Yup.
But you may get thermal throttling with a laptop too and I'd rather get numbers of what the capabilities of the system are.
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.
[CPU only] Apple M1/2(Max/Ultra) (TSMC 5nm) vs AMD Zen 4 (TSMC 5nm) - Technical Analysis
- Thread starter exoticSpice
- Start date
- Sort by reaction score
Yup.
But you may get thermal throttling with a laptop too and I'd rather get numbers of what the capabilities of the system are.
Comparing M3 to anything right now is a bad comparison...
AMD 7000 was built on a known processor architecture, the m3 was built on a new architecture with terrible results both in yields and in quality. When TSMC switches over to N3E and apple makes a chip on that node, it will smoke anything AMD has on their roadmap for years to come.
N3B which all M3 processors are based upon is pure crap, infact apple is the only retailer who accepted any N3B products every other company turn down N3B because it was pure crap.
AMD 7000 was built on a known processor architecture, the m3 was built on a new architecture with terrible results both in yields and in quality. When TSMC switches over to N3E and apple makes a chip on that node, it will smoke anything AMD has on their roadmap for years to come.
N3B which all M3 processors are based upon is pure crap, infact apple is the only retailer who accepted any N3B products every other company turn down N3B because it was pure crap.
As a resident Apple Silicon fanboy;
It’s perfectly valid to compare two current competing products with each other. What you’ve suggested is like what another user suggested but in reverse (that comparing Intel v. Apple Silicon isn’t fair because of process node).
We should hold every product to the same standards.
Seriously dude?
OK, let me be very clear. When I say noise I mean there is nothing TECHNICALLY interesting in such small differences.
If you find such differences fascinating for whatever reason, go for it.
But don't be surprised when other people are simply UNINTERESTED in your trumpeting such numbers. They'd not interesting for clarifying tech differences between designs. They're not interesting for deciding to buy a new machine.
HotChips just uploaded the presentations from the last conference.
AMD made two presentations that may be interesting for this thread.
- AMD Next Generation “Zen 4” Core and 4th Gen AMD EPYCTM 9004 Server CPU
- AMD Ryzen 7040 Series Mobile Processor
AMD made two presentations that may be interesting for this thread.
- AMD Next Generation “Zen 4” Core and 4th Gen AMD EPYCTM 9004 Server CPU
- AMD Ryzen 7040 Series Mobile Processor
Can anyone confirm and explain this comment?
Can anyone confirm and explain this comment?
For fastest performance, you want your data to be in cache. Cache usually works in small blocks (e.g. 64 bytes). This means that if you are processing a sequential list of data elements, only the first of the 64-byte block will actually result in a DRAM access — entire cache block is loaded and subsequent n elements can be loaded from fast cache. Even better though, if you know that you will be processing a bunch of such elements, it makes sense for the CPU to load the relevant blocks into the cache even before you get to processing the relevant item, this reduces waiting times. A while ago we had dedicated prefetch hints for this (e.g. before doing a large memory read you could instruct the CPU that you will do it, prompting it to start loading the data from DRAM into the cache). Nowadays this is done with automatic prefetches that try to learn your access pattern and prefetch data accordingly. Detecting linear access is simple (e.g. if the CPU sees that you have accessed N subsequent locations in memory it can assume you are doing array processing and start fetching ahead). Apple goes one step further and also prefetches indirect accesses — that is, if you are processing an array of pointers — it will detect it and start loading the data at the subsequent pointer addresses.
There is in fact still an instruction for explicit memory prefetch, because maybe it is sometimes still needed. I am not seeing any flush/invalidate instructions (though with all the layers of caching, that would be somewhat fraught); perhaps those are effected through MSR?
The DC instruction has modifiers that will perform a wide range of cache maintenance operations.
Learn the Architecture - A-profile
Free how-to guides and tutorials on the Arm A-profile CPU architecture, including Armv8-A and Armv9-A.
developer.arm.com
Prefetching on Apple chips is in fact extremely sophisticated. I'd be surprised to see a real-world example of a data pattern that's both predictable enough for SW prefetch to be worthwhile, but isn't caught by one of the many Apple hardware prefetchers.
Well, look at DCZVA: the program tells the processor, I am going to fill this whole line with stuff, so zero it out and don't bother to load it. That is just excellent.
If you are calling DC ZVA a *prefetch* instruction then I'm out of this conversation.
You're obviously more interested in "winning" debate games by playing stupid word tricks than in understanding technology.
No, it is not a prefetch, it is a do-not-fetch, because the program only wants to write. It saves the fetch cycle that would normally happen when a program starts writing stuff. Of course, AS might well have that in their memory optimization logic, so that a program would not need to issue the instruction at all. In fact, I would not at all be surprised if the other designs, including x86, have it as well.
