Is ARM the future?

[2FA]

2 hours ago, Zodiark1593 said:

Internally, high performance “x86” CPUs break down x86 instructions to instructions that are remarkably reminisce of RISC. Intel had done this since the Pentium Pro.

Yes, that's the complex part of the CISC name. The instruction is complex internally doing many things so that the machine code doesn't need to handle a lot of that, the instruction set does it for it.

 

2 hours ago, Zodiark1593 said:

On the other end, ARM had increased instructions within their ISA over the years. Both ISAs implement similar techniques to improve performance (pipelining, superscalar execution, OoOE, caches, etc). The performance of ARM’s SVE (the parallel to AVX) will likely be determined by the implementation by ARM themselves, or those with an architecture license. 

Unfortunately SVE hasn't really been implemented widely yet unlike AVX which is in pretty much every CPU made by Intel and AMD except for extremely low power chips thanks in part to ARM focusing on low power uses for so long. Yes it's a few years newer but even original AVX was more widespread at the equivalent point in time.

[Bombastinator]

On 3/12/2021 at 12:28 PM, .Apex. said:

The M1 has demonstrated that ARM could have a much higher IPC than even Zen 3 at low frequency, if an M1 equivalent reaches desktop for the mass market (not just Apple devices) then it could be clocked much higher, though currently for Snapdragon the IPC is about the same as Zen 3 if taking rough numbers into account, I wouldn't want the switch to ARM to happen anytime soon and hopefully not ever, mainly for software backwards compatibility, but it's ignorant to say that it has no potential in single-threaded performance

Doesn’t Rosetta fix the x86 software issue though?  I heard 20% performance hit. So about 1.5 generations of cpu.

[Donut417]

6 minutes ago, Bombastinator said:

Doesn’t Rosetta fix the x86 software issue though?  I heard 20% performance hit. So about 1.5 generations of cpu.

Yeah but we are talking Microsoft here. They have a compatibility layer for their ARM surface tablet. Last I checked their implementation was lacking quite a bit. I mean they could improve it, but even Rossetta 2 doesnt necessarily run everything flawlessly. Microsoft has to cater to Enterprise users. Apple does not. Apple knows they have a cult following and knows that they can do what they want and people will still buy their products. Microsoft on the other hand doesnt really have that. Sure they have the market share, but mostly due to businesses adopting their OS and software back in the early days. There are many Linux Distro's out there now days. Microsofts market share has been slipping, Im sure they have noticed. 

[Eigenvektor]

1 hour ago, Bombastinator said:

Doesn’t Rosetta fix the x86 software issue though?  I heard 20% performance hit. So about 1.5 generations of cpu.

Apple's Rosetta 2 is very good, but as far as I know it isn't perfect. There are programs that have compatibility issues or don't run at all. Apple also seems to have included some form of hardware acceleration in their CPU, which is why its performance impact is as low as it is. Microsoft has no such control over the hardware, so their competing solution is software only, as far as I know.

 

Even worse, Microsoft's emulator has only supported 32 bit for the longest time. Support for 64 bit emulation has been available as a preview only since Dec 2020. Unless they can get it working perfectly, I don't see many businesses moving to emulation. To get them to move, it would not only have to work, the cost of new hardware and the running costs due to higher efficiency would have to be very compelling compared to x86 to offset the potential risks.

[saltycaramel]

What if we’re all missing the point and the “future” (for mainstream personal devices and some work devices) is not about this or that ISA per se but about the “appliance”-ization of the silicon inside a PC? 

 

Like extreme customization (year after year) of the SoC to achieve supreme efficiency and speed in doing what the software dictates. Like it’s a washing machine or a smart fridge. 

 

Apple is now in the position of “playing Lego” with blocks of the SoC year after year. It’s not just about the CPU cores and the ISA. 

[Bombastinator]

5 hours ago, saltycaramel said:

What if we’re all missing the point and the “future” (for mainstream personal devices and some work devices) is not about this or that ISA per se but about the “appliance”-ization of the silicon inside a PC? 

 

Like extreme customization (year after year) of the SoC to achieve supreme efficiency and speed in doing what the software dictates. Like it’s a washing machine or a smart fridge. 

 

Apple is now in the position of “playing Lego” with blocks of the SoC year after year. It’s not just about the CPU cores and the ISA. 

Your describing IOT?  I avoided “smart” stuff because it seems to have really severe security issues. 

[Obioban]

1 minute ago, Bombastinator said:

Your describing IOT?  I avoided “smart” stuff because it seems to have really severe security issues. 

Not if you’re in HomeKit and using a HomeKit secure router— you can deny the IoT devices internet access, but still access them remotely through your AppleID.

[Bombastinator]

1 minute ago, Obioban said:

Not if you’re in HomeKit and using a HomeKit secure router— you can deny the IoT devices internet access, but still access them remotely through your AppleID.

