Hacker News Reader

Comment by fxtentacle | Hacker News Reader

To me, that move makes a lot of sense. But judging from the other comments, I'm the only one with that assessment.

I'm my opinion, the secret sauce that makes Intel dominate certain industries is software. And it has been for some years already.

If you need really fast mathematical number crunching, e.g. high frequency trading or realtime audio filtering, then you need MKL, the Intel math kernel library.

If you want to further reduce latency with parallelism, you need TBB, the Intel thread building blocks.

Raytracing? Intel embree.

Once you are locked in that deeply, the raw Intel vs AMD performance becomes meaningless. You only care about how fast the Intel libraries run

So a CEO with experience building high performance low level software seems like an amazing fit.

Edit: And I almost forgot, the Intel compiler used in pretty every PC game to speed up physics. Plus some people have seen success replacing GPUs with icc+avx for huge deployment cost savings in AI.

View on HN

To me this description sounds like a specialist high performance computing company rather than a consumer technology company. That may be a perfectly reasonable market to be in, but is that type of company worth $200bn? I'm not sure.

Roughly 40% of their revenue is consumer chips where, apart from some games optimisation, they are no longer standing out from the crowd, and the leader is arguably Apple, with AMD doing well. The next ~30% of their business is servers, where there may be a significant number of HPC clients, but the bulk of this is again likely to be VMs running non-Intel specific software, and this market is starting to realise that Intel is nothing special here.

Looking at their revenue breakdown, I struggle to put more than 20% into the things that you mention they are great at. Should they focus on this? It would lose them much of their market cap if they did.

I agree with your market breakdown, but surely not with your assessment.

In the consumer segment, you have regular people trying to make vacation videos with software like Adobe Premiere and Adobe Media Encoder, or Magix. Nvenc quality is bad. AMD is horribly slow. The only fast high quality encode is with Intel's dedicated CPU instructions, which both apps heavily promote to their users.

And the 30% that you mention that run VMs... Wouldn't they be pretty happy if Intel added dedicated CPU instructions to make VMware better?

I agree that for the work that I do, AMD is as good as or better. But people doing highly parallelizable tasks like compiling are the minority.

I think you might over estimate the prevalence of video editing software like this. Adobe don't appear to sell consumer versions anymore, it's only pro subscriptions now. Magix is sold at a "vacation video friendly" price, but doesn't mention Intel in their marketing material.

I just don't think the market for home devices is thinking about their video encoding time when they buy a laptop, but I do think they'll use an M1 Mac and find it surprisingly fast, or hear from a friend or family member that they are really good.

Intel just haven't been optimising for the main user experience seen by these people, or those writing "normal" server software either. They've been pushing AVX512 instead, which looks good for video or things like that, but not for regular use-cases.

Another good example is how fast the M1 chips (and the A chips in iPhones) perform at Javascript benchmarks. Those benchmarks look a lot more like what most people are doing most of the time than video encoding benchmarks.

> but I do think they'll use an M1 Mac and find it surprisingly fast, or hear from a friend or family member that they are really good.

That only happens in California.

MAGIX heavily mentions Intel in their marketing.

"4K Ultra HD video editing with Intel and MAGIX"

"Enjoy HD Video editing with Magix Movie Editor Pro and Intel Iris Graphics"

"Edit in 4K Ultra HD" + Intel Logo

"Finish and Share videos quickly with Intel Iris Graphics"

https://www.youtube.com/watch?v=eT9KOtN7KFM

Plus, as a user of the software, I can tell you that if you tick the "Hardware Acceleration" checkbox on AMD, a popup will tell you to buy a supported Intel CPU and then turn the checkbox off again.

BTW I'm picking Magix here because in the local electronics store, that's the video software that you can buy as a box and that is featured in bundles with Intel laptops. So if someone clueless walks in there and says they need video editing, this is most likely what they will end up with.

It's not just marketing but also uses Intel QSV. AMD RADEON has similar feature VCE but its quality is relatively bad and its adoption isn't well, compared to QSV/NVEnc.

>if Intel added dedicated CPU instructions to make VMware better

They (and AMD) did years ago. Intel VT.

> And the 30% that you mention that run VMs... Wouldn't they be pretty happy if Intel added dedicated CPU instructions to make VMware better?

They have this today. What would make them happier is cutting power utilization by half or more, which is looking quite possible with non-Intel Silicon.

> AMD is horribly slow. The only fast high quality encode is with Intel's dedicated CPU instructions,

You might need a source for that.

