[CPU only] Apple M1/2(Max/Ultra) (TSMC 5nm) vs AMD Zen 4 (TSMC 5nm) - Technical Analysis

[Xiao_Xi]

Does anyone have some recommendations which tech review outlets to follow now that both Andrei and Ian aren’t with Anandtech anymore?

Chips and Cheese write in-depth reviews of microarchitectures.

Chips and Cheese

Technology news, reviews, insights and leaks with a real technical twist, and the occasional rant or two...

chipsandcheese.com

Hopefully they will write about Zen4 soon.

You can find out more about them in the video in which they talk about Tachyum with Ian Cutress.

 

[thenewperson]

You can find out more about them in the video in which they talk about Tachyum with Ian Cutress.

Sorry to derail but man YouTube face is such so cursed goddamn

 

[altaic]

Chips and Cheese write in-depth reviews of microarchitectures.

Eh, it’s weird that they have a focus on architectures yet no articles about Apple Silicon. Even the “Arm or x86, ISA doesn’t matter” article only mentions the A4 & A5 in passing because Jim Keller also worked on Zen, and said article leans hard x86. Also, they founded their site in Dec 2020, just after the M1 was released 🤨

 

[leman]

Eh, it’s weird that they have a focus on architectures yet no articles about Apple Silicon. Even the “Arm or x86, ISA doesn’t matter” article only mentions the A4 & A5 in passing because Jim Keller also worked on Zen, and said article leans hard x86. Also, they founded their site in Dec 2020, just after the M1 was released 🤨

I remember the author from Chips and Cheese saying that they are mostly interested in x86.

 

[kvic]

Independent of the question if the cell library is custom, even the standard TSMC N5 cell library offers cells with vastly different performance and power characteristics for the same cell function. The general rule here is, the slowest cells have the lowest leakage. In a mobile design you try to minimize leakage and tend to favor slow cells - in an HPC desktop design you can shift your cell distribution easily towards the high leakage cells.

Ryzen processors benefit from the process optimisation meant for EPYC. Now 7000 series on 5nm+6nm are about to show some true colour on performance and efficiency. Apple silicon, including Mn Max die, will be always mobile first. Would be interesting to see how Apple addresses the competition on desktop/workstation, and defends their position on laptops as Zen4 mobile chips would be a direct pressure next year.

Thanks, very helpful! Hope to see some in-depth reviews once the products are out.

Does anyone have some recommendations which tech review outlets to follow now that both Andrei and Ian aren’t with Anandtech anymore?

Rumours said Ryzen 7000 embargo will be lifted on September 20, one week before products go on sale. I expect media outlets will be full of presentation slides that AMD showed the media privately in the past few days. We're likely to see high level architectural changes in Zen4. Some info already leaked officially or unofficially:

• FCLK (Infinity Fabric clock) is not as important as in Zen2 & Zen3
• DDR5-6000 is the sweet spot of Zen4's memory. FCLK:UCLK:MCLK ratio 2:3:3 i.e. optimal FLCK will be 2000MHz same as in Zen3
• Doubled L2 cache per core (1MB)
• 50% increase in L2 TLB entries for data pages
• 5GHz+ sustained all-core frequencies

We could speculate AMD should have increased number of Infinity Fabric lanes inside Zen4. Also people are speculating number of instruction decoders increased from 4 to 5 or perhaps 6.

I'm sure Ian Cutress will go through the architectural details on his Techtechpotato channel on Sept 20. But I'm not so sure who we can go to for detailed tests and wide range of benchmarks on day 0.

 

[TSE]

Can someone give me a TL;DR of how the new AMD 7000 series efficiency & performance compares to Apple's latest and greatest?

 

[Xiao_Xi]

But I'm not so sure who we can go to for detailed tests and wide range of benchmarks on day 0.

Maybe Igor's LAB or LTT?

 

[leman]

Can someone give me a TL;DR of how the new AMD 7000 series efficiency & performance compares to Apple's latest and greatest?