So still security issues but with a firewalled trusted router.  Not for me.  I don’t trust the trusted systems concept.  It’s a lot like quarantine.  If something gets behind the firewall there are all sorts of problems.   This thread isn’t about IOT though.  That’s a different thing entirely.  The concept that generalized computing will be totally abandoned in favor of a series of smart devices strikes me as wildly unlikely.

[Bombastinator]

10 hours ago, Donut417 said:

Yeah but we are talking Microsoft here. They have a compatibility layer for their ARM surface tablet. Last I checked their implementation was lacking quite a bit. I mean they could improve it, but even Rossetta 2 doesnt necessarily run everything flawlessly. Microsoft has to cater to Enterprise users. Apple does not. Apple knows they have a cult following and knows that they can do what they want and people will still buy their products. Microsoft on the other hand doesnt really have that. Sure they have the market share, but mostly due to businesses adopting their OS and software back in the early days. There are many Linux Distro's out there now days. Microsofts market share has been slipping, Im sure they have noticed. 

Businesses usually have pretty specific individual requirements though.  A given business will need to run a particular set of programs which generally isn’t all that large, but can be very obscure.  If all the programs happen to run on rosetta2 they’re good for mac.  They won’t necessarily do it though and it could very likely need to be tested in each instance of an unusual program.  I would think the big barrier would be infrastructure there.  With windows stuff it just isn’t ready for prime time and windows history for making desperately needed improvements in something is abjectly awful. Their more or less refusal and foot dragging to create a decent multitasking operating system is more or less what created the multithreading environment we are in. The industry basically had to go around them because they refused to keep up.

Edited March 14, 2021 by Bombastinator

[Donut417]

2 minutes ago, Bombastinator said:

If all the programs happen to run on rosetta2 they’re good for mac.

Not necessarily. Apple doesnt plan to support Rosetta 2 forever. They are even discontinuing support in select markets coming up soon. Microsoft would be a required to support the comparability layer long term. Because some businesses are not going to make their software work on ARM, especially if its software they created internally for their use only. 

 

 

[Bombastinator]

1 minute ago, Donut417 said:

Not necessarily. Apple doesnt plan to support Rosetta 2 forever. They are even discontinuing support in select markets coming up soon. Microsoft would be a required to support the comparability layer long term. Because some businesses are not going to make their software work on ARM, especially if its software they created internally for their use only. 

 

 

Hmm.. didn’t know that.  I can see that as being a potentially massive problem.  Doesn’t strike me as a good idea.  If rosetta2 is going away in future OS updates me buying a Mac may not be a good idea.  I could be left with non functional software and a computer that while still technically works, in practice doesn’t for me.

[saltycaramel]

2 hours ago, Bombastinator said:

Your describing IOT?  I avoided “smart” stuff because it seems to have really severe security issues. 

No, I meant what if the speed of the CPU cores plateau or becomes a given and the new way of making new hardware better consists of adding custom modules at the chip design level to serve the everchanging needs of the OS and the software in a bespoke way. 

 

Of course to do this you need to have a vertical control of

- chip design

- OS 
- software (or have a fast reacting dev community that quickly adopts the APIs/frameworks/etc. you introduce)

 

Like Apple. 

[Bombastinator]

5 minutes ago, saltycaramel said:

No, I meant what if the speed of the CPU cores plateau or becomes a given and the new way of making new hardware better consists of adding custom modules at the chip design level to serve the everchanging needs of the OS and the software in a bespoke way. 

 

Of course to do this you need to have a vertical control of

- chip design

- OS 
- software (or have a fast reacting dev community that quickly adopts the APIs/frameworks/etc. you introduce)

 

Like Apple. 

Seems far fetched.  Speed of cpu cores plateaued or nearly plateaued some years ago.   There are still tricks to pull.  They were collecting them way back before multi-core turned out it worked.  You’re looking for a long chain of fairly unlikely things to happen.

[NineEyeRon]

I heard LEG are the next big thing, once they get a foothold in the market 

[Vishera]

X86_64 is a powerful instruction set,I don't see it going away anytime soon.

ARM is more useful for low power applications.

[Bombastinator]

4 minutes ago, Vishera said:

X86_64 is a powerful instruction set,I don't see it going away anytime soon.

ARM is more useful for low power applications.

Part of the problem with this is x86 has gotten more RISClike and ARM has gotten more x86like so there’s less of a gap than there appears.  Neither remains the same.  ARM has advantages in wait state power use.  That’s about it though.

Edited March 14, 2021 by Bombastinator

[Zodiark1593]

17 minutes ago, Bombastinator said:

Part of the problem with this is x86 has gotten more RISClike and ARM has gotten more x86like so there’s less of a gap than there appears.  Neither remains the same.  ARM has advantages in wait state power use.  That’s about it though.

Pretty much this. Wouldn’t call it a problem though so much as progression. 
 

