RISC or CISC

[ManuelGomes]

OK, I think I'll fire up a bloody discussion.
Back in the old days, workstations were mostly RISC machines.
Who doesn't remember the good old Alpha, MIPS, Sparc...
And those beautiful SGI machines?
And the predecessor to the MacPro (NeXT computer) was about to be a RISC machine too. NeXT Step would become OS X!!
Personally I believe x86 should have been long gone now, as much as it's efficient and fast, but it's old tech and there should be an alternative (good one that is) already.
MIPS seems to be making a comeback in the hands of Imagination, but not for this kind of application.
Intel has (not for long it seems) the Itanium 9500 (Poulson) which seemed a great opportunity to kill off x86 but I guess higher values were more important. I believe only HP still uses these, and maybe not for long.
The Chinese seem to be very active developing RISC processors, too bad the big players don't follow the lead on this one.
Is anyone going to mention ARM? Still too far off.
I guess we'll be stuck with x86 for a long time still.
Any thoughts?

 

[netkas]

x86 is risc inside, there is decoder that decodes x86 cisc opcodes into series of internal risc opcodes. At least for intel it's true.

 

[mcnallym]

CISC / RISC is irrelevant to most people.

What they care about is what software the machine runs.

http://sunnyeves.blogspot.co.uk/2009/07/intel-x86-processors-cisc-or-risc-or.html

Is well worth a read as shows back then that the line between the two was blurring. Intel have adapted the x86 line to use RISC like features and RISC processors have adapted CISC like features as the Instruction set grew.

Intels Core series are even using a RISC core inside them which is hidden and you have to use CISC instruction set that is then broken down and executed on this RISC core.

If something is to replace x86 and x86-64 instruction set then the supplier will have to find way to provide backward compatibility with the software, and buy-in from the software developers.

Instead what is most likely to occur is that the x86 will develop further extensions etc and remain dominant in the mainstream market. Intel already had a go at replacing x86 with the Itanium range however AMD brought out x86-64 that meant could improve performance but didn't have to get the Software Developer to recode for the new instruction set. Code has then moved gradually from x86 to the x86-64 and Intel started adding Itanium features into some of the Higher End Xeon processors and Itanium hasn't replaced masses of x86 processors.

There will still be a place for other Architectures, i.e. PowerPC still found a lot as embedded controllers, ARM will continue to grow in power however whether Apple will move towards it for there lower end machines will depend on how well runs software that users want.

I want my Mac as a general Internet, iTunes and PVR system. If it can do this with an ARM or A series CPU or some other Architecture altogether in it I don't care, as long as it does what I need it too.

With things like phones, tablets etc where creating new Apps then easy to bring in a new CPU style initially as doesn't have to do as much as mainstream desktop/laptop does.

Look at how the ATV moved from Intel and an Nvidia GPU to the ARM based A series processors. Easy to do as didn't have to port much in terms of Applications etc.

I think still a long way off from x86 instruction set systems being replaced simply through software compatibility.

 

[kwikdeth]

dont forget PowerPC which was also a RISC architecture!

 

[ManuelGomes]

Yes, AVX is already in that direction.
But this is exactly what Intel has been doing over the years, for the sake of compatibility, adding stuff on top of an old architecture.
This will never end.
And Apple is known to have the b***s in place so as to cut with the old radically and embrace new tech for the sake of innovation, without concern for legacy compatibility - this is what I like most about Apple.
Of course people will always complain when changes happen but if this is not so we'll always be in the dark ages.
Itanium was in fact my hope of an Intel comeback with a fresh tech but it failed unfortunately.
PowerPC was once inside Macs but performance and heat issues required a change I guess.

 

[AidenShaw]

PowerPC was once inside Macs but performance and heat issues required a change I guess.

Does this not say that the RISC instruction set on the PowerPC was a failure against the CISC instruction set on x64?