Based on the information provided by AMD, their new high-end desktop CPUs are faster than Apples current ultra-mobile chips designed for passively cooled chassis. We can’t say anything about efficiency.

 

[pshufd]

The Raptor Lake T chips look interesting as they are 35 Watt parts. I don't expect reviews anytime soon but something that I will keep an eye out for, particularly if they're cheap. I'm curious as to what AMD's 35 watt parts look like as well.

 

[theorist9]

I think you are not taking into account that the Studio was a replacement for the 27 iMac. At $1,999, it is already priced $200 higher than the old iMac 27" and it has no screen! Even higher and the Studio Display also being $200 higher than an Apple screen and that would make the Studio even more problematical coverage for the old iMac 27" entry point.

The 27" iMac survived just fine with a "better"/"best' iMac 21.5" price in the $1,400-1,500 range. There is zero rational motivation that presents as M2 Mini Pro in the $,1400-1,500 as being hugely flawed. It worked before well for Apple. It would work again at the same gaps between model offerings. It is the same thing only they have pruned the screens off the offerings. ( so the completed system price for users is even higher for those who need to buy a new monitor).

The Studio starting with at least 24 GPU cores and the M2 Pro likely toping at 20 GPU cores is a product differentiation gap. They can make the gap wider once get to M2 Max that has the same GPU core count bump. The Studio is going to run a bit quieter for folks who want to extra pay for that also.

M2 Pro will work because many will want to spend that gap money on more RAM , more Store , and/or something else because the M2 Pro is "fast enough". Apple 'pulling' folks into spending more on BTO options actually likely gives them higher profits.

Not sure why surprised. The iMac 27" was likely one of the better selling desktop models Apple had. The overall iMac ( non Pro 21.5-27") was by far the best selling desktop model. So a major part of that segment probably would be best (or at least minimally 'better' ) selling also.

What were iMac 27" folks suppose to do? Dump it for a M1 Mini ( with gimped Display out) or smaller screen 24" or way too expensive and increasingly dated/stale Mac Pro 2019? Probably not. [ the 16 core CPU + W5700 configuration that was most popular Mac Pro configuration is actually slower than the up configured Studio for thousands of dollars less. Why wouldn't that sell well? There are lots of folks who will gladly take the multiple thousand dollar savings. ]

Where Apple placed it was right on top of approximately the middle of the old iMac 27" configurations ( $1,999 . Not the cheapest but also not on the BTO upscale price either. It was a standard configuration that likely sold very well off the shelf. ).

P.S. The iMac Pro sold much better than many folks though it did. In its heyday it was about 5th or so best selling 27" iMac configuration at B&H ( could rank at their website). upscale Studio + Studio Display and back in iMac Pro zone price wise. Folks cross upgrading from iMac Pro -> Studio would be an easy pipeline also.

Yup. The last few times I called AppleCare, I asked the advisors what kind of Mac Apple issued them. Their responses were always the same: a 27 inch iMac. That shows you that Apple itself understands how useful a 27 inch Retina display is for general desktop use. And yet they don't currently provide a way for consumers to have an Apple system with the functionality of a 27" iMac with anything approaching reasonable consumer pricing.

The starting price of a 27" 2020 iMac with 256 GB SSD/8 GB RAM was $1800. The starting price of a Studio Display + Mini with the same RAM and storage is $2300, which is reasonably close. But, unlike with the iMac, you can't drive a total of three monitors, nor can you install 32 GB RAM (both of which are in my setup), which means you need to go up to a Studio, which puts you at $4500. So they really need either a large iMac, or a Pro Mini and a consumer-priced 27" Retina display. I personally think they should offer all three.

 

[pshufd]

I just watched MLID on Intel having to cut back on pricing for Raptor Lake because Ryzen 7xxx is better than expected in performance and more aggressive in pricing. Intel wanted to raise prices by 5-30% but it looks like Ryzen 7xxx will be roughly the same as Raptor Lake. Ryzen 7xxx will also allow you to reuse your motherboard going out to Ryzen 9xxx, which, to me, is a very good reason to go with AMD over Intel. I will reserve final judgement until I see efficiency numbers but that could take a while.

