Build a PC while you still can

[GabenJr]

Arm CPUs are taking over. Apple Silicon showed us that desktop computers need not be power hogs - Why haven't AMD, Intel, and Nvidia done the same, and would you want it?

 

 

[Asbo Zapruder]

Page me in 5-10 years when the thumbnail for this has to add clown makeup.

[Unedited_Mind]

One thing people hate is change.

 

 

-Anyone with life experience and an ounce of a brain cell.

[doubleflower]

It really doesn't matter anyway simply because both ARM and x86 has evolved so much that it barely resembles their original implementations. 

 

https://chipsandcheese.com/2021/07/13/arm-or-x86-isa-doesnt-matter/

 

It's very possible to make efficient x86 chips or even hardware too (see the APUs on the new consoles for a good example, or the new Quadro GPUs), but the main problem is that people wanted performance at any cost so companies just cranked the power consumption to ridiculous levels just to out edge the competition by a small margin in terms of performance.

 

...and of course, there's always the large companies or even government organizations (including potentially mission critical systems) that still relies on decade-old IT systems so it's not very likely for ARM CPUs to completely takeover the x86 incumbents soon, and even so it's not like AMD/Intel are incapable of competing with them either (they have bountiful cash and resources for that)

 

If anything else the uprising of ARM will just motivate the incumbents to innovate and ultimately the consumers will win from the competition spawned.

 

My two cents

[Zodiark1593]

If the consoles (PS5 and Series X) had moved to a different ISA, I’d be pretty bearish on the near-intermediate-term future of x86. However, this currently isn’t the case. 
 

It will still be a pretty long time before we start seeing the first inklings of an earnest transition. 
 

Rosetta does offer the possibility that we don’t need to sacrifice compatibility to finally migrate off x86 (IE, we needn’t drag folk kicking and screaming). So it’s (x86’s) days are definitely numbered. Just a matter of getting everyone on board. 

[darknessblade]

As long as I can add as much RAM and Storage as I want, I do not really care for ARM or normal CPU's.

 

But pulling a Apple, and making it Impossible to upgrade is 1 step too far for me.

 

But then again I am the type to only buy a PC once every 6-8 Years, buying the "best of the best" at that moment. Upgrading storage and Ram if needed

[StephanTW]

I get the premise of this video, but i feel people would want to have options and might not want to overpay for a die that has a great cpu on it that they need but a way overprized and oversized gpu. so a 1 die cpu/gpu like apple might not be liked by the consumers, like me personally would not easily swap to a 1 die design.
on top of that there is the potential for either the cpu or gpu part to die and needing replacing, with the proposed single die design it would mean spending alot more due to basicly buying a gpu and cpu at once instead of 1 of the 2

[Zodiark1593]

1 minute ago, darknessblade said:

As long as I can add as much RAM and Storage as I want, I do not really care for ARM or normal CPU's.

 

But pulling a Apple, and making it Impossible to upgrade is 1 step too far for me.

 

But then again I am the type to only buy a PC once every 6-8 Years, buying the "best of the best" at that moment. Upgrading storage and Ram if needed

One thing that was neglected to be mentioned was that, just because the ISA is different, doesn’t necessarily mean that modularity is lost. ARM chips can be equipped with PCIe that can be used for both built-in communications (iPhones actually do this to access storage), there’s no reason why you couldn’t direct those lanes to a standard slot. 

 

Same thing with memory to a degree, though the high speed LPDDR variants (such as LPDDR4X) apparently need closer and more solid connections than can be done via slot. Still, nothing that widening the memory interface can’t overcome. 

[Sauron]

The death of desktop computers has been announced multiple times in the last few decades and yet here they still are. It's been a bit of a niche since the early 2000s but a strong niche nonetheless. It's possible we'll see parts be more integrated as we've seen with math coprocessors but I don't think the desktop is anywhere near dead yet.

