• Please review our updated Terms and Rules here

What do you think of Nvidia's ARM-based PC idea?

C

carangil

Experienced Member
Joined
Jun 3, 2009
Messages
285
Location
Oakland, CA
I just read this:

http://www.computerworld.com/s/article/9203538/Nvidia_unveils_Denver_its_first_CPU_for_PCs

Nvidia is developing an ARM cpu for desktop PCs, and Microsoft is making a ARM version of Windows to match. When Apple switched to x86, I thought that was weird. I like x86, but I thought the loss of PowerPC Macs really started to limit how interesting consumer hardware is. Now its either its x86 or it's Apple-branded x86. I seem to see more x86 stuff running Solaris than actual sparc hardware.

I don't think all manufactures would jump to ARM, but if Windows Arm catches on, that would be pretty cool. I just think having two architectures to choose from for desktop PCs would make things more interesting. (And as a software engineer, it gives me more work to do, haha.)
 
I'm all for this. Don't know how ARM Windows will turn out, but there's always Linux, and I've really been hoping the PC market will diversify in the near future - not only is it boring having only x86-based computers anywhere near being in the mainstream, the whole Intel PC architecture is definitely showing its age. It's high time for things to branch out in new directions again.
 
Well, ARM is used for netbooks and mobiles, so I suppose it's possible that it might succeed.

But this isn't the first time a RISC has been deployed as a PC platform (remember the IBM Power PC and the RS6000?).

Quite frankly, if we're starting from scratch, it'd be nice to see a new CPU architecture, not one that's nearly 30 years old.
 
I expect there to be some kind of portable computer using a real Windows OS that is bigger then a cell phone but doesn't look like a notebook/netbook and needs a very low powered CPU for web/video use. So maybe ARM will fill in that product slot.

X86 is here to stay for desktop use, there is just too much behind it for anything else to take over. Either we evolve away from a powerful desktop/laptop centric computer to another CPU type or we are stuck with generations of x86 clones.
 
carangil said:
Nvidia is developing an ARM cpu for desktop PCs, and Microsoft is making a ARM version of Windows to match.
Click to expand...

There is a growing market of (chinese) ARM-based Windows CE netbooks, and I think the move is focused on this market. The talk about "desktop PCs" is probably meant to bring desktop software makers to develop for ARM, in order for the netbooks to have the same software options as desktop Windows (that means less pain and trouble for users switching from PC to netbook). But that's just my guessing :).
 
carangil said:
I don't think all manufactures would jump to ARM, but if Windows Arm catches on, that would be pretty cool. I just think having two architectures to choose from for desktop PCs would make things more interesting. (And as a software engineer, it gives me more work to do, haha.)
Click to expand...
Well, I'm bored with the current unification of general-purpose computers too, but I'm definitely not an average PC-user.
Average PC-users are the people who don't know and don't care about different architectures, they just want all the software run on their hardware, and want everything to be as cheap as possible.

I expect those desktop ARM PCs to have lower performance, and lower performance/price ratio than those with x86, and unable to run most common off-the-shelf software (or, if they include software x86 emulation, running them at sluggish speed). Why would anybody want such a thing???

There were attempts at bringing new CPU architectures to PC - in the mid-90s IBM poured a lot of money into their PowerPC&OS/2 project, and later Intel seemed to think about replacing x86 with ia64. All in vain.
What does Nvidia think ARM has that PowerPC didn't have?
 
Marrr said:
What does Nvidia think ARM has that PowerPC didn't have?
Click to expand...

Microsoft's blessing, it would seem!

I'd be interested in a non-x86 desktop, though it would probably run Linux. In the past, I've used several non-x86 workstations with Linux and was pretty happy with performance (a Sun Ultra 30 and SGI MIPS based workstation). If ARM really gets pushed for the desktop market and not just the netbook market, it wouldn't surprise me to see hardware x86 emulation add-on boards come back to some degree. Of course, with Microsoft working to release a full version of Windows for ARM, support for an x86 coprocessor might be integrated seamlessly into Windows.
 
glitch said:
Microsoft's blessing, it would seem!
Click to expand...
This is not going to mean much if x86 also has that blessing...
I'd be interested in a non-x86 desktop, though it would probably run Linux. In the past, I've used several non-x86 workstations with Linux and was pretty happy with performance (a Sun Ultra 30 and SGI MIPS based workstation).
Click to expand...
True, that's the power of open-source: just re-compile and it works at full speed no matter what hardware you have.
Unfortunately, common software for average users is still closed-source.
If ARM really gets pushed for the desktop market and not just the netbook market, it wouldn't surprise me to see hardware x86 emulation add-on boards come back to some degree. Of course, with Microsoft working to release a full version of Windows for ARM, support for an x86 coprocessor might be integrated seamlessly into Windows.
Click to expand...
The problem is: such add-ons are sure to cost money.