As others have already said, current Intel chips have a very RISC-like core (lower case "c", since the design came with the P6, long predating the "Core" series of processor) as the basic architecture. There is a layer that dynamically and efficiently translates the x64 CISC instructions into the RISC-like µops that the processor actually executes. This makes the whole RISC vs. CISC debate pointless (as if the mass failure of pure RISC systems hasn't already).

In addition, since Intel x64 chips present a virtual instruction set as the exposed interface, new chips can (and have) change the µarchitecture at will - without causing any incompatibilities with the published x64 instruction set.

"CISC vs RISC" is a last millennium debate - the world has moved on.

 

[neomorpheus]

I miss those sweet Alpha CPUs....

 

[Ph.D.]

Who doesn't remember the good old Alpha, MIPS, Sparc...
Any thoughts?

Ask anyone under 30 and the answer is... no-one remembers or knows about anything other than an iPhone. It's irritating.

My first workstation was a MicroVax running BSD Unix. As configured with its $15,000 color monitor and dual Winchesters, it cost $60,000. Now that, at least in terms of its instruction set, is CISC!

To cover my bases, my next workstation was a SPARCstation 1. RISC. I think its monitor only cost half as much. ;-)

 

[monokakata]

My own pathway was PDP-11, VAX, Alpha.

Just a couple of points:

I think I read that the Chinese supercomputer that carried off the crown a few years ago was running a Chinese version of the Alpha architecture.

And I'm sure I remember that DEC and Intel had some kind of patent fight and in the end, Intel paid DEC for a licence to use some aspects of the Alpha architecture. This would have been in the 90s, I think.

Finally, who remembers that RISC was commonly said to actually stand for "Really Important Stuff in Compiler."

 

[Ph.D.]

No. But "CP/M" originally stood for "Control Program / Monitor" not "Microcomputer". Also, "get off my lawn!" ;-)

 

[ManuelGomes]

Yeah, too bad DEC (Digital) couldn't keep it going. It was great, at the time.
The Chinese (military) are very active with Alpha-like proc development for super computers.
The Chinese seem to like RISC a lot, they have also MIPS based processors for common devices.

 

[theluggage]

PowerPC was once inside Macs but performance and heat issues required a change I guess.

The only reason the PowerPC had "performance and heat issues" was that Apple were the only customer who wanted to put PPC in a laptop, so the makers of PPC chips (IBM & Motorola) had no real incentive to develop a fast, low-power version of the G5 PPC.

The problem with RISC is very simple: the huge corporate IT sector that is completely dependent on Windows and hence x86 code (probably pre-Windows NT 16 or 13 bit code, at that). Its no good building a super-fast, super-low-power RISC superchip if your biggest customer base then has to depend on software emulation of x86 to run their code, defeating any performance gains. The scientific workstation market got away with it because they tend to run Unix/Linux which has an ingrained culture of supporting multiple architectures and shipping application software as architecture-independent source code.

One killer may have been that Microsoft, at one time, had Alpha, MIPS and PPC versions of Windows NT - but then dropped them - probably because too much of their customer base relied on x86 code.

The x86 market is so huge, with such economies of scale, that it could keep up with RISC by throwing money and technology at the problem (including - as several posters have pointed out - incorporating RISC techniques into the x86).

Even Intel couldn't compete with x86 when they launched Itanium, and had to suffer the ignominy of adopting AMDs 64-bit x86 bolt-on instruction set.

Likewise, ARM was originally designed for fast desktop machines (and the ARM 2/3 beat the stuffing out of the contemporary 286) but there with no market for non-DOS/Windows machines they wisely decided to concentrate on mobile and embedded devices.

So, the problem is, and always has been DOS/Windows - even for Microsoft as they try and wean corporates off Windows XP (or even earlier).

What should be happening around now is the death of shipping software as architecture-dependent binaries in favour of virtual machine bytecode (like Java, Android's Dalvic, Microsoft's .NET) or high-level language with just-in-time compilation (Javascript, Python etc) - but its going to be a long time before that becomes universal, and probably forever before all the legacy x86 code is flushed out of the system.