[Blademaster91]

The desktop PC was supposed to be in decline with laptops and phones replacing desktops, except desktop PC's are still around, and the DIY desktop PC market has gotten more popular.

I doubt the X86 ISA is going anywhere anytime soon, not while servers and IT still use x86, and AMD and Intel have a lot invested in x86. Also the claim that desktops are wasteful is interesting, an x86 desktop PC is a modular platform, I can replace a failed ram stick or a power supply, or upgrade ram, SSD,and GPU. But with an integrated SOC that has non-replaceable RAM and SSD, you'll have to replace the whole system if anything breaks.

[Nystemy]

To be fair, this video is a bit inept.

 

Yes, Apple's SoC is "good", ie having about the same power efficiency as contemporary processors when benchmarked. And usually at a much lower performance.

 

Power efficiency is however not always the most critical specification in a system. Sometimes other factors like memory capacity, peak performance, expandability (IO), and reliability/reparability is more worthwhile features. And it is here that desktop computers with their more modular approach has major benefits.

 

Also, building an SoC isn't magic either. To reach higher performance we will need a larger chip to fit all the different parts. This has a couple of major downsides.

1. A larger chip has lower yield in manufacturing.
2. A larger chip has a larger failure rate out in the field due to thermally induced stress between the substrate and flip chip. _{(ie, the balls crack as a result of fatigue from thermal cycling.)}
3. A larger chip with the same W/mm^2 is harder to cool.

And going with a more multi chip approach has advantages, but interconnects rapidly suffer in performance. And here the value add of flexibility is often times worth more to the end user than an integrated solution.

 

However, yes, the "console" PC form factor is likely to become a bit more prevalent over time, since just like so many other things. All applications do have some upper limit where the application will eventually reach diminishing returns and more than X amount of performance won't be required for Y application. _{("application" here being a type of workload, not a specific program. Think video conferencing, or media playback, or editing a text document, etc. Sound is a good example of an application that just don't really need much more performance today than it did 20 years ago.)}

 

Gaming however haven't shown any signs of hitting said upper threshold where we start reaching into diminishing returns. Since graphics can still get a lot better, physics too, as well as NPC path finding and decision making, among a slew of other things.

 

And a fair few other applications haven't reached that upper practical limit yet for that matter. Be it relatively normal tasks like video editing, photo editing, 3d modeling/animating, etc. _{(Could drag in other more compute heavy applications, but doubts most people do such.)}

 

But in the end.

I do not think the desktop computer will disappear in the next century.

However, X86 probably will become less common over time. If it completely disappears in the next 50 years is a different debate.

ARM however isn't the only contender in the ring, there is PowerPC, RISC-V. _{(this I doubt will go anywhere due to fragmentation, something ARM already suffers)} But there is also other architectures on the rise, however few that are currently out in the field. _{(If things go my way, then my own architecture is on the shelf by the end of the decade.)}

[Uttamattamakin]

I 1000% agree with this as I have been saying for a while APU's are going to soon be so good that for 90 -95% of people they will give good enough video performance for any game that now exist.  Wait 10-15 years from now and an APU will be all of it. 

As for a PC just being an SOC.  That'll take a while.  I think it will be possible to buy an Intel, or AMD X86 SOC , or Nvidia ARM or Risk V SOC with CPU GPU RAM and storage on it and socket it in.  Then ADD to it more ram, more storage and more of a GPU if you like. 

 

As for the state where EVERYTHING is just an SOC.  I can't see that being a thing.  There will ALWAYS be a need to add more capacity.  

 

I 1000% disagree that ARM will take over.  I'd believe that more efficient x86 designs will become a thing.  Remember that X86 was once represented by the HOT power consumptive Net Burst arcitecture.  Then a series of architectual revolutions lead us to cooler, faster CPUs with better IPC. 