BTW: ARM already was in the desktop market - anybody remember Acorn Archimedes? I reckon they were popular only in the UK.

Let's face it: x86 has already smashed all the competition in the desktop/laptop market, most of the competition in small-to-medium servers, and still seems to conquer new areas - currently mostly netbooks and similar small systems.

I can see only two chances to break this tendency:
1. Some completely new technology which would immediately produce chips by at least one order of magnitude faster than those already on the market, so that existing software would work fast enough even under software emulation. Maybe photonic computing?
2. The end of desktop computing as the mainstream. In the foreseeable future even the most simple pocket devices can have enough processing power and factory-installed software to fulfill all the needs of an average user. And if there's no need to install any additional software, why worry about incompatibility?
 
Marrr said:
BTW: ARM already was in the desktop market - anybody remember Acorn Archimedes? I reckon they were popular only in the UK.
Click to expand...

That was my reference to "nearly 30 year old technology"--the Acorn came out, what, 1983 or so? ARM is a nice simple design, but it's not the x86-slayer that the article thinks it is. And there are probably at least twice as many ARM-based applications out in the world as x86 ones, since the ARM finds its way into most mobile devices and lots of other things like routers, GPS and tons of embedded applications.


1. Some completely new technology which would immediately produce chips by at least one order of magnitude faster than those already on the market, so that existing software would work fast enough even under software emulation. Maybe photonic computing?
Click to expand...

A new architecture wouldn't hurt one bit. But not ARM.

2. The end of desktop computing as the mainstream. In the foreseeable future even the most simple pocket devices can have enough processing power and factory-installed software to fulfill all the needs of an average user. And if there's no need to install any additional software, why worry about incompatibility?
Click to expand...

Not going to happen at least in the corporate world. Control over access and data is important and that's hard to do wihen mobile devices (heck, even laptops) are involved.
 
I will have to see what the final ARM systems look like and compare it to what Intel and the rest are shipping then. Hopefully, Intel will finally spend efforts to really cut the power consumption of their chipsets. But Nvidia hasn't completed the supporting chipsets to go with the ARM CPU. Or, like current ARM netbooks, most everything will be handled by external USB devices which are less efficient than the internally mounted drives. The resultant ARM system might well have very low power consumption on the CPU when playing Blue-Ray discs but the external Blue-Ray drive could waste more power than the ARM CPU saves.
 
Since ARM is much simpler and cleaner than x86, a good engineering team can probably optimize an ARM chip better than x86. x86 architecture is ridiculous in its complexity, mostly stemming from having to keep compatibility with all previous mistakes. Still even with all its ugliness, look how fast x86 can run today. I'm not sure ARM is the best architecture to make PCs out of, but I welcome any change in the industry.

What would be really interesting (and weird) is if PowerPC makes a comeback and we get PPC Windows. Then it would just be as if Mac and PC switched places.
 
carangil said:
Since ARM is much simpler and cleaner than x86, a good engineering team can probably optimize an ARM chip better than x86. x86 architecture is ridiculous in its complexity, mostly stemming from having to keep compatibility with all previous mistakes. Still even with all its ugliness, look how fast x86 can run today. I'm not sure ARM is the best architecture to make PCs out of, but I welcome any change in the industry.

What would be really interesting (and weird) is if PowerPC makes a comeback and we get PPC Windows. Then it would just be as if Mac and PC switched places.
Click to expand...
My HP PDA 4700 (at least 5 years old) has an ARM processor and runs a form of embeded Windows. My net book has a dual core Atom processor and all that little box is good for is surfing the net. It does a fairly good job of doing that and is pretty doggy for anything else - it fills a niche. I think those type of processors will be dedicated to certain tasks and will never be in the mainstream as far as high end data processing or gaming. I may be wrong but time will tell.
 
Various RISC architectures also had Microsoft's blessing in days of yore. Windows NT 3.1, 3.51, and 4.0 ran on various RISC platforms (MIPS, Alpha, and PowerPC). Itanium also had Microsoft's blessing, but it ultimately lost out to AMD64 (now known as x86-64 or even, ironically, Intel64).

