my 2019 16” and I can now use my 5500M to render. My PC’s RTX 2080 Super is significantly faster but it draws more than 200W.
This indeed is great progress!
Nobody bothers to guess what's the potential of some new graphic card once all drivers, games, OS etc are fully tweaked and optimized. You decide on performance now.
We knew this would happen for everything using Rosetta, that's just part of transitioning to a new CPU architecture. For Vulkan vs Metal that's self-inflicted on Apple's part.
Well it worked for Microsoft
1: https://docs.microsoft.com/en-us/windows/win32/opengl/portin...
These days you're better off using OpenGL via ANGLE, built on top of Direct3D. That will probably also happen analogously on Apple's platform with Metal, so I don't quite get the kvetching about Apple deprecating OpenGL.
As if it matters whether Apple ships an OpenGL implementation as part of the OS or purely as a normal user space library.
If Apple wants people to take their claims seriously and if weak engine support for Metal and Rosetta support is a problem they should pay someone to port a popular graphics intensively game properly. Of course it works in their favor not to because fans can point to it and say we'll it's just not optimized for it! Like we all did back in the PPC days until they admitted how crap those chips were when switching to Intel.
34s on ultra (64 core gpu) vs 11s on rtx 3090
So 3 times slower but taking ~4 times less power (350W 3090 70W ultra [on 48 core gpu so about 90 on 64 cores])
But i can agree results are good but we need also to remember Samsung 8nm [rtx 3090 node] is 4 years old so 2 generations different from TSMC 5nm 4090 will be probably on 5nm node in few months and presumably will have 3x teraflops (probably 3x tf32 precision but it could be fp32 using 2/1.5x(?) more power).
(also I'm not sure about optimization it was write by apple employees and apple likes to drop open source support and focus on proprietary software)
But it is impressive that they get high GPU performance with LPDDR. Consoles also use one big SoC and unified memory – but with GDDR memory!
Really, if you can fit the memory on top of the GPU, then there is no meaningful distinction between a discrete and integrated GPU because the fundamental difference on the PC platform is whether the iGPU uses the system DDR RAM or whether it uses dedicated GDDR RAM. That distinction really doesn't exist on Apple's M1 platform because the memory is placed directly on top of/next to the SoC.
It's just due to apple being the biggest customer they have access 1 year earlier.
And who's fault is that?
Of course, this is marketing material. But as far as I am concerned, the jury is still out, I'd like to see a benchmark with native binaries (non-Rosetta) and some optimization for Metal.
Though even if the performance was half of the RTX 3090, that would be mighty impressive. I have an RTX 3090 on my desk and under full load it is loud and consumes a lot of power. Mac Studio + M1 Ultra's fans are apparently barely audible.
Bring on the marketing and PR battles!
It's either this or no comparison at all. And it's a game that apple often uses in their promotional materials.
Because it's not being marketed as a gaming computer, and other use-cases aren't run under Rosetta (e.g. 3D rendering, video editing etc).
e.g. Apple's marketing doesn't claim gaming performance, it claims this:
> You can run complex particle simulations or work with massive 3D environments that were previously impossible to render. And with twice the media engine resources, M1 Ultra can support up to 18 streams of 8K ProRes 422 video playback — something no other personal computer can do.
Their marketing page doesn't even mention games...
With M1, I think they can move beyond these shady marketing tactics entirely and focus on what they're developing. In other words, promote the merits of their architecture and systems (which are excellent) rather than try to compare Apples with oranges.
I mean, they turned these chips into desktops, because that's how good they are. Wtf are you talking about?
I agree. M1 with Metal is already quite good. Just leave it at that. You're not going to be able to compete with a high-end discrete GPU in most areas, so stop pretending.
(more barriers to be able to run with existing GPU API semantics)
edit: so i get downvoted for commenting that a benchmark needs to control external variables (like emulation vs native) in order to get a valid comparison of the underlining hardware? Never change apple fanboys, never change!
Until either Asahi/Windows support for M1 gets further along, there are more M1 Ultra equipped machines, or more workloads get ported to Apple native (either because there's incentive enough despite the difficulties Apple introduces, or because Apple reduces those difficulties), it's hard to have an apples-to-apples comparison of the M1 Ultra vs 3090.
I fall more to the camp that we have to evaluate the package of the Mac Studio as presented for now. Especially as it's not like Apple is going to sell this standalone, so the only other deployment in the near future is a likely M1 Ultra Mac Pro. This means it can have the advantages (lower power draw, OS specifically optimised for it) and disadvantages (emulation for common workloads) that come with that.
Games are always less optimized for Mac/Metal than for Windows. It would have been better to use a content creation benchmark.
Why and how would these GPUs deliver more fps with higher resolution. 4K is 2160p or 2x1080. Shouldn't 1440p (2x720) deliver more fps? Someone please tell me what is going on here.
M1 performance-per-watt ruins any GPU by Nvidia on any such benchmark.
Older GPUs with larger node sizes would fare even worse.
Why does the FPS go down with lower resolution?
https://www.theverge.com/22981815/apple-mac-studio-m1-ultra-...
Such real-world benchmarks are thin on the ground for the M1 Ultra thus far and are clouded by things like programs and drivers not being optimized and/or translation layers. This will improve over time.
But I've seen a couple of game benchmarks for the M1 Ultra that seem to show it's >50% of the FPS of the 3090. That's honestly astounding.
I'm still confused by the power draw of the M1 Ultra. It's claimed to be 60W but the Mac Studio seems to draw >300W so what's the reality?