I am really happy with Apple Silicon but I am looking forward to M2 systems with 24 GB of RAM. I really like the M1 mini for its performance but I need somewhat more than 16 GB of RAM to run everything on it.

 

[Xiao_Xi]

A little more information about Zen 4.

https://twitter.com/i/web/status/1566349465740627969

https://twitter.com/i/web/status/1564405596303962113

 

[deconstruct60]

Yup. The last few times I called AppleCare, I asked the advisors what kind of Mac Apple issued them. Their responses were always the same: a 27 inch iMac. That shows you that Apple itself understands how useful a 27 inch Retina display is for general desktop use. And yet they don't currently provide a way for consumers to have an Apple system with the functionality of a 27" iMac with anything approaching reasonable consumer pricing.

The starting price of a 27" 2020 iMac with 256 GB SSD/8 GB RAM was $1800. The starting price of a Studio Display + Mini with the same RAM and storage is $2300, which is reasonably close. But, unlike with the iMac, you can't drive a total of three monitors (which is what I'm doing with mine), which means you need to go up to a Studio, which puts you at $4500. So they really need either a large iMac, or a Pro Mini and a consumer-priced 27" Retina display. I personally think they should offer all three.

A substantially large number of folks are going to get back to the 'old' iMac 21.5/27" entry prices by just avoiding the Studio display.

The trade-off for the end user is that with the Mini and Mac Studio they can pick their own display. I suspect that is in part why Apple went to an "even more expensive display docking station" pricing with the Studio Display. They probably know they are going to sell substantially fewer panels so are putting a "low unit volume" tax on it similar to the Mac Pro 2019. Crank the entry price higher and sell fewer at higher margins.


If Apple finds a way to deliver better screen and super mature 5K Retina panel prices , then they'll bring the large screen iMac back. They won't get back to the old iMac panel sales volume, but may be necessary if that gets to be a weak spot in the line up as desktop competition creeps up over the next 2-3 years.

However, if Apple 'sleeps' on the iMac 24" for well over 2 years and they can't get the costs down to old mature 5K levels, then I'm not so sure the iMac willl come back. At least, a lower end (sub $2,300 ) version. If the iMac 24" gets an M2 update before the end of 2022 that would be a good sign for a large screen iMac later. If Apple just gives up on the small screen iMac (no M2 update at all far into 2023) ... the large screen model could be on equally 'thin ice'. If Apple can get to higher aggregate unit volume sales with the M2 Pro Mini + Max/Ultra Mac Studio than they got with the old 21.5"+27" iMac combo then they might give up ( or minimally highly deprioritize) the iMac.

Or the large screen iMac comes back to a set up similar to the MBP 14/16" where largely share almost exact same motherboard and it is mainly a screen size difference of mostly the same thing (to lower costs that way). And both the 24" and 27" are on the slow motion, plain "Mx" upgrade pace as kind of "hobby products" that are pretty close to just being super-sized iPads on a stand.

 

[deconstruct60]

A little more information about Zen 4.

https://twitter.com/i/web/status/1566349465740627969

https://twitter.com/i/web/status/1564405596303962113

The combination of those two (smaller than Intel ROB) and larger micro op cache , and very limited L2 cache increase is likely very substantive contributors to the core size being about half the size of Intel Golden Cove cores.
To reorder more things you have to store more things which will take more space.

The micro ops cache is a better trade-off on increase space cost because if cover more tight loops the front of the x86 decode engine doesn't have to run as much. ( Intel is about 40-50% of time not having to decode anything from scratch. ). That is part of their "front end" feeding more instructions into the execution engines they already had improvements of Zen 4.

300 entries is still a huge bucketload of entries. There are limits in the structure of the code to how much reodering can actually get done. At some point hit the diminishing returns zone where the extra ROB entries are only good/production on fewer and fewer workloads.