 

[Comp_J]

I think many of you are way too optimistic about the continuing existence of desktop PCs in their current form. Hardware manufacturers have consistently shown a pattern of taking control away from consumers via a 'we know best for you' approach, as is especially evident with the smartphone industry and Apple. Any loss of customizability or repairability is arguably a good thing FOR MANUFACTURERS, as it forces consumers to spend more and more often. The increased profits from these policies and cut costs from eliminating unnecessary middlemen (i.e. separate GPU or motherboard manufacturers) will vastly overshadow any slight decreases in yield from the larger sizes of the SoCs.

 

As much as we all hate it, the average layperson won't ever be able to understand the nuances of 'losing customizability and repairability', as their ape brains are wooed by the 'lightest, thinnest, shiniest' new device. We are the last of a soon-to-be dying breed of PC hardware enthusiasts. 

 

 

[Ultraforce]

26 minutes ago, Nystemy said:

To be fair, this video is a bit inept.

 

Yes, Apple's SoC is "good", ie having about the same power efficiency as contemporary processors when benchmarked. And usually at a much lower performance.

 

Power efficiency is however not always the most critical specification in a system. Sometimes other factors like memory capacity, peak performance, expandability (IO), and reliability/reparability is more worthwhile features. And it is here that desktop computers with their more modular approach has major benefits.

 

Also, building an SoC isn't magic either. To reach higher performance we will need a larger chip to fit all the different parts. This has a couple of major downsides.

1. A larger chip has lower yield in manufacturing.
2. A larger chip has a larger failure rate out in the field due to thermally induced stress between the substrate and flip chip. _{(ie, the balls crack as a result of fatigue from thermal cycling.)}
3. A larger chip with the same W/mm^2 is harder to cool.

And going with a more multi chip approach has advantages, but interconnects rapidly suffer in performance. And here the value add of flexibility is often times worth more to the end user than an integrated solution.

 

However, yes, the "console" PC form factor is likely to become a bit more prevalent over time, since just like so many other things. All applications do have some upper limit where the application will eventually reach diminishing returns and more than X amount of performance won't be required for Y application. _{("application" here being a type of workload, not a specific program. Think video conferencing, or media playback, or editing a text document, etc. Sound is a good example of an application that just don't really need much more performance today than it did 20 years ago.)}

 

Gaming however haven't shown any signs of hitting said upper threshold where we start reaching into diminishing returns. Since graphics can still get a lot better, physics too, as well as NPC path finding and decision making, among a slew of other things.

 

And a fair few other applications haven't reached that upper practical limit yet for that matter. Be it relatively normal tasks like video editing, photo editing, 3d modeling/animating, etc. _{(Could drag in other more compute heavy applications, but doubts most people do such.)}

 

But in the end.

I do not think the desktop computer will disappear in the next century.

However, X86 probably will become less common over time. If it completely disappears in the next 50 years is a different debate.

ARM however isn't the only contender in the ring, there is PowerPC, RISC-V. _{(this I doubt will go anywhere due to fragmentation, something ARM already suffers)} But there is also other architectures on the rise, however few that are currently out in the field. _{(If things go my way, then my own architecture is on the shelf by the end of the decade.)}

Yeah I don’t know if it needed more revisions or maybe just needed someone else to take a look at it but I feel like my hardware prof in third year and 4th year made a better argument and helped explain why x86 should die.

 

Also yeah form factor type discussions and upgradability feels like a separate video that could have been made.

[mini-z1994]

If they can truly become their own ecosystem first while being compatible in certain ways natively with Windows, Linux & phones before chasing the goal of becoming desktop replacements sure i can see it taking a spot in the market that's pretty solid.
Because otherwise that's at best a fever dream.

With what we have now there are so many solutions too make it convenient using your phone or current computer too consume content or work with or on.
I honestly don't see it as anything more then an alternative solution too an already solved problem.