The only meaningful differences now lie in vendor and software implementation. Apple has shown that ARM compatible cores are more than capable of delivering on compute intensive tasks. There’s absolutely no debate on that. 
 

Legacy software support is a pretty massive hurdle however. Having ARM cores of merely comparable performance to x86 cores (or even a moderate to solid performance advantage) of the same generation won’t cut it. For ARM compatible cores to begin to take desktop share, they need a pretty drastic performance advantage over a same-gen x86 core, so as to translate x86-64 code fast enough and put it on even footing. Only Apple has really come close. 

[straight_stewie]

On 3/12/2021 at 11:57 AM, Amias said:

ARM is just better in almost everyway.

What? Have you actually looked at real benchmark scores? The most common M1 chips have only like a quarter the single or multithreaded performance for most non synthetics that the big boys (i7's and threadrippers) have.

 

 

On 3/12/2021 at 11:36 AM, curiousmind34 said:

lagging behind a bit in the generation to generation performance uplift that apple's arm chips do

Just wait a bit on that one. The big limitation for x86/64 chips is particle physics. The big limitation for ARM is that no one has tried big, fast, ARM chips before.

Let's just see what happens with the benchmarks over the next decade. Apple's ARM chips have HUGE ground to cover to catch up to current x86/64 offerings.

[Bombastinator]

26 minutes ago, Zodiark1593 said:

Pretty much this. Wouldn’t call it a problem though so much as progression. 
 

The only meaningful differences now lie in vendor and software implementation. Apple has shown that ARM compatible cores are more than capable of delivering on compute intensive tasks. There’s absolutely no debate on that. 
 

Legacy software support is a pretty massive hurdle however. Having ARM cores of merely comparable performance to x86 cores (or even a moderate to solid performance advantage) of the same generation won’t cut it. For ARM compatible cores to begin to take desktop share, they need a pretty drastic performance advantage over a same-gen x86 core, so as to translate x86-64 code fast enough and put it on even footing. Only Apple has really come close. 

Close enough for me I suspect but I’m not a super demanding user. This whole “we’re only going to do Rosetta for a while though” thing implies that Apple is going to back off from that though. 

[Bombastinator]

4 minutes ago, straight_stewie said:

What? Have you actually looked at real benchmark scores? The most common M1 chips have only like a quarter the single or multithreaded performance for most non synthetics that the big boys (i7's and threadrippers) have.

 

 

Just wait a bit on that one. The big limitation for x86/64 chips is particle physics. The big limitation for ARM is that no one has tried big, fast, ARM chips before.

Let's just see what happens with the benchmarks over the next decade. Apple's ARM chips have HUGE ground to cover to catch up to current x86/64 offerings.

Re: just better

while I agree, your example compares a thin-and-light laptop CPU to an enterprise level CPU.  When compared thin-and-light to thin-and-light the m1 does really well.  Scalability may be a factor though.  Enterprise level RISC chips are being made though which implies it doesn’t have to be.

 

Re: too early to say

again I agree, but while full blockage bugbears may emerge, you attempt to predict how long the life expectancy of x86 is.  I would say that is unknown as well.

[whm1974]

41 minutes ago, Bombastinator said:

Re: just better

while I agree, your example compares a thin-and-light laptop CPU to an enterprise level CPU.  When compared thin-and-light to thin-and-light the m1 does really well.  Scalability may be a factor though.  Enterprise level RISC chips are being made though which implies it doesn’t have to be.

 

Re: too early to say

again I agree, but while full blockage bugbears may emerge, you attempt to predict how long the life expectancy of x86 is.  I would say that is unknown as well.

Well x86 lasted far longer then Intel expected. Not bad for a Stopgap design. Everyone in the Industry did accept the x86 will be replaced by both RISC and VLIW ISA designs. Even Intel.

 

AMD gave the IA-32 ISA a new Lifespan by extending it to 64 Bits.

[RoseLuck462]

Most likely 

[Vishera]

8 minutes ago, whm1974 said:

Not bad for a Stopgap design.

What are you talking about?

[Bombastinator]

20 minutes ago, Vishera said:

What are you talking about?

Might be referring to AMD64. 

[whm1974]

38 minutes ago, Vishera said:

What are you talking about?

https://en.wikipedia.org/wiki/Intel_8086

Quote

The 8086 project started in May 1976 and was originally intended as a temporary substitute for the ambitious and delayed iAPX 432 project. It was an attempt to draw attention from the less-delayed 16- and 32-bit processors of other manufacturers (such as Motorola, Zilog, and National Semiconductor) and at the same time to counter the threat from the Zilog Z80 (designed by former Intel employees), which became very successful. Both the architecture and the physical chip were therefore developed rather quickly by a small group of people, and using the same basic microarchitecture elements and physical implementation techniques as employed for the slightly older 8085 (and for which the 8086 also would function as a continuation).

See?, A Stopgap...