 

[kvic]

AMD's roadmap diagram says that Zen 4 family is spread over N5 and N4.

I believe Zen 4 on 4nm is for mobile if it follows the pattern of Zen 3. Rumors said the ultra portable (Phoenix Point) will get 4nm up to 8 cores, paired with a beefy iGPU, and the high-performance portable (Dragon Range) will get 5nm up to 16 cores, paired with a dGPU. Both rumoured to be out in 2023.

Seems make lots of sense that M2 Pro/M2 Max will be on 3nm.

Maybe Igor's LAB or LTT?

Igor's Lab has some cool contents. For example, he has setup to precisely measure power consumptions of a system at various points. Only Igor can tell you how much power a motherboard consumes at idle and under different workloads. TL;DR: a typical ATX motherboard (with DRAM, M.2) consumes ~15W at idle and ~30W under load (if I recall correctly).
--

The rumoured info from RetiredEngineer seems to corroborate AMD's own presentation:

AMD said IPC increase (8-core chips both tested at 4GHz) from Zen 3 to Zen 4 is 13%. Overall single-core performance increase from Zen 3 to Zen 4 is 29%. So other than IPC uplift, frequency uplift contributed the remaining 16%.

Among the IPC uplift, as shown in the above break-down, contribution from Front End improvement is ~40%, Load/Store ~24%. What major change in Front End could bring great impact? Perhaps from 4 instruction decoders in Zen 3 to 6 in Zen 4.

Anandtech found M1 has 8 instruction decoders. By now x86_64 chips (Intel already did; Zen 4 likely the moment for AMD) pretty much have caught up and close.

 

[theorist9]

A substantially large number of folks are going to get back to the 'old' iMac 21.5/27" entry prices by just avoiding the Studio display.

The trade-off for the end user is that with the Mini and Mac Studio they can pick their own display. I suspect that is in part why Apple went to an "even more expensive display docking station" pricing with the Studio Display. They probably know they are going to sell substantially fewer panels so are putting a "low unit volume" tax on it similar to the Mac Pro 2019. Crank the entry price higher and sell fewer at higher margins.

The problem with that is they don't get the Retina display and, particulary with the loss of subpixel text rendering after High Sierra, MacOS needs Retina to look its best. A Mini or Studio plus a 4k 27" display is a sigificantly downgrade in user experience, at least visually, compared to a 27" Retina iMac. At least that's what I see, and I'm running both displays side-by-side.

 

[pshufd]

The problem with that is they don't get the Retina display and, particulary with the loss of subpixel text rendering after High Sierra, MacOS needs Retina to look its best. A Mini or Studio plus a 4k 27" display is a sigificantly downgrade in user experience, at least visually, compared to a 27" Retina iMac. At least that's what I see, and I'm running both displays side-by-side.

I concur. I have three 4k 27 inch Dell Ultrasharps next to a 2014 5k iMac.

 

[deconstruct60]

The problem with that is they don't get the Retina display and, particulary with the loss of subpixel text rendering after High Sierra, MacOS needs Retina to look its best. A Mini or Studio plus a 4k 27" display is a sigificantly downgrade in user experience, at least visually, compared to a 27" Retina iMac. At least that's what I see, and I'm running both displays side-by-side.

That is not a 4K can't be Retina problem. That is a macOS is crappier than Windows on text rendering problem. It is a software problem, not a hardware one. The problem with that is how long Apple can run with the "we build a crappier OS than them" OS when selling an even higher number of headless desktops than before.

Apple can force all iMac 27' buyers into looking at the Apple 27" display. However, some folks are not going to buy 27" 4K display. Some folks will buy 32" 4K display and put them further back . Some will buy ultrawides that Apple doesn't sell. Some folks will have 27" 4k displays with better contrast ratio (IPS Black) and/or more dynamic range; probably not going to get most of those folks to backslide for more money spent. Most of the folks who already own a modern , $400+ 4K display are more likely to keep it (on a sunk cost basis) than run off and buy an Apple version that costs 4x as much.