Like a Bluetooth keyboard & mouse, chrome cast or similar device capable of sharing your phone display & boom you got a streaming or web browsing machine for your tv or screen that pulls a low amount of power vs a pc or dedicated arm based hardware on the side.
The price of those things might be like 80$ - 100 in total vs buying something like a dedicated pc for it.
Sure my phone isn't going too do 4k being a Samsung galaxy a12 nacho.
But my tv that was a cheap model from like 2012 i got for free isn't going too do 4k either.

Even as a game console similar too nvidia shield or the nintendo switch that was mentioned is more realistic I'd imagine if there is proper game support.
Together with arcade machines being purpose built with arm chips in mind maybe that emulates older arcade games well with minimal input lag I'd see that as a potential option.

Like i can see something like an internet cafe running specifically made machines based on ARM with headsets & gamepads running a version of android or Linux with a chip that pulls maybe 15w put behind the 23 inch 1080p ips display in the booth there for people running like genshin impact or a mobile version of like pubg, cod or whatever.

They might have a better time developing something that sells even if the market might be niche too start with, like maybe they can aim at making cheap machines for web development & store displays which plenty of times for store displays it's literally an old raspberry pi hooked up too the tv's display input running a photo slideshow 24/7, we've all seen them for years & thats what ? 50$ of hardware for the tv ?
ARM based arcade machines, NAS storage with an m.2 drive as additional cache for the sata ssd's or mechanical harddrives too stream 4 - 8k content i can see maybe for now.

[Sauron]

1 hour ago, Nystemy said:

the balls crack

hate when that happens

[NomBread]

Anthony I have to disagree with this nobody in the pc space will want to buy a SoC box that can't be upgraded or when it breaks have to go out and buy a new one.

 

ThatGuyPete in the youtube comments section has a point * It's funny that LMG has been pushing consumer rights and then the moment Apple makes a half-decent product, we get a video telling us about how it's "the future." Genuinely this video should've gone back to the drawing board. Telling enthusiasts—hell even consumers—that their PCs are dead if even a single component is shot and that it's a "good thing" was the absolute wrong way to go. *

[n0stalghia]

Anthony is absolutely on point with this, but there is another vector of development that will also start to come to consumers - PCs as "terminals". Small, ultra-lightweight devices that are nothing more than a screen and a processor with minimal other hardware, and all computing will be done in the cloud. When quantum computing becomes broader - and it will, right now QC is following Moore's law on increasing its performance every couple years - we'll have to access QCs for our compute tasks, and we'll be able to offload stuff to them. Cellular and connections in general will be fast enough by that point to make the experience seamless.

[n0stalghia]

14 minutes ago, NomBread said:

Anthony I have to disagree with this nobody in the pc space will want to buy a SoC box that can't be upgraded or when it breaks have to go out and buy a new one.

Well then those enthusiasts will have to fab their own chips, as there sure as hell won't be any x86 being made when the manufacturers make the jump... The enthusiast market is a tiny portion of the PC market which has been stagnating/shrinking for a decade.

[hishnash]

13 minutes ago, NomBread said:

Anthony I have to disagree with this nobody in the pc space will want to buy a SoC box that can't be upgraded or when it breaks have to go out and buy a new one.

The vast majority of PCs purchased by people are never upgraded and it is very important for the right to repair movement to not conflate upgradability with repairability.

 fully upgradable system does not mean a repairable system! These are very different things, you are not repairing your SSD when you take the NVMe drive and throw it way (along with a perfectly working SSD controller and dram package). A reparable system is a system were you can buy board level parts to replace and get the software to be able to reset the firmware on the many little controllers. In this aspect apple are in fact ahead of the industry since they provide anyone who want them the software to fully re-flash all the firmware on the modern Macs (that includes your SSD controller). so the SSD is more repairable on a Mac than on a windows machine were your really going to struggle to get the software to reset the NVMe controller and your also going to find it hard to apply that to the controller as its unlikely to just do this flashing over a USB-C port. What apple are missing is the providing of board level parts and schematics that are just as critical to repair. 

