CPU Development Progress (Low End Value)

[s-p]

Having recently pondered about an upgrade to my ageing system (purchased in 07), I thought I'd take a look at what my money purchased me back then relative to what we would get today with the same level of cash.

So my current CPU is a E2140, purchased back in July 07 for £43. In comparison, that amount of cash today roughly buys you a Intel Pentium G620 2.60GHz.

On first inspection, it appears we have a higher clocked CPU, same number of cores and what I expect is a more efficient CPU (i.e. less CPU cycles per unit of work). A quick look at a CPU benchmark comparison would indicate this to be the case.

Anandtech Benchmark: http://www.anandtech.com/bench/Product/70?vs=406

A look at the results shows us that performance has approximately doubled within this 5 year time frame. Using Moore's Law, or preferably David House's statement that performance of CPUs was to double every 18 months, then we appear to be lagging behind somewhat.

As I'm sure many of you are aware, the E2140 is a fantastic chip for overclocking. In fact, the majority of chips were able to double their clock speed allowing for performance not too shy from today's G620. Compare this to the Sandy Bridge architecture of the G620 where increases are limited to approximately 5-7% (via the BClk) and I'm left scratching my head as to how far behind the E2140 actually sits behind the G620 once we've performed our tweaks.

From my admittedly very naive viewpoint, I'm struggling to see the massive leap in performance I would have expected over this period of time. I probably would have "liked" to see Quad core as default at this price level 5 years on from my original purchase, with maybe a small bump in clock speed and a more efficient architecture.

Am I looking at this all wrong, or is the value I found with the E2140 no longer available in today's marketplace? This is a genuine question as I know nothing about today's crop of CPUs, but my initial findings bring me to the conclusion I can't perform the same value based upgrade as I did when moving from a 2800+ to my E2140.

Thoughts?

 

[edscdk]

Moors law is to do with the number of transistors not CPU power

 

[FreeStream]

Moors law is to do with the number of transistors not CPU power

This.

For the sake of argument, if you go from a dual to a quad it is twice the transistor count (although can be as much as three times), so in theory compute power should have doubled at-least in that time frame.

Clearly the quads haven't made it down to cheap cheap status yet, mainly because duals actually do what 95% of people need them for and the market is shifting very much towards laptops.

Looking at the high end, even the jump from Core i7 970 to Core i7 980X wasn't double and that was 18 months.

 

[beejjacobs]

There is a clear improvement at stock speed, and bare in mind as you say efficiency [in IPC and power consumption] have improved too - raw speed is not the aim of these inexpensive CPUs, as like Concorde Rules says, a cheap dual core will do what most people want anyway.

The other problem, is that you can overclock your current CPU so well. The Pentium is probably good at overclocking too, based on what we see people getting with i5 2500Ks. It's not that its bad at overclocking, it's that Intel have deliberately limited its overclocking to stop people buying cheap chips and OCing rather than buying their more expensive chips.

 

[neo-omega]

Another big thing is that Moore's law doesn't depend on pricetag.

 

[s-p]

Moors law is to do with the number of transistors not CPU power

I was careful in my original post not to make reference to Moore's Law directly as being the doubling of performance, but instead took the reference to the Intel executive David House's quote which factors in Moore's Law to quote the 18 month period. Given performance gains aren't just applicable to increases in transistor count but other factors as well (IPC), would we not expect performance increases to happen over a shorter period of time than Moore's Law dictates?

The other problem, is that you can overclock your current CPU so well. The Pentium is probably good at overclocking too, based on what we see people getting with i5 2500Ks. It's not that its bad at overclocking, it's that Intel have deliberately limited its overclocking to stop people buying cheap chips and OCing rather than buying their more expensive chips.

As an end consumer, I care only for the performance I'm able to achieve from my purchase. This is why the discussion is not purely about the technical limitations of progress, but understanding where the market currently is. So if Intel decide to limit the overclocking ability of some of their Sandy Bridge chips, then that immediately reduces their value to me. Why Intel do this is kind of irrelevant as I'm only concerned about the value I can extract from my CPU purchase.

 

[FreeStream]

I was careful in my original post not to make reference to Moore's Law directly as being the doubling of performance, but instead took the reference to the Intel executive David House's quote which factors in Moore's Law to quote the 18 month period. Given performance gains aren't just applicable to increases in transistor count but other factors as well (IPC), would we not expect performance increases to happen over a shorter period of time than Moore's Law dictates?

Not necessarily. The lower end CPUs now include GPUs, memory controllers (with more channels), more PCI-E lanes, etc. Thereby increasing the transistor count without actually increasing performance much if at all but reducing overall system power instead.

The Core i7-980X to the i7-3960X is double the transistors but the performance isn't double. About 10-20% in reality.