It would be different if Apple sold an even halfway decent affordable display, but they don't. $1,599 gets you a non ergonomic stand or no stand at all. The attachment rate for Studio display is probably way less than 50%. Apple is selling as many as they make, but lots of folks are looking at something else.

 

[deconstruct60]

I concur. I have three 4k 27 inch Dell Ultrasharps next to a 2014 5k iMac.

And what does Dell sell more of non Ultrasharps 27" or Ultrasharp 27" 4ks ?

The question is what are most people going to buy, not what some people are going to buy. There is highly likely that Apple is going to force someone who has a $800 (or less) monitor budget into buying a $1,999 monitor (presuming they don't want to give up an adjustable stand.) .

 

[leman]

That is not a 4K can't be Retina problem. That is a macOS is crappier than Windows on text rendering problem. It is a software problem, not a hardware one. The problem with that is how long Apple can run with the "we build a crappier OS than them" OS when selling an even higher number of headless desktops than before.

Text will always look sharper on a 5k screen compared to a 4’ scrim at the same resolution. What about it is a “software problem”? You don’t use subpixel rendering on high-DPI displays and Apple does care about low-DPI. And given that sub pixel rendering is a hot mess that is difficult to implement and breaks GPU composition, retiring it was only a question of time.

Regarding Windows text rendering… Windows has long been known to render fonts incorrectly to improve perceived clarity, which is hardly “better text rendering”. Fortunately modern displays don’t need these hacks.

 

[pshufd]

And what does Dell sell more of non Ultrasharps 27" or Ultrasharp 27" 4ks ?

The question is what are most people going to buy, not what some people are going to buy. There is highly likely that Apple is going to force someone who has a $800 (or less) monitor budget into buying a $1,999 monitor (presuming they don't want to give up an adjustable stand.) .

I've been doing a lot of playing around with 4k monitors this evening and what seems to work with scaled resolutions is Apples USB-C output which supports 6k. I think that support for 4k (as in the HDMI output from the mini and even the MacBook Pros) doesn't have the nice HIDPI font sharpness that you get from USB-C output. My guess is that it's rendered at a higher resolution and then downscaled.

Whatever the case, USB-C output looks great and HDMI output looks fuzzy. Yes, it could be done in software but I wonder if Apple does this in hardware so that they don't have to bother with software. I think that this is fixed with a Mac Studio as it has 4 USB-C ports that can output to monitors.

At the moment, I have the USB-C set up in scaled mode and the other two set up as native 4k. That will have to do until I get a Mac Studio.

 

[theorist9]

Text will always look sharper on a 5k screen compared to a 4’ scrim at the same resolution. What about it is a “software problem”? You don’t use subpixel rendering on high-DPI displays and Apple does care about low-DPI. And given that sub pixel rendering is a hot mess that is difficult to implement and breaks GPU composition, retiring it was only a question of time.

Regarding Windows text rendering… Windows has long been known to render fonts incorrectly to improve perceived clarity, which is hardly “better text rendering”. Fortunately modern displays don’t need these hacks.

Yes, Retina displays (~220 ppi for external monitors) don't need these hacks to look good. But other modern displays—like 4k@27" (~160 ppi)—do (at least subpixel text rendering). [Saying only Retina displays qualify as "modern" would be, IMO, way too Apple-centric.]

MacOS vs. Windows text rendering is a long-standing subject of debate. I think the summary is that Windows sacrifices font fidelity for sharpness. If you're a graphic artist, you'll probably value font fidelity. If you're someone who doesn't have a Retina display, and needs crisp-looking fonts to avoid eye fatigure, you'll probably value sharpness.

At least that's the general concept. I haven't seen Windows on a 4k 27" to compare.

 

[leman]

Yes, Retina displays (~220 ppi for external monitors) don't need these hacks to look good. But other modern displays—like 4k@27" (~160 ppi)—do (at least subpixel text rendering). [Saying only Retina displays qualify as "modern" would be, IMO, way too Apple-centric.]

