Of dozens of articles about computer value I've read over the years, and many dozens more about performance, this is easily the best. Many kudos.

When someone is choosing a cpu they would best consider likely uses. For games for example, only certain current games need cpu speed above the low end dual cores. Low end dual cores also handle common heavy multitasking of "power" users at businesses except for certain (less common) applications where speed can be all-important. There's no point in having a 5600+ (also my personal favorite) if you just don't need that level of speed for what you do. There's no point in settling for only a q6600 when all possible speed is critically needed and fundamental to immediate return (profits or productivity) for one's business. Therefore the idea of value winners in segments is useful for those able to figure out how much cpu speed they need and without special needs, and who can also guess how much cpu speed they'll need for 2-3 more years.

But often just plain price and performance needs trump all else. Realizing what you can spend is often important for home users. Further complexities have to do with upgrading the cpu in your current motherboard, etc., and it's all grist for the mill.

I also enjoyed comments below. Again Kudos to TechRep for this excellent article.

An excellent article, but as an economist I would like to make a few remarks to put it in a larger context.

Technically speaking, you offer a cost-benefit analyisis of using a computer, with CPU price as a proxy for cost and benchmark performance as a proxy for benefits.

As far as the benefits are concerned, benchmark performance is probably the best proxy you can have, since benefits vary significantly from one user to another. For the cost, however, there are some better choices.

First of all, lets say that your analysis would be perfetly applicable for an upgrade situation where you replace only the CPU. If, however, you buy a new system you should definetly consider the system cost. As suggested in #39, you could for example evaluate price performance for three types of systems - value, mainstream and performance. In this case you will most likely have different winners in each category, which makes perfect sense.

If you wanted to find an even better measure of cost, you would have to concider the life-cycle cost (LCC) and big companies will often do just that when they evaluate theit IT purchasing options. But in our case LCC would probably be an overkill, so the system's initial cost is the best option for the cost benefit analysis.

Once again, this is not to criticize the choices you made in this article, but to suggest how you future articles can be made even better.

I think the value proposition of the Q6600 is understated. I understand that conjecture is bad in a conclusion, but surely with the increase in multi-threaded code that should happen in the next few years, this is possibly just as wise an investment as the better dual cores?

My argument goes something like as follows:
- Right now, you get better performance, just not quite as much value.
- For a lot of business applications, system downtime / instability / even the risk of instability is money. Lots of it. Not having to move to a whole new computer (upgrades are so pre-millennium) for a year or two longer is possibly worth a lot of money to some people.
- This suggests to me, that a lot of people will buy a fast dual core, but will stick with it for that year or two longer anyways, and miss out on all the quad-threaded performance they would have had anyways.

If I had the money now, that seems like good odds that I'd come out ahead. To me at least.

Nice work gang!

I still want to know if this was Cyril's first big review here.
He knocked it out of the park.

This was an interesting read and a well written article.

However, I think there is a flaw in the analysis.
We cannot compare performance per dollar looking only at the cost of the CPU. We need to take into account the other components of the system (without monitor, speakers, keyboard), but we have to include the cost of the motherboard and the cost of the memory at a minimum. For games' performance we also have to include the cost of the GPU, though if we benchmark them at low resolutions we should perhaps use a cheap/mainstream card to avoid skewing the numbers in the other direction.

The problem is that how you defined performance/$ in this article, it corresponds to the slope of the lines that go through the origin and each of the scatter points. By considering only the cost of the processor, the cheapest CPUs receive an unrealistically high score. Performance is a function of more than just the CPU, and we simply cannot have a CPU running benchmarks without a motherboard, some RAM and a power-supply at a minimum.

In these tests you used similarly priced components with all CPUs. By adding this fix cost to all systems, all your scatter graphs will slide somewhat to the right on the x axis. The value results would be a bit more interesting in that case and I think the X2 3600+ would not win that many tests anymore. If we assume that our system sans CPU costs $400 and we can add either a $100 or a $200 processor, then the cost of the second system is only 20% higher and not double the price of the first system. If it gets 20% better performance, then it has a better value.

Of course, you could use cheaper components with a cheaper CPU to reflect how they are used in practice. Therefore, I think this analysis works best to analyze full systems. And perhaps a very good use of this analysis would be to compare the performance/$ value of the systems recommended by TR at various price points.

I don't mind this review, but what about
e2140
e2160
e4400
e6320
e6420

for completeness
Some of these are newer, and some of these have been out for awhile (why use e6300 or e6400?).

I'd like to see how the chart would look if overclocking (safe, with default voltage) were taken into account.

EXCELLENT article. This is the kind of analysis you'll never see at sites like Anandtech. They'll spend pages and pages going over 3-5% differences in memory/motherboard but won't use a multitasking test when measuring multi-core CPU's. And quantifying value? Never

Great job, TR

Are we going to get a graphics card version now?

That would be awesome.

Another factor, is I don't care about saving $20 to drop to an inferior architecture.

I think at this point everyone knows you can buy a bottom end Intel Core 2 Duo and easily overclock it to obliterate the entire AMD lineup.