[Dawid11]

I want to say that Anthony is partly wrong in case that Nvidia isn't offernig any ARM SoC for costumers, since they have whole lineup of products called Nvidia Jetson. Yes of course their main focus is AI and robotics, but they can be used as normal computers. Their newest entry even have hefty specs like 12 core cortex A78AE CPU integrated with 2048 core ampere GPU with 32GB of LPDDR5x memory and also it all comes with hefty price tag of $2000 its called Nvidia Jetson Orin AGX developer kit.

https://www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-orin/

[hishnash]

1 hour ago, Uttamattamakin said:

I'd believe that more efficient x86 designs will become a thing.

The massive bloat of the x86 instruction set means if they want to continue to have hardware decoding of this thing then it will be very hard for them to take over.  It is much much harder to build a slow and wide x86 cpu core (that decodes and executes many instructions at once) than it is to do the same for a fixed width instruction set like ARM.  If you can double your width you can 1/2 your clock speed and power draw is very non linear with clock speed so 1/2 your clock speed means 1/4 or even less the power!  Vendors like intol/amd could build x86 hybrid cpus were os etc uses ARM and legacy apps use the x86 decoder pathways but this would still require all those transistors for those x86 apps very quickly it just becomes better to do this translation in software (like apple this does not need to be done as an interpreter you can do static recompilation that is cached on disk). 

[hishnash]

2 hours ago, Nystemy said:

• A larger chip has lower yield in manufacturing.

Being able to target a much lower frequencies do to having much better IPC means apple might well end up with higher yields than you think.

2 hours ago, Nystemy said:

1. A larger chip with the same W/mm^2 is harder to cool.

Yep however by being integrated the W/mm^2 is lower as you save a lot on the power draw for IO (this is a large power draw in some systems getting close to that of the compute). 

 

 

2 hours ago, Zodiark1593 said:

Same thing with memory to a degree, though the high speed LPDDR variants (such as LPDDR4X) apparently need closer and more solid connections than can be done via slot. Still, nothing that widening the memory interface can’t overcome. 

The reason apple are using on package soldered LPDDR is power draw for the bandwidth. To get the bandwidth they need with socketed memory they would need many many sockets of DDR not only would this take up a LOT of space and power (more than the total volume of the systems they are selling in some cases). 
 

[hishnash]

2 hours ago, darknessblade said:

But pulling a Apple, and making it Impossible to upgrade is 1 step too far for me.

 

If you have a high perfomance SoC your not going to be able to upgrade the memory attached to it since like your gpu that SOC will need a LOT of bandwidth for its memory and to get that backwdith with socketed memory you would be talking 16 or more memory dims! not only would your upgrade of this memory cost a LOT but also this would draw more power than the rest of the system combined. There is a reason GPUs do not used socketed memory and that reason applies to SOCs with good integrated GPUs. 

Some vendors (I expect apple will do this for the macPro) might go with a 2 tier system memory were the SOC gpu/npu/cpu shared pool is soldered high bandwidth and there is an additional expansion region of memory that is just for lower bandwidth (cpu) address spaces that you can expand. But I would not expect many systems to offer this option out side of the workstation market. 

[Sauron]

23 minutes ago, n0stalghia said:

Anthony is absolutely on point with this, but there is another vector of development that will also start to come to consumers - PCs as "terminals". Small, ultra-lightweight devices that are nothing more than a screen and a processor with minimal other hardware, and all computing will be done in the cloud. When quantum computing becomes broader - and it will, right now QC is following Moore's law on increasing its performance every couple years - we'll have to access QCs for our compute tasks, and we'll be able to offload stuff to them. Cellular and connections in general will be fast enough by that point to make the experience seamless.

This has been a trend for years, it really has nothing to do with Apple's new hardware. It's actually starting to show its limitations with speculative execution vulnerabilities and internet speeds that are just not catching up fast enough in large parts of the world. It still doesn't really do anything to replace what a diy desktop offers.