Everyone in the PC world wants to compare themselves to the MBA. This is the comparison between Zen 5 and M3.
www.tomshardware.com
AMD deep-dives Zen 5 architecture — Ryzen 9000 and AI 300 benchmarks, RDNA 3.5 GPU, XDNA 2, and more
Zen 5's 16% IPC improvement floats all boats.
Everyone in the PC world wants to compare themselves to the MBA. This is the comparison between Zen 5 and M3.
View attachment 2397230
AMD deep-dives Zen 5 architecture — Ryzen 9000 and AI 300 benchmarks, RDNA 3.5 GPU, XDNA 2, and more
Zen 5's 16% IPC improvement floats all boats.
Pretty cool though I'd love to see a comparison with the M3 MacBook Air in a fanless thin and light comparison. I read through the whole thing and didn't see any mention of battery life either.
It looks like third-party reviewers are embargoed from providing any interesting benchmarks right now.
Nothing about battery life on the ASUS website either. All they say is "Zenbook S 16 has the day-long stamina you need, and more." Whatever that means.
Everyone in the PC world wants to compare themselves to the MBA. This is the comparison between Zen 5 and M3.
View attachment 2397230
AMD deep-dives Zen 5 architecture — Ryzen 9000 and AI 300 benchmarks, RDNA 3.5 GPU, XDNA 2, and more
Zen 5's 16% IPC improvement floats all boats.
Shadow of the Tomb Raider? They couldn't find a newer native game? 😆
Last edited:
By the way, Cheese interviewed the person in charge of Zen development during the AMD event.
chipsandcheese.com
A Video Interview with Mike Clark, Chief Architect of Zen at AMD
Today’s “article” is a little bit different to what you readers are used to.
chipsandcheese.com
Not an excuse for using a non-native app. Benchmarks are not the only way to test game performance. How do you think PC sites test the performance of Resident Evil games, Lies of P or or Baldur's Gate 3?
One difficulty PC oriented review sites have had is lack of access to (or at least knowledge of) the tools needed to make such measurements on the Mac platform. They know how to do it under Windows, they don't know on Mac. So when they decide to dip their toes into Mac benchmarking, they naturally gravitate to games that have built-in benchmarking tools, even if they're not native.
We’re not talking about PC review sites wanting to benchmark Mac games. Here we’re talking about AMD which is very familiar with Macs since they were Apple’s GPU provider before. It is also common knowledge how you can activate Metal Performance HUD in games. A Google search shows you easily. You can also simply use Steam fps counter that now works in Sonoma too. The graphs are also only showing a comparison by percentage, no Min/Max/Avg fps so no fancy special tools needed for benchmarking Mac games, just Metal Performance HUD.
Yeah, more benchmarking BS. They're comparing a 12-core processor (AMD Ryzen AI 9 HX 370) to an 8-core (the M3 in the Air), using MT CPU workloads. E.g., for GB6, the AMD is 2,879/14,888, while the 8-core M3 is 3,082/12,087 (all scores taken from tomshardware.com for consistency), so of course they're going to report the MC scores, and conveniently omit the difference in SC performance.Everyone in the PC world wants to compare themselves to the MBA. This is the comparison between Zen 5 and M3.
View attachment 2397230
AMD deep-dives Zen 5 architecture — Ryzen 9000 and AI 300 benchmarks, RDNA 3.5 GPU, XDNA 2, and more
Zen 5's 16% IPC improvement floats all boats.
And what kind of apps are people most commonly using on thin&lights? That's, right SC.
Doesn't the tech press use the same video games to make historical comparisons?
They are not comparing SoCs, they are comparing their laptops to the market leading ultrathin laptop. It's not their fault that the market leading ultrathin laptop only has 8 cores. Price should determine whether the comparison is fair or not.
You missed my point. The chart isn't BS because they compared an 8-core and 12-core processor. It's BS because they showed the kinds of tasks on which the AMD is faster (MC), but conveniently omitted those on which the M3 is faster (SC)—and it's particularly BS b/c the overwhelming majority of apps used on thin&lights are SC rather than MC.
And it's not just that they showed MC rather than SC. They additionally concealed the fact that the scores were MC by labelling them as, e.g., "Geekbench - CPU Score" rather than "Geekbench - MC CPU Score".
Now you might argue that mfrs. do that all the time—they show the scores on which their devices do best. If it were just a matter of cherry-picking certain apps, that's expected. But to omit an entire class of tasks, when that's the most common class used on that category of device, and to conceal that, is an entirely different level of marketing BS.
Last edited:
It took me a while to understand what you mean by „mfrs“ 😂