Why not to use a Pentium 4 in 2013
Which offers better performance, an Atom server or a Pentium 4 one?
It's an interesting comparison, and I don't know the direct answer.
The core problem with CPU performance comparison is that people tend to latch on to clock rate as an easily-understood measure. This doesn't work because clock rate is essentially only very loosely correlated to performance. It's like saying that a 125cc motorbike is 4x more powerful than a 16 litre truck engine, because the motorbike engine does 16,000 rpm and the truck engine maxes out at 4,000rpm. The relative speeds are true; the conclusion drawn from them is utterly wrong. Yes, one revs faster, but it is in no way more powerful. It is less powerful.
The correction factor - and it's still only approximate - is instructions per clock (IPC).
Let's say that a Pentium did 0.25 IPC. That's a figure from thin air, incidentally.
So a Pentium running at 200MHz did a quarter that many instructions per second - 50 MIPS. (It did not actually do so, note.)
A Pentium Pro (totally different, new architecture, unrelated to the Pentium) might do 0.5 IPC, but run at 233 MHz - that's 116.5 MIPS.
A Pentium II (an improved PPro) might do 0.75 at 350MHz - 262 MIPS.
The Pentium III, the same, but faster MHz - maxing out at 1GHz: 750 MIPS.
The P4 was specifically designed to rev fast - i.e. to have a very high clock rate. Its IPC is very very low - let's say 0.1. (Yes, I do mean VERY low.)
So a 2.3GHz P4 looks quick, right? Compared say to a 1GHz PIII.
But it isn't. It actually returns only 230 MIPS because it needs many clocks to complete a single instruction.
I've not looked at actual IPC figures for the P4 and so on, but the general trend I've indicated is broadly in line with reality - the IPC for Intel x86 chips climbed slowly but steadily from the 386 on upwards. The 486 was better, the 586 (Pentium) better still, the PPro much better but only on certain 32-bit workloads, the PII a bit better than the PPro, the PIII just a faster-clocked PII with better MMX.
Then there was a huge drop with the P4, which was a design disaster. It was meant to run at many gigaHertz but the technology didn't deliver and so its performance was dire.
Also, it came with a chipset that only supported RAMbus RAM, which was very expensive and which also did not deliver the promised performance.
Then there was a hasty version 1.1 chipset which supported SDRAM, but over a converter, further reducing performance. Only late models supported DDR RAM natively and started getting half-decent RAM performance - but still woefully abysmal CPU performance.
Then AMD launched its Sledgehammer design as the Athlon64 and Opteron, with way better IPC, plus a 64-bit instruction set with twice as many registers, and completely kicked Intel's arse. They ran at slower clock speeds but were actually dramatically faster than P4s.
And Intel did a huge, vastly-expensive U-turn, fired and closed its whole Indian design centre, reassigned most of its American design team, and adopted Intel Israel's mobile CPU for notebooks, the Pentium-M - a revised Pentium III design with a Pentium 4 bus on the front.
Microsoft compelled Intel to licence AMD's 64-bit instruction set (a humiliating climbdown for Intel).
The initial desktop version of this that was rushed out was the Core Solo and Core Duo, with just the 32-bit instruction set. Intel also bolted the 64-bit instruction set on to the last couple of models of P4 but they were still utter dogs and to be avoided at all costs.
Then came a revamped Core chip with 64-bit instructions: the Core 2 series.
The rest is history.
Now the original question was, Atom versus P4 performance. The Atom is a very basic CPU core - in terms of IPC, it's a bit better than a Pentium 1, but it's nowhere near as good as a PIII. It's roughly PPro to PII level. But, of course, Atoms run at way higher clock speeds than the PII ever could (which was 166 to 350MHz, IIRC.)
At the same clock speed - which almost never happens, because the fastest Atoms run at about the speed of the slowest P4s - I'd expect an Atom to give a P4 a good run for its money. Atom chipsets' memory controllers are vastly better than the P4's, which helps.
But P4s run at about twice the speed of Atoms - typically the surviving mid-to-late period ones run from 2.4 to about 3GHz.
The P4 might be a bit quicker, but it will burn about 4x, maybe even 6x as much electricity to do it.
[Original poster], if you had an old Athlon64 or Opteron box, or a Core 2 box, I'd say go for it - so long as they were models that supported hardware virtualisation.
You don't.
You have an old box with the worst x86 CPU ever made in it. It's slow, inefficient and crap. It was crap when it was new - all P4s were. Now it's old and it's junk. It's a liability.
Give it to ComputerAid and spend £100-£150 on a cheap new motherboard with the cheapest Core i3 or i5 you can find that supports hardware virtualisation. If you're really desperate to save money, the AMD APUs are OK - their CPU performance is merely barely OK but the graphics are good, not that that will do you any good in a server. But they're cheap.
They'll deliver many times the performance of your P4 while drawing a quarter of its electricity.
And that means that, if you get one with a 3Y warranty on it, it will save more money than it cost to buy before the warranty runs out.