Intel Core i7 6700K is double the FPS of AMD Ryzen 5 1600X https://www.magix.info/de/forum/ein-performancetest-zwischen...

Seems like the clock, RAM, and GPU differences there may have had an effect? Comparing between systems that different seems unusual.

this is article from 2017 which compares 5 generation old cpu vs other 5 generation old CPU

Uh, those numbers are more than three/five years old at this point. Beyond comments on the test bench not being properly set up, Ryzen has improved significantly since then.

AMD's latest consumer-level chips significantly outperform Intel's chips in both price and performance. When talking about prosumer video editing performance, the Ryzen 9 5900x, the second most expensive "new" chip from AMD is a 3.4% performance improvement over Intel's most expensive "new" chip 10980XE. Additionally, the 5900x retails for $549 USD while the 10980XE retails for about $1,000 USD.

https://www.pugetsystems.com/labs/articles/Adobe-Premiere-Pr...

On first generation Ryzen AVX2 instructions were executed as 2 AVX instructions as the AVX pipeline was 128 bits wide. This was fixed in Zen 2 and nowadays we are at Zen 3.

> AMD is horribly slow

Not sure where you're getting that these days? Absolutely in the days of Bulldozer, but AMD's Zen 3 architecture has taken even the single core lead from Intel, not to mention the multi core lead they've held for several years now.

The encoder?

AMDs GPU encoder still lags a way behind Nvidia for example

>Roughly 40% of their revenue is consumer chips where, apart from some games optimisation, they are no longer standing out from the crowd, and the leader is arguably Apple, with AMD doing well.

You lost me at Apple. Apple owns around 15% of the PC market space and almost the entirety of that is Intel-based systems. Outside of HN, nobody cares about the M1 chip, it isn't a selling point to my mom or her friends. If someone at the Apple store recommends it they might buy it instead of an intel-based system but it definitely isn't something they're seeking out.

The only threat Intel has right now in the consumer space is AMD, and it's a very real threat. AMD won both Sony and Microsoft console designs, and the mobile Ryzen 5000 chips released at CES look to have enough OEM design wins to put a serious hurt on Intel in 2021.

Even if Apple goes 100% M1, there's the other 85% of the market that Intel is likely far more concerned about.

I think you are vastly underestimating the revolution of what the M1 represents to the PC industry.

I think you are vastly overestimating the "revolution" of what the M1 represents to the industry. Apple isn't selling it to any other PC makers, and corporations aren't pivoting away from Microsoft for a CPU. Every single ARM chip that's been targeted at the Windows world has produced yawn-inducing performance.

M1 is amazing. I think people are also underestimating how fast Intel can catch up.

I think you are vastly underestimating how hard IT departments will kick you if you request or bring a device that cannot properly execute the company-critical legacy Windows x64 software. Like SAP, for example.

(SAP is the largest non-American software company by revenue and does business management, workflow automation, and bookkeeping)

My prediction is that outside of hipster startups, M1 will have no effect on business laptop sales.

Isn't sap a database? Why would you want users to run db on laptop, especially on windows?

No, SAP is more like an Operating System for your factories. It contains EVERYTHING, from payroll to inventory management. Think of it more like an Exchange server plus all Microsoft office apps combined. To connect to the Exchange server and get all features, you need Outlook. It's the same with the SAP database and SAP client GUIs.

The official GUI is C++ and Windows only. They do have a Java port for other OSes, and some 3rd party GUIs, but none of that is feature-complete or even halfway there.

SAP is so many things that their Products page has search functionality and is broken down by first letter: https://www.sap.com/products-a-z.html

So no, SAP is not a database and has many client applications that would run on a laptop.

SAP is lots of enterprise software stuff, not just a DB.

But then again, the current version of SAP is S/4 HANA, and unless you are a developer or admin for that, you will be using their Fiori based web clients, so a normal browser is enough. I am a developer in an S/4 rollout project in a Windows-only shop, but for our future system landscape I could see the normal people using the SAP systems using any kind of laptop or tablet. Even we are testing iPads and laptops at least.

> I think you are vastly underestimating how hard IT departments will kick you if you request or bring a device that cannot properly execute the company-critical legacy Windows x64 software. Like SAP, for example.

Many companies have long ago set up some beefy Citrix servers for those application.

M1 with Rosetta emulation is still faster or comparable to top Windows laptops, with room to run for M2.

>My prediction is that outside of hipster startups, M1 will have no effect on business laptop sales.

This is what everyone always says but the iPhone kicked off a whole BYOD trend that has ended up with many high-value employees caring a lot about what tools they have to use, and a lot of software engineers want Macs.