 

[Maxx Power]

Yeah, too bad DEC (Digital) couldn't keep it going. It was great, at the time.
The Chinese (military) are very active with Alpha-like proc development for super computers.
The Chinese seem to like RISC a lot, they have also MIPS based processors for common devices.

The Chinese (and others) love RISC because there are no licensing fees with many instruction sets and therefore no lawsuits that follow such as with x86.

Another crucial difference is that while not many have agreed on a specific set of RISC instructions (PPC, MIPS, ARM, whatever), there is a much higher degree of homogeneity in the x86-x64 world. Changes to the instruction comes few per generation (like AVX, AVX 2.0, etc) and are not required to run any existing programs. In short, imagine x86-x64 as a common language that many already speak and agree with each other on the syntacs and vocabulary. The RISC camp has yet to achieve that.

Ultimately, as others have already pointed out, the instruction set is becoming abstracted in a sense because the underlying architecture can efficiently translate whatever instruction (CISC, say) into native microcode operations that execute, shuffle, schedule, parallelize and retire nearly independently of the incoming stream of CISC instructions. I think an advantage in this approach is that knowing the larger CISC instruction coming in (say a square root operation) gives the scheduler/decoding hardware an advantage for better scheduling of the execution pipeline.

 

[sittnick]

It will happen when . . .

We get rid of the QWERTY keyboard first.

 

[displaced]

Let's not forget the role of compilers

The stuff that modern compilers do is incredible. A 'complex' instruction set, implemented efficiently, might actually give a very smart compiler more options in generating the most efficient/performant machine code.

In fact, unless you're programming directly in assembly, the compiler's the key. Your CPU may be clean simply-instructioned, massively pipelined, etc., but if your compiler does a crappy job of keeping the pipelines full, then you're out of luck.

But, like so many of these 'great divides', great engineering and unusually smart people made it irrelevant. Turns out, in reality, neither 'CISC' nor 'RISC' as dogmas were the right answer. The best design turned out to be somewhat more nuanced.

(also, in general, let's not conflate the general brain-damagedness of x86 with the CISC concept in general)

 

[blesscheese]

"CISC vs RISC" is a last millennium debate - the world has moved on.

When I first saw the title of this thread, I thought, "Hello, 1996 is calling, and they want their thread back..."

 

[QuantumLo0p]

Brings back memories.

When I first started using cad/cam/cae software, everything was running on risc and cisc was little more than a glorified calculator. It would be years before cisc could even think about competing. I ran HP's with the HP-UX OS with CRX-48Z graphics and they were super nice. Their desktop UI was called VUE which IMO is what Apple copied for the OS-X tool bar. We had full hardware and software support so if anything happened a HP tech would be there within a couple hours. I had friends who used Silicon Graphics stations while others were on IBM and there was always a rivalry going on, lol.

Before the HP-UX I used Domain but that OS gives me the willies just thinking about it. Using tapes with the tar command was so hit and miss. *shudders*

 

[ActionableMango]

Ask anyone under 30 and the answer is... no-one remembers or knows about anything other than an iPhone. It's irritating.

My first workstation was a MicroVax running BSD Unix. As configured with its $15,000 color monitor and dual Winchesters, it cost $60,000.

$60,000? Why didn't you old timers just buy iPhones? They have more computing power, cost less, and come with a higher resolution color display!

And did you have to get a background check for the dueling Winchesters?

 

[kevink2]

Another thing I recall from those days is the rest of the architecture mattered too. PCs were built for DOS then Windows 95, which didn't really require good hardware interfaces that supported multitasking well. So you had hard drives using PIO. serial cards that shared interrupts. ISA slots that were limited in speed.

While the RISC workstations could be designed for high performance with the OS developed along with the hardware. SCSI. Bus mastering. Higher performance graphics. Platforms that could support higher amounts of memory. I remember PCs that would cache only the first 64 meg of RAM, for instance. Or some OSs would only support 16MB of RAM if there was a SCSI controller in the PC on the ISA bus. The PC world didn't really start to improve until Windows 9X went away. And Intel started moving the better technology down to the desktop.