26 Desktop Processors Compared 192
theraindog writes "The number of different CPU models available from AMD and Intel is daunting to say the least. The Tech Report's latest CPU review makes some sense of the landscape, exploring the performance and power consumption characteristics of more than two dozen desktop processors between the $999 Core i7-975 and more affordable sub-$100 chips. The article also highlights the value proposition offered by each CPU on its own and as a part of the total cost of a system. The resulting scatter plots nicely illustrate which CPUs deliver the best performance per dollar."
Correlation is positive (Score:4, Insightful)
Re:Correlation is positive (Score:2, Insightful)
You perhaps aren't cynical enough. I wonder how much this chart would change if the chips were overclocked. That would tell you if some of the slower processors were simply underclocked versions of the faster ones.
Re:Seems pretty clear: (Score:4, Insightful)
Not embarrassing, just not as fast. However, a Ferrari is vastly more powerful for the money, but a Corvette is still the much better deal -- do I really need to explain this on here?
Re:Seems pretty clear: (Score:4, Insightful)
Well now, they are faster. They're just not $800 faster. Nowhere near it.
Not to mention that the corresponding hardware you have to buy with a Core i7 (the expensive motherboard, the expensive triple channel ram) makes it even more not worth it.
Re:Seems pretty clear: (Score:3, Insightful)
Plot isn't that clear to me at all.
Price/performance is slope on that graph, and AMD looks pretty good at first look, occupying a band mostly on the top-left.
They ought to have plotted performance on the independent axis vs. price/performance on the y axis. And put that sucker on a log-log plot. Then we'll see who's got the best price/performance for various performance levels.
Re:Only Intel compatible .... (Score:3, Insightful)
They'll cover ARM when someone sells a motherboard with a socket I can stick a 2+ GHz quad-core ARM in and get performance equal to an equivalent AMD or Intel chip.
As it stands, ARM is irrelevant outside the embedded/pda/non-x86 netbook scope.
Re:A better solution. (Score:4, Insightful)
Re:Correlation is positive (Score:3, Insightful)
Aside from the possitive correlation, the other main point I would take away is the apparent 'knee' in the chart at about the $300 mark.
Seems that is the point of diminising returns.
Re:Seems pretty clear: (Score:4, Insightful)
Not if you factor in the motherboard price difference - about $100 for x58 over AM2+. The i7 is brilliant for heavy, multi-threaded workloads but I question how much of that performance is really worth it for an average consumer or even Slashdot reader.
Re:Seems pretty clear: (Score:5, Insightful)
The Phenom II X4 955 beats the i7 920 in 3 out of the 4 games they tried. The only one it lost was Cryis Warhead and it was a narrow loss (48 vs. 46 FPS).
These price difference of these two chips is about $35 on Newegg. I think for gamers, getting the X4 955 and putting that extra $35 towards the video card will net better results. This isn't counting the additional cost of DDR2 vs. DDR3 memory which has minimal effect on performance right now but still has a big price difference.
Re:Seems pretty clear: (Score:4, Insightful)
Personally I don't find it that trivial; never before has the best choice been so extremely dependent on what you need the performance for.
Say you've got some eminently parallel task, like ray tracing. With the huge price/performance difference between low end and high end you might beat the high end in performance by buying two cheap systems rather than one expensive. Look at the i7-940; it's just barely twice as fast as the cheapest CPU, yet it costs six times as much. That price would easily accommodate a cheap motherboard and memory and you'd still be shelling out less.
On the other hand, say you have very few parallel tasks, then you may still be as well off with a cheap CPU. Without several tasks you're not going so see much difference between a dual core CPU with good per-core performance and a quad core. Not an entirely rare situation when talking about desktop systems.
Or you might be in the sweet spot of latency sensitive parallel tasks, possibly applicable to some games, in which case a more expensive quad core CPU might definitely be what you're looking for.
And add to that the need for actual application specific benchmarks to determine actual performance as opposed to generic benchmarks... well, I wouldn't call the choice trivial.
Comment removed (Score:5, Insightful)
Re:Seems pretty clear: (Score:4, Insightful)
That's one way to go, but the most economically sound way to do it has always been (slightly) more frequent upgrade cycles and lower range hardware.
Re:What I'm seeing... (Score:0, Insightful)
No shit Sherlock. Please link to a motherboard that supports 2 X2s. Not Opterons, X2s.
What you say? There are none? Guess your whole post just became pointless.
Re:Seems pretty clear: (Score:3, Insightful)
I agree so much I'm going to take it to the next level:
Almost all hardware today is ridiculously powerful for ridiculously cheap. Review sites implicitly agree by reviewing at full detail 1600x1200 or higher.
Re:What I'm seeing... (Score:3, Insightful)
Re:Seems pretty clear: (Score:5, Insightful)
In science we call that honesty. You can make a tiny little difference look REALLY BIG by cropping your graph so that it only shows a very tiny range. By showing the origin tiny differences look, well, tiny.
Re:Seems pretty clear: (Score:5, Insightful)
My issue with the graph is someone needs to take a class on "how to make a graph". 0% performance and $0 cpu.... why? Was there a $0 cpu? Did any of the cpus get a 0%?
Probably because the person who made that graph for The Tech Report wanted all the proportions to be honest.
Did you ever read the book How to Lie with Statistics [wikipedia.org]? Or the book How to Lie with Charts [amazon.com]? Or a nice, short blog post called Graphs That Lie [gyford.com]?
When you chop out some of the "wasted space" in a graph, you distort the graph. Unless people are careful and check where your axes begin, and then mentally visualize where the axes go, they'll get a misleading idea of the data from the graph.
Suppose the bottom part of the graph was sliced off, at the 90% line, to make you happier. Imagine what it would look like. The AMD X2 6400+, sitting at the 100% line, would have very little white space under it; and the Intel i7-920, sitting a bit below the 200% line, would now appear to be ten times faster than the AMD X2 6400+. The numbers would be the same, but the visual impact would be that the Intel chip totally blows away the AMD chips.
The graph is good the way it is.
I'll meet you halfway, though: it wouldn't have hurt for them to have put in a second chart, zooming in on just the most crowded areas.
steveha
Re:Seems pretty clear: (Score:1, Insightful)
Meh, that's not long enough to even be new tech in 6 months. Not worth the money you lose. I seriously doubt you're getting 95% for a used system. Then factor in the inconvenience, shipping (if any), plus your labor and it's actually costing you a lot more than you think.