I’ve used (and developed software for, and helped administer) SAP from my Mac on and off for 20 years. Tends to be easier these days now that everything is mobile and web.

A lot bigger companies than “Hipster startups” use Macs, this tends to start with the C-suite and people follow suit.

My point also wasn’t that everyone was going to switch to Mac. It was that M1 proves you can build a “better in every way” PC with an ARM architecture. Linux ARM is also being pushed by AWS heavily from the server side with impressive price/performance numbers.

Windows ARM has been failing for many years, but I suspect this is going to change. Microsoft has a talented virtualization group, where the HyperV roots go back to the Connectix Virtual PC team that built PPC/x86 emulation for the Mac. I suspect they can pull off something like Rosetta - they just need a chipmaker to collaborate with. Might even be Intel! Pat Gelsinger is an outside of the box thinker.

You remind me of folks that thought the iPhone / iPad would have no impact on Blackberry sales, as real businesses need keyboards.

Not sure of the source of your 15% but I'm willing to bet that by value it's more - no Celerons in Apple's line up. Plus Apple wouldn't be going down this route if it didn't expect to grow market share - and although people don't care if it's M1 or i5 they do care if the experience is better.

Then Apple's success with the M1 will spur others - I would not be surprised if Microsoft follow them down the same route.

> Apple wouldn't be going down this route if it didn't expect to grow market share

Marketshare is not what Apple is about. Apple is about profitability and control. Their move to own silicon is driven by improvements in the reliability of their build pipeline (no more waiting for tic-tocs and whatnot) and tighter control / integration of their whole stack (same arch on phones and pc). That these chips happen to perform so well that they are potential market-growers, is a welcome coincidence.

Growing marketshare but profitably and without impairing the brand is what Apple is about. That's why we have the iPhone SE. The M series lets them do that with the Mac now. And more Macs implies more Apple services sales.

It's certainly partly defensive - they were frustrated with Intel - but Apple would only make a move of this scale if it thought it created business opportunities for them.

>it isn't a selling point to my mom or her friends

Gargantuan battery life isn't a selling point? For laptop? In what universe?

the m1 isn’t competing on many of the same axes as an intel or amd cpu, because it’s necessarily packaged inside of an entire computer built around it. that computer is a mac, which might be different from the purchaser’s current os so they decide not to switch, or they already bought software for windows and want to use it there, or they’re married to the microsoft ecosystem, etc.

Office runs on Macs too!

Seriously, I predict we will see Apple successfully attack the sub $1000 laptop market within two years. They sell the iPhone SE with an A13 for $399 so they could easily do so now they no longer have the 'Intel tax'. And the products will be a lot better than the Windows equivalents.

Most home users might use Office and that's about it. The allure of the Apple ecosystem will be strong especially for iPhone users.

I’m skeptical.

Apple’s bread and butter, as far as Macs go, is the MacBook Air. And by all accounts, they sell a lot of those, and will presumably sell even more, with better margins, now that they’ve gone ARM.

Do they really want to undercut that with a cheaper laptop? I suppose it’s possible, if the volume/margins works out, but I’d bet they just keep plugging along with $999–$1500 13-inch laptops.

All fair points but I think that with higher margins it tips the balance towards market share growth. Key issues are 1) can they make an acceptable margin on a good $800 laptop and 2) can they genuinely significantly grow market share rather than lowering average selling price - i.e. can they maintain distinction between $800 and $1000 products. Given what we've seen them do on iPhone and iPad I bet then answer is yes to both of these.

Bear in mind too that after a generation or two they can put the last gen M chips in cheaper products.

I think there's an economic principle here (and I don't know the sign), but this is all assuming a frictionless vacuum - in practice, Apple cannot sell 25% more M1 Macs if they lower their price to $800, or whatever, since their marginal costs rise in that case (because TSMC is totally booked!).

That's today but I'd expect next year's sub $1000 Macs will use previous year's M series chips in due course. (Exactly the iPhone and iPad playbook).

I'm not really sure to be honest. We've moved on from laptops that you can't watch movie on without charging few years ago. I don't really care if my laptop works for 10 hours or 15.

It’s the same reason battery life is so critical in EV’s. Smaller batteries need to be charged more often which eats up more of their remaining lifespan. It’s a downward spiral that means a 50% extra lifespan up front can be worth 100% extra lifespan in 3 years.

Laptop batteries are also expensive in terms of money, weight, and bulk which puts Intel into a much larger bind.