Personally I would characterize a 4K 27-32” monitors as “retina”. Viewing distance does play a role and typical distances to a desktop monitor tend to be larger than to a laptop screen. In my experience at least, a good quality 4K 27” works fine with macOS text rendering. I have an AOC 32” 4K display at home and it’s ok, not as sharp as the LG Ultrafine of course, but perfectly acceptable to my eyes.

 

[theorist9]

Personally I would characterize a 4K 27-32” monitors as “retina”. Viewing distance does play a role and typical distances to a desktop monitor tend to be larger than to a laptop screen. In my experience at least, a good quality 4K 27” works fine with macOS text rendering. I have an AOC 32” 4K display at home and it’s ok, not as sharp as the LG Ultrafine of course, but perfectly acceptable to my eyes.

I think this is one of those YMMV things. At my typical viewing distance (19"-20"), I was perfectly happy with text on my 27" 4k with High Sierra (last OS with subpixel text rendering). But I disliked text on that monitor with any OS after that. It's like my eyes were always trying to focus on it to make it sharp, when it couldn't be made sharp, which is fatiguing.

Hence when I had to upgrade to a newer OS, I also had to upgrade to a Retina display, and got a 2019 27" iMac, which gives me sufficiently sharp text without the subpixel rendering. I still use the 27" 4k, but only as a side-monitor, where my eyes are on it a sufficiently low percentage of the time that it's tolerable.

Because a lot of people like to use their laptops as desktop replacements at home, pairing them with a external monitor, I think Apple should offer consumer-priced Retina displays. There's a lot of people in your camp, but a lot in my camp as well. And Apple itself believes MacOS only looks optimum on a Retina display.

 

[kasakka]

Yes, Retina displays (~220 ppi for external monitors) don't need these hacks to look good. But other modern displays—like 4k@27" (~160 ppi)—do (at least subpixel text rendering). [Saying only Retina displays qualify as "modern" would be, IMO, way too Apple-centric.]

MacOS vs. Windows text rendering is a long-standing subject of debate. I think the summary is that Windows sacrifices font fidelity for sharpness. If you're a graphic artist, you'll probably value font fidelity. If you're someone who doesn't have a Retina display, and needs crisp-looking fonts to avoid eye fatigure, you'll probably value sharpness.

At least that's the general concept. I haven't seen Windows on a 4k 27" to compare.

Yes, Windows generally forces text to the pixel grid so there is no blurring of fonts like MacOS has at the expense of font shapes. I think it's a very working compromise for better looking text especially on lower res displays. Windows does tend to do worse for oddball pixel structures like what LG and Samsung OLED panels have (WRGB and triangular RGB).

MacOS uses a very simplistic way to handle HiDPI scaling. Target resolution (aka "looks like AxB") -> render at 2x -> downscale to native resolution. So e.g "looks like 2560x1440" -> render at 5120x2880 -> downscale to native res. Which just happens to perfectly fit the 5K Apple Studio Display, resulting in no loss of fidelity. Using any other scaling level you will have quality loss but with the high PPI of the ASD it's unlikely you will notice it much.

Personally I don't mind the same scaling using a 4K display even if it is less sharp and appreciate the massively cheaper cost, more flexible inputs, higher refresh rate and better stand options etc of my 28" 4K 144 Hz screen.

I'm not quite sure how Windows handles this by comparison but it seems much more complicated and more in the hands of developers on how exactly to set it up. Results in a mixed HiDPI experience but most apps nowadays are fine with this, I tend to only see scaling issues in things like installers looking blurry.

A big issue on MacOS is that Apple imposes arbitrary rules on "what is HiDPI capable". Apps like BetterDisplay would not need to exist if it wasn't often doing the wrong thing. For example the LG DualUp 28" 16:18 2560x2880 display has the same PPI as a 32" 4K screen but because it's horizontal resolution is less than 3840 then MacOS just decides "not a HiDPI display". They should simply adopt Windows' "the user can decide for themselves" way for this stuff.