By: Brett (ggtgp.delete@this.yahoo.com), August 14, 2013 8:17 pm
Room: Moderated Discussions
Mark Roulo (nothanks.delete@this.xxx.com) on August 13, 2013 10:11 am wrote:
> Brett (ggtgp.delete@this.yahoo.com) on August 12, 2013 6:50 pm wrote:
> > Drazick (DrazickHellNo.delete@this.yahoo.com) on August 11, 2013 3:27 am wrote:
> > > Does "Power" architecture hold any advantage over ARM / MIPS?
> > > How do they all compared (In their 64 Bit variants)?
> >
> > Playstation 2 was 300 MHz MIPS, Playstation 3 was 3.2 GHz PowerPC, ARM does not go that high even today.
>
> But ARM *does* go to 3.1GHz ... with an Out of Order core
> (which the Playstation 3 and Playstation 2 did not have).
> http://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do?action=detail&language=E&newsid=6781
>
> It is unlikely that the frequency is *architecture* limited ... rather that
> ARM hasn't been aiming for high frequency markets (this is changing).
As I mentioned the worst architecture (x86) has the highest performance and also highest clock.
ARM has some warts/features like a free shift on one operand for most instructions, that has to add a bunch of gates and limit clock for simple designs. For high end you could just crack the few instructions that actually use the shift and pay an extra cycle in dependency, like x86.
Predication is apparently a problem child also as it was deleted from the 64 bit version. Most instructions had 4 bits of predication, this was not a small feature to conditionally execute almost every instruction. This deletion was a OMG revelation, NOT compatible upgrade. How horrid is 32 bit mode going to be on 64 bit ARM? Will it suck massively but none should care with 64 bit being so much faster? Everyone uses variable length Cortex instructions (not the old fixed width 32bit real ARM instructions) which already deleted those 4 bits, so it really does not matter? (YES) Does ARM64 only require Cortex compatibility?
> > Both ARM and PowerPC sold architecture licenses, while MIPS sued cloners.
> > Lots of ARM and PowerPC choices, few MIPS choices.
>
> I don't know how far back this goes, but MIPS does have architectural licensees.
I neglected to mention that MIPS rolled out all new CPU designs, to late to save the company, Imagination Technology bought the company as you pointed out. I did expect that MIPS was licensing, but it is the lawsuits that make the news, and that distorts the reality of what is going on.
> > POWER is trivially different from PowerPC, different marketing
> > label for workstation/server class performance and cost/price.
>
> Starting with "Power ISA v.2.03" the PowerPC lines and the POWER lines are converged.
> At least on paper. The implementations for the high end POWER line are obviously
> different from the implementations for the low end embedded lines.
I also neglected dozens of other CPU architectures, AVR32 and AVR (8bit) which is so cool to have a pipelined 8bit part, 68000->Coldfire, SH-2/3/4, PIC flavors, and the old 8 biters Z80/805?/6502, etc.
http://en.wikipedia.org/wiki/Atmel_AVR
And PowerPC has some whacky optional extensions like a floating point accumulator, likely used in some ColdFire part that was ported up to PowerPC. (If I remember this correct...)
> Brett (ggtgp.delete@this.yahoo.com) on August 12, 2013 6:50 pm wrote:
> > Drazick (DrazickHellNo.delete@this.yahoo.com) on August 11, 2013 3:27 am wrote:
> > > Does "Power" architecture hold any advantage over ARM / MIPS?
> > > How do they all compared (In their 64 Bit variants)?
> >
> > Playstation 2 was 300 MHz MIPS, Playstation 3 was 3.2 GHz PowerPC, ARM does not go that high even today.
>
> But ARM *does* go to 3.1GHz ... with an Out of Order core
> (which the Playstation 3 and Playstation 2 did not have).
> http://www.tsmc.com/tsmcdotcom/PRListingNewsAction.do?action=detail&language=E&newsid=6781
>
> It is unlikely that the frequency is *architecture* limited ... rather that
> ARM hasn't been aiming for high frequency markets (this is changing).
As I mentioned the worst architecture (x86) has the highest performance and also highest clock.
ARM has some warts/features like a free shift on one operand for most instructions, that has to add a bunch of gates and limit clock for simple designs. For high end you could just crack the few instructions that actually use the shift and pay an extra cycle in dependency, like x86.
Predication is apparently a problem child also as it was deleted from the 64 bit version. Most instructions had 4 bits of predication, this was not a small feature to conditionally execute almost every instruction. This deletion was a OMG revelation, NOT compatible upgrade. How horrid is 32 bit mode going to be on 64 bit ARM? Will it suck massively but none should care with 64 bit being so much faster? Everyone uses variable length Cortex instructions (not the old fixed width 32bit real ARM instructions) which already deleted those 4 bits, so it really does not matter? (YES) Does ARM64 only require Cortex compatibility?
> > Both ARM and PowerPC sold architecture licenses, while MIPS sued cloners.
> > Lots of ARM and PowerPC choices, few MIPS choices.
>
> I don't know how far back this goes, but MIPS does have architectural licensees.
I neglected to mention that MIPS rolled out all new CPU designs, to late to save the company, Imagination Technology bought the company as you pointed out. I did expect that MIPS was licensing, but it is the lawsuits that make the news, and that distorts the reality of what is going on.
> > POWER is trivially different from PowerPC, different marketing
> > label for workstation/server class performance and cost/price.
>
> Starting with "Power ISA v.2.03" the PowerPC lines and the POWER lines are converged.
> At least on paper. The implementations for the high end POWER line are obviously
> different from the implementations for the low end embedded lines.
I also neglected dozens of other CPU architectures, AVR32 and AVR (8bit) which is so cool to have a pipelined 8bit part, 68000->Coldfire, SH-2/3/4, PIC flavors, and the old 8 biters Z80/805?/6502, etc.
http://en.wikipedia.org/wiki/Atmel_AVR
And PowerPC has some whacky optional extensions like a floating point accumulator, likely used in some ColdFire part that was ported up to PowerPC. (If I remember this correct...)