For who? My mom uses her laptop at home 99% of the time, if the battery gets low she plugs it in. She needs a battery that will last 1-2 hours for the 3 times a year she flies.

You can find a place to plug in at basically any coffee shop or library you go to. My mom isn't spending 10 hours in a datacenter, so it doesn't really matter to her if the battery life is 3 hours or 12. For the average consumer, battery life has just been another stat on the spec sheet for years now.

Even if your mom doesn't care about battery life, she will probably buy a product that is also sold to buyers who do care about battery life, and if the product can meet those buyer's needs with a smaller battery, then your mom's laptop will be lighter.

So, does your mom also not care about the weight of her laptop?

The Macbook air with intel CPU weighs 2.75 lbs, the Macbook air with M1 weighs 2.8 lbs. The macbook pro is 3.0 vs 3.1 lbs.

The weight is a non-factor. Quite frankly until you start cracking 5lbs nobody even cares in my experience. Apple's maniacal focus on making laptops skinnier and lighter has done a disservice to the entire product line, which they seemingly acknowledged with the 16" Pro.

Apple had the option of making the M1 Air lighter (by choosing a smaller battery) but decided instead to greatly increase battery life. The point remains that Apple has choices that vendors of laptops reliant on Intel CPUs do not have, which might end up eating into Intel's market share.

Strongly disagree. I'm writing this on a 12" MacBook with a partly broken screen - by far my favourite machine (I have a 2020 MacBook Pro too). I'm not alone too.

I expect we'll see an M series MacBook again but this time it won't be underpowered.

> For the average consumer, battery life has just been another stat on the spec sheet for years now.

I'm not sure I agree with this. I think if you asked someone whether battery life was a priority, they might say no. And if you asked them to rank tech specs I'm not sure it would necessarily be that high either. But the experience of using a laptop with a noticeably better battery is, for me, quite likely to be one of those things that you didn't know you were missing, even if you just charge it every now and then.

This resonates with me too - anywhere it would be comfortable to work with my laptop is going to have somewhere to plug it in to charge.

For consumers? It’s a race to the bottom. Mom wants to pay $200 if anything. My in-laws do their taxes on their phone.

> it isn't a selling point to my mom or her friends

Really? That's surprising to me. I'd imagine that for the demographic of her and her friends, quality of life increases for their phones are far more material than for their computers.

It sounds like you're agreeing with your parent comment.

> phones are far more material

Thus they don't really care about laptop battery life.

That's not what I'm getting at. What I'm getting at is that to say that "even if Apple goes 100% M1, there's the other 85% of the market that Intel is likely far more concerned about." is somewhat far-fetched. Apple can make their own desktop chips because it's an easier problem than making a phone chip (performance/thermal efficiency), but Intel can't make because they've sacrificed thermal efficiency time and time again -- not just this time but a decade ago. Remember Prescott?

I think that is why Intel should be (and probably is) worried about Apple. They will make Intel redundant by having solved a harder problem which their own problem becomes a subset of.

I get your point, but I think the M1 is more significant as proof of what is possible than because I think everyone will buy a Mac.

I can absolutely see Qualcomm offering laptop chips off the back of the M1's success. They may not be as good, but they might be much cheaper. I can also see Microsoft pushing Windows on ARM harder, and rolling out their own chips at some point.

Also once the market gets "used to" multi-architecture software (again), I think we'll see a renaissance of chip design as many more players crop up, because of the lower barrier to entry.

Maybe. Apple has solved the chicken and egg problem regarding software compatibility by forcing everyone to move on to ARM in the near future. Microsoft will not abandon x64 though, so there are far less incentives to port things. Also Microsoft cares far more about compatibility (e.g. 32-bit software). That means means a lot of things will run under a Rosetta 2 like system (probably less efficient if you need to support 32-bit as well). If you add the fact that Qualcomm is unlikely to match Apple in performance, the resulting product might not be very appealing compared to a classic x64 system.

An ARM transition isn't a fait accompli just because Apple introduced M1 at the lower end of the Mac lineup. There's a huge lump of inertia there.

I'm not so sure, since a lot of software is somewhat cross-platform now anyway. You need to port the JVM, .NET, and Chromium (for electron), and you've effectively ported a large part of the desktop application space already.

Web applications are portable if you have a browser

Many modern languages that compile to native, like Rust, trivially recompile for multiple target architectures.

Others are dynamic and don't need recompilation.

Of course some major software is written in non-portable C and C++. But the question is whether some emulation isn't acceptable here.