I ran NT4 on Alpha and liked it, but ran into two problems. Driver support wasn't nearly as good under the RISC architectures, and software was a problem too. Now, there's a lot more open-source software out there that you can recompile and run, but ARM-compiled web browsers aren't going to get very far if Adobe doesn't release a compatible Flash plugin. And there are always going to be people who are going to be hopping mad that they can't run software they bought out of the $5 bargain bin at the local big box store.

For that matter, I used to get at least one support call per week from people with Macintoshes who couldn't run the joke programs that PC users were sending them via e-mail. One user actually complained to my managers that I was refusing to support her, because I was unable to fix this "problem." So you have to be careful what kind of person you give an alternative machine to. Most Mac users famously see the glass as half full and treat their computer like a religion, but bad, bad things happen when you give one to a glass-half-empty type of person. Windows running on ARM will run into the same problem.

I'm interested in something with a reasonably fast ARM CPU in it and the ability to connect it to a proper keyboard and display. But I'd run Linux on it and use it as a quiet, low-wattage server. That's not something that the average user is going to do.

I think it might stand a chance IF the right software is available for it and if there's enough hardware support for it and if the price is significantly lower. But Windows on ARM faces the same challenges that Linux on x86 faces. It still has the perception of not having good hardware support and not having any interesting software to run on it (which really means no native version of Microsoft Office and Adobe Creative Suite).

If the price difference between an ARM-based PC and something similar based on an AMD Fusion is small, people will go with the AMD Fusion just out of convenience. Just in case they ever want to run something on it that only runs on x86.
 
If Nvidia upgrades their ARM core design with enough cache, a x86 software emulator could potentially run at about 80% of the ARM's core speed. With current best ARM chip speeds, that could leave emulated x86 code running at close to the same speed as code running on Atom or Bobcat based Fusion systems. The higher end AMD and Intel CPUs would outperform any likely ARM chip but should cost much more. Though a big question is whether Nvidia can release a very thrifty and cool running GPU system; their previous GPUs were lacking in that aspect.
 
Dave Farquhar said:
Various RISC architectures also had Microsoft's blessing in days of yore. Windows NT 3.1, 3.51, and 4.0 ran on various RISC platforms (MIPS, Alpha, and PowerPC). Itanium also had Microsoft's blessing, but it ultimately lost out to AMD64 (now known as x86-64 or even, ironically, Intel64).
Click to expand...

i860 even had Microsoft's blessing. I don't know about i432, but it wouldn't surprise me. Basically, BillG would bless anything that Intel came out with.

On the other hand, M$ demonstrated a version of NT running on ARM at the 2011 CES, so I suppose that they're serious.
 
f Nvidia upgrades their ARM core design with enough cache, a x86 software emulator could potentially run at about 80% of the ARM's core speed. With current best ARM chip speeds, that could leave emulated x86 code running at close to the same speed as code running on Atom or Bobcat based Fusion systems.
Click to expand...

Actually modern x86 internally translates the instructions into micro ops and then schedules and perform those ops (sometimes out of order) on a simpler risc core. An x86 CPU is more like a hardware accelerated JIT x86 emulator running on a risc-like chip. I don't see why you couldn't create a CPU that could decode both ARM and x86 instructions into the same micro-op scheduler. At first it seems far-fetched you could run two completely different instruction sets in the same chip, but if you think about it, x86 already works that way: You have 16, 32 and 64 bit modes, each with different addressing modes, register count, and even register size. Its already 3 archs in one, why not just pile another one on top! (Just wait, x128 will be here before you know it!)
 
Hi
Nvidia sees the end of life coming and must do something.
You also have to realize that for the most part, we've been running
RISC processor in all of our machines for some time. It just isn't
seen as a RISC processor to the software.
Today, the biggest bottleneck to speed is main memory. CISC
processors at that level have an advantage. More compact instructions
in main memory.
If Nvidia is to make it work, they really need to think about instruction
and data flow compaction at the hardware level.
If they don't do this, they will be missing the boat.
Both Intel and AMD realize that the separate GPU is a thing of the
past. This has been the main market for Nvidia. The first nail is
in the coffin. The have to get out of the hole quickly before it is
too late. They also need to be working on software translators,
not just emulators.
Dwight
 
Dwight Elvey said:
128 memory pointers.
Dwight
Click to expand...

Huh? :huh: Are you perhaps referring to 128-bit memory address space? Let's see, that's about 10**42, about enough to number all of the stars in the universe (best guess about 10**23) twice over, not quite enough to number all of the atoms on Earth (about 10**50). 256-bit double-precision would be about 10**85, sufficient to grid off between 16 and 32 copies of the known universe into cubic meter blocks and assign a number for each block.
 
You must log in or register to reply here.
Back
Top