But if a CPU/GPU combo does in fact draw 60W (compared to the 350W+ for the 3090 alone and another 100W+ for any likely CPU) and gets over 50% of the performance? That is absolutely huge (if true). This is still a first-generation product. Many expect the M2 later this year. That may bring respectable (but definitely not cutting edge) GPU performance to the Mac Mini. Apple still has a lot of headroom here.
I think that’s because it has to power all those ports, too. Google tells me Thunderbolt 4 can go up to 100W.
Despite this m1 ultra gpu is a big step for all people intrested in apple market and DL/ML workloads
While the M1 Ultra is the latest CPU, it still uses A14/M1 cores from 2020. It's likely that Apple made a lot of progress on the microarchitecture since then, plus rumors are that M2 will switch to a 4nm node.
Apple is on TSMC 5nm here comment from 2years ago about 8nm samsung vs 7nm TSMC
"Hell, Samsung 8nm is really a 10nm process, it's just an extension of Samsung 10nm with a ~15-20% improvement in transistor density.
TSMC 7nm (EUV 4 layer) is nearly twice as dense as Samsung 8nm. non-EUV (first gen TSMC 7nm) is still far denser than Samsung 8nm."
Well as a mostly Apple user, I don't read Apple propaganda. Or nvidia propaganda for that matter.
The M1 Max mac studio's full system consumption on load seems to be ~50 W. I wonder how much performance nvidia can deliver with that. I think ... they simply don't have a product in that power budget?
I know most people only care about FASTEST CONSUMER GPU (yay, it doubles as a leaf blower and a space heater) but I still think nvidia's best product was the 1050Ti ... decent 1080 performance in 75 W.
It's out of stock just about everywhere and the few places that you can get it have scalpers charging thousands for it.
A standalone GPU that consumes multiple times more power than the M1, is a massive brick of hardware and is impossible to get for most consumers performs better? Shocking.
4K: 60fps
1440p: 96fps
1080p: 108fps
The Verge doesn't mention what graphics setting SoTR is set at, but assuming it is Ultra, the M1 Ultra's performance is about on par with an RTX 3060 Ti (MSRP $400):
4K: 54fps
1440p: 96fps
1080p: 130fps
Here's what a 3070 Ti (MSRP $600) gets:
4K: 67fps
1440p: 111fps
1080p: 162fps
Here's what a 3080 Ti (MSRP $1200) gets:
4K: 83fps
1440p: 147fps
1080p: 210fps
GPU performance in games doesn't scale linearly. You pay a lot more for a small boost in fps. Sourced from gpucheck.com which loosely matches up with the reviews I checked.
Now, Apple’s benchmark, like all benchmark releases by their own manufacturer including Intel, use tests that’s favorable to their specific hardwares. No 2 workloads are the same, there probably exists some workload that your hardware run fastest comparing to the competition.
In this case, relying on GPU alone doesn’t seem to make the M1 Ultra comes first, but they have a secret weapon: unified memory. Anyone who uses GPGPU for example will know that the latency to copy back and forth between main and GPU memory can be a deal breaker in some workload.
So to choose a a benchmark making M1 Ultra stands out then would involve picking some intensive workload that involve moving data between the GPU and CPU. That movement would diminish a bit of the gain from 3090 and if choosing the mix right, M1 Ultra can come out on top.
I’m not accusing Apple to be disingenuous, at least not in the sense that is not a common practice in the industry.
Apple’s chart is very uncommon though. Eg at a few points they keep showing the same statistics comparing it to another thing, where consecutive charts are just changing the “other thing”. Any sane person would put all of them in the same chart. The reason? Probably because nothing is in scale so drawing all of them in the same chart is meaningless.
People using such applications where additionally power efficiency is paramount, such as memory hard proof-of-work, should be very keen in seeing whether the M1 Ultra lives up to its promise...
Apple, with such capabilities you need to invest in gaming/APIs/consoles.
Sadly you can't take advantages of Apple's CPU designs unless you buy into their ecosystem. Here's hoping ASAHI linux (Marcan's port to the M1) goes well.
There's little things like ditching the 32 bit ISA, not a big deal, but does make each new iteration that much easier from a design and testing standpoint.
ARM also has a loser memory model, which makes it easier to use more effectively use existing bandwidth.
But there's no magic. It's a straight up aggressive design, wide instruction decode (yes that's easier for arm than x86-64), large re-order buffer, wide issue, etc. Anandtech has a nice summary comparing various M1 features to the Intel and AMD competition. Apple has a high IPC and they make the most of their lower clocks (usually around 3.2 GHz). The chips from Intel/AMD that they are compared against are often in the 4-5 GHz range and burn significantly more power because of that.
Given how much time I spend in the browser/Electron apps, browser benchmarks are the only benchmarks that are relevant to me (tongue placed firmly in cheek, I'm sure the entry-level M1 would suit me just fine).
I’m not hopeful to be honest, at least not without a significant thermal reduction penalty. The M1 Ultra in the Mac Studio has a 1kg heavier heat sink vs the Max due to the extra heat.
(the M1 variants are awesome, but as per the article, their marketing... stretches the truth)
(nvidia had the same issue with miners still preferring their LHR consumer card to their attempts to push miners to their business tier cards)
I'm not saying Apple gimps their hardware. I'm asking because you are saying Apple gimps their hardware by not releasing "workstation drivers". Thus I ask you why would apple Gimp their hardware.
What about power usage? ;)
At equal power usage, M1 decimates the RTX 3090
¯\_(ツ)_/¯