Porting the JVM to a new CPU architecture is hard but it has already been done for AArch64.

I know, I'm just thinking in today's day and age an architecture change isn't that hard anymore. We're already paying for all these abstractions, why not use them.

I don't think that there is a high performance JVM implementation for RISC-V. A vendor would have to do a lot more work if they picked an architecture other than AArch64.

True, but Apple has been the harbinger for a LARGE number of advances in consumer computing adoption. Mouse/windows desktop interface? Apple. Font libraries? Apple. USB? Apple. Touch interfaces? Apple. Modern smartphone? Apple.

The list goes on and on. M1 could be such a harbinger. Who knows if anyone else can compete, and likely Apple will be in the pole position for some time, but Apple has shown everyone it can be done.

At WWDC last year they made it clear that every Mac is going to run on ARM eventually.

> An ARM transition isn't a fait accompli just because Apple introduced M1 at the lower end of the Mac lineup. There's a huge lump of inertia there.

If you're just referring to Apple's future, the endgame is already a done deal. They don't roll back something this fundamental once it's public.

Their last Intel computers will be manufactured and sold within 3 years; I'd wager a month's salary on that if I had it lying around.

If they can't make it perform at the top level like AMD, which seems unlikely but obviously it's too soon to tell, they'll make up for it in other ways or simply live with the performance hit until they can.

Apple doesn't believe in inertia. Their partners and customers can either make the transition or get left behind. Microsoft's business model depends on making enterprises happy; Apple doesn't care.

Large parts of Windows 10 are still Win32 not UWP.

They can't even remove decades old legacy code from their own products, so good luck everyone else.

You misunderstand the point of the M1 out of Apple, and for that matter the graviton2 instances out of AWS. What was demonstrated in the marketplace is that the biggest tech companies are now able to develop in-house processors that are more cost efficient and more performant. These processors are based on ARM and have minimal overhead licensing costs, as compared to buying Intel or AMD chips for their vast fleets / products.

If AWS and Apple can do it, soon other very large companies will, but in a few years, even OEMs will be able to develop their own chips. The market for high end gaming is unlikely to be touched, but the vast consumer market is going to be eaten by custom made ARM-based chips.

So in a world where processor design becomes a commodity, what does that mean for Intel and AMD? And what does that mean for the overall datacenter, consumer markets?

> So in a world where processor design becomes a commodity, what does that mean for Intel and AMD? And what does that mean for the overall datacenter, consumer markets?

Processor design is already a commodity, and has been for many years. Any company with the cash can buy a license to the ARM64 instruction set and the reference core design, and have someone like TSMC or Samsung manufacture it.

These designs haven't taken over desktop market from x86 yet because those designs simply weren't performance-competitive with what AMD and Intel are pumping out, and it's not clear that that'll change anytime in the foreseeable future.

Apple knocked it out of the park with the M1, but they've been kicking ass for years, including their competition in the ARM processor space. Just because Apple's processors happened to use an ARM instruction set, doesn't imply that an ARM revolution is upon us.

Apple isn't using a reference core, though. They are using entirely custom cores.

My mom would love to be able to use the same apps on her phone and her computer. I've been thinking about suggesting her next computer be an apple one since she got her first iphone and this makes it easier. Your Mom May Vary.

The compiler tricks can only get you so far. I administered a HPC cluster and we have a lot of software dependent on MKL and BLAS. However, with the lucrative performance boost AMD seems to put out, open source libraries like BLIS and open BLAS are attempting to fill gaps. Trust me, no one likes the intel lock-in if there is an alternative that is even close enough in performance.

> I'm my opinion, the secret sauce that makes Intel dominate certain industries is software. And it has been for some years already

Intel's secret sauce is inertia.

The thought that Intel's is not challengeable, and the world doesn't need a company to dethrone it either.

But that assumption is no longer true, and the counter movement is in its full swing.

The future of computing is on not CPU if you ask me. It would move from general computing to heterogeneous computing, and possibly application-specific chips/FPGA. MKL is fast, probably, but GPU and ASIC would be even faster.

> If you want to further reduce latency with parallelism, you need TBB, the Intel thread building blocks.

That's not how latency works and there is nothing too special about Intel's TBB library. It is a big bloated group of libraries that doesn't actually contain anything irreplaceable. Don't be fooled by marketing or people that haven't looked under the hood. It should also work on amd cpus.

> Raytracing? Intel embree.

Embree is a cool convenience, but also doesn't marry anyone to intel cpus.

mkl runs fine on amd now once you un-handicap it

Reply on news.ycombinator.com