It is about getting the biggest bang for a reasonable buck.

i would have liked to have seen a pentium d in there
mainly because it's intel's cheapest offering
but also just to see how far we've come in a year
(and how much current processors are leaving netburst in the dust)

yay! exactly the kind of article I love on TR.

next up, memory, graphics, and MB's.

I could care less about best performance, I care more about bang/buck/value (when purchasing for personal reasons.)

Dude, you're getting Sloshdatted!

The lack of any Exx20's seems a little blearing... Thanks for the effort tho, looks great! I just wish the E6420 I just bought was somewhere on the graph.

Great article, doing something a little different this time around was nice, and reading the charts was interesting.

However I didn't understand how the overall platform cost issue was handled, did I miss the explanation of how differences in cost there were factored in?

Great article but the gaming section needs some RTS tests. Can we see some Company of Heroes or SupCom info?

Google doesn't buy fancy servers. They buy the cheapest workstation that fits their needs and clusters them together.

I think if your software is "clusterable", then the 3600+ is KING!!!

If not, 50$ or 100$ more for a processor is a good trade-off.

Great review, its nice to see TR still comes up with the goods.

For the next one could you do this for an "Office style PC" review i.e. mobo with integrated graphics, etc maybe 690 vs 965G as this is more likely to be what the cheaper CPUs are used for.

see, that's why i like this site. different takes from what everyone else is doing.

I have nothing to add other than to say thank you for this article. Great info, great effort. yet another reason why TR is the best tech site on the net. :)

The final graph should display price as a percentage of the 3600+'s, just as it does performance.

Very good and relevant article. Some people here seem to believe that overclocking is custom practice, on the contrary its very uncommon for people to take that risk, even among techsavvy people its not that usual to overlock.

Thanks Cyril. Applause for TR. You guys have been burning the candle at both ends lately.

My only big question about not only this review but the coverage of the CPU world since the launch of th AM2 platform: why no coverage of AM2 Opterons? They all come with 1meg cache and their prices dip down to $120ish. 939 Opterons were known for overclocking and for being a great value - is the picture the same for AM2 Opterons? If AM2 Opterons had been included in this test is there a chance that one of them would have ended up being one of the highest value CPUs?

Just wondering because I've seen zero coverage of AM2 Opterons and yet they're readily available and very attractively priced.

Respectfully,
flip-mode

i liked how, in the final chart, the processors between intel & amd were kind of bunched together - you could see which was competitive with which other CPU.

it made for an interesting second read, checking those "pairs" against one another throughout the previous section of the article.

We cannot compare performance per dollar looking only at the cost of the CPU. We need to take into account the other components of the system (without monitor, speakers, keyboard), but we have to include the cost of the motherboard and the cost of the memory at a minimum. For games' performance we also have to include the cost of the GPU, though if we benchmark them at low resolutions we should perhaps use a cheap/mainstream card to avoid skewing the numbers in the other direction.

This entire line of reason is specious at best, irrelevant at worst. Look at the systems in question; the only cost variation in the systems of any consequence is the motherboard, which from a cost standpoint are within a few dollars based on a cursory Newegg search. Everything else is the same--same OS, same hard drives, same clocking memory, same video card. They've removed the possible variables to isolate and expose the relative CPU performance benefits quite well. Including things like peripherals into the equation serves zero purpose at all with regard to cost/benefit of the CPU.

Further, given the randomness and variability associated with overclocking considerations, such analysis would only serve to obfuscate what is already a difficult process of determining CPU value based on cost/performance metrics because it injects non-standard, non-repeatable and random occurrences into the value proposition analysis. The TR crew did a great job of clearly illustrating the cost/performance factor without needlessly complicating the exercise.

I posted this before, and it's applicable here again:

http://www.xbitlabs.com/articles/cpu/display/dualcore-roundup_9.htm...

Extremely well written aritcle, it's nice to see a different style of article as well.

Good article.

I think it confirms what most people have suspected - that AMD and Intel are both good value at the lower end, and "Extreme" CPUs from both companies are a waste of money unless they can literally save you minutes, and the sum of those minutes is worth $$$ to you. I.e., not for a general purpose computer.

I think it is fair to exclude overclocking right now - there's enough article as it is, we don't want to overcomplicate it. Overclocking would probably benefit the low-end Intel products more though, but don't forget that not every one overclocks the same... and lots more people prefer to silence their machines than overclock these days.

The only bang for the buck interesting to me is after over-clocking.

I have been building my computers since the 80486 days (when I switched from Amigas to PC's) and I have NEVER run a CPU at stock speed, but I don't tend to invest in super cooling solutions. Though my next build I will probably get a slightly better air solution than stock.

So that is the only thing I am really interested in. What is the simple overclock for most of the these procs. What is the dead simple, drop in and overclock speed without needing liquid cooling, or super expensive ram.

Because bang/buck at stock speed is pretty much dead obvious and bloody useless to a large amount of us who will never run stock speed.