If that is not argument enough, I don't know what is.
It's an interesting comparison, and I don't know the direct answer.
The core problem with CPU performance comparison is that people tend to latch on to clock rate as an easily-understood measure. This doesn't work because clock rate is essentially only very loosely correlated to performance. It's like saying that a 125cc motorbike is 4x more powerful than a 16 litre truck engine, because the motorbike engine does 16,000 rpm and the truck engine maxes out at 4,000rpm. The relative speeds are true; the conclusion drawn from them is utterly wrong. Yes, one revs faster, but it is in no way more powerful. It is less powerful.
The correction factor - and it's still only approximate - is instructions per clock (IPC).
Let's say that a Pentium did 0.25 IPC. That's a figure from thin air, incidentally.
So a Pentium running at 200MHz did a quarter that many instructions per second - 50 MIPS. (It did not actually do so, note.)
A Pentium Pro (totally different, new architecture, unrelated to the Pentium) might do 0.5 IPC, but run at 233 MHz - that's 116.5 MIPS.
A Pentium II (an improved PPro) might do 0.75 at 350MHz - 262 MIPS.
The Pentium III, the same, but faster MHz - maxing out at 1GHz: 750 MIPS.
The P4 was specifically designed to rev fast - i.e. to have a very high clock rate. Its IPC is very very low - let's say 0.1. (Yes, I do mean VERY low.)
So a 2.3GHz P4 looks quick, right? Compared say to a 1GHz PIII.
But it isn't. It actually returns only 230 MIPS because it needs many clocks to complete a single instruction.
I've not looked at actual IPC figures for the P4 and so on, but the general trend I've indicated is broadly in line with reality - the IPC for Intel x86 chips climbed slowly but steadily from the 386 on upwards. The 486 was better, the 586 (Pentium) better still, the PPro much better but only on certain 32-bit workloads, the PII a bit better than the PPro, the PIII just a faster-clocked PII with better MMX.
Then there was a huge drop with the P4, which was a design disaster. It was meant to run at many gigaHertz but the technology didn't deliver and so its performance was dire.
Also, it came with a chipset that only supported RAMbus RAM, which was very expensive and which also did not deliver the promised performance.
Then there was a hasty version 1.1 chipset which supported SDRAM, but over a converter, further reducing performance. Only late models supported DDR RAM natively and started getting half-decent RAM performance - but still woefully abysmal CPU performance.
Then AMD launched its Sledgehammer design as the Athlon64 and Opteron, with way better IPC, plus a 64-bit instruction set with twice as many registers, and completely kicked Intel's arse. They ran at slower clock speeds but were actually dramatically faster than P4s.
And Intel did a huge, vastly-expensive U-turn, fired and closed its whole Indian design centre, reassigned most of its American design team, and adopted Intel Israel's mobile CPU for notebooks, the Pentium-M - a revised Pentium III design with a Pentium 4 bus on the front.
Microsoft compelled Intel to licence AMD's 64-bit instruction set (a humiliating climbdown for Intel).
The initial desktop version of this that was rushed out was the Core Solo and Core Duo, with just the 32-bit instruction set. Intel also bolted the 64-bit instruction set on to the last couple of models of P4 but they were still utter dogs and to be avoided at all costs.
Then came a revamped Core chip with 64-bit instructions: the Core 2 series.
The rest is history.
Now the original question was, Atom versus P4 performance. The Atom is a very basic CPU core - in terms of IPC, it's a bit better than a Pentium 1, but it's nowhere near as good as a PIII. It's roughly PPro to PII level. But, of course, Atoms run at way higher clock speeds than the PII ever could (which was 166 to 350MHz, IIRC.)
At the same clock speed - which almost never happens, because the fastest Atoms run at about the speed of the slowest P4s - I'd expect an Atom to give a P4 a good run for its money. Atom chipsets' memory controllers are vastly better than the P4's, which helps.
But P4s run at about twice the speed of Atoms - typically the surviving mid-to-late period ones run from 2.4 to about 3GHz.
The P4 might be a bit quicker, but it will burn about 4x, maybe even 6x as much electricity to do it.
[Original poster], if you had an old Athlon64 or Opteron box, or a Core 2 box, I'd say go for it - so long as they were models that supported hardware virtualisation.
You don't.
You have an old box with the worst x86 CPU ever made in it. It's slow, inefficient and crap. It was crap when it was new - all P4s were. Now it's old and it's junk. It's a liability.
Give it to ComputerAid and spend £100-£150 on a cheap new motherboard with the cheapest Core i3 or i5 you can find that supports hardware virtualisation. If you're really desperate to save money, the AMD APUs are OK - their CPU performance is merely barely OK but the graphics are good, not that that will do you any good in a server. But they're cheap.
They'll deliver many times the performance of your P4 while drawing a quarter of its electricity.
And that means that, if you get one with a 3Y warranty on it, it will save more money than it cost to buy before the warranty runs out.
If that is not argument enough, I don't know what is.