Slashdot Mirror
Performance Comparison of Current Intel Core 2 CPUs
crazyeyes writes "Intel has way too many Core 2 processor models. No one really knows if it's worth paying $100 more for a Core 2 Quad, instead of a Core 2 Duo. And when tech websites start interjecting codenames like Wolfdale, Kentsfield and Yorkfield, you know the battle is lost. All we want is a simple guide on the REAL WORLD performance differences between the many Intel Core 2 processors. How do they perform in games like Crysis, 3D rendering software, video encoding software, etc.? Fortunately, there is such a guide — just simple comparisons of the relative performance of these CPUs."
Someone needs to do the same for nvidia graphics cards...
I went out and bought an 8600 card, only to find out later that a 7900 is actually faster (despite being lower model number and previous generation tech).
http://spamdecoy.net - free throwaway anonymous email - avoid spam!
Please, intel. Codenames are fine for internal development but for public products it's just insane.
Do we really need that many different types of CPUs anyway?
It's easy to understand "Core" vs "Core 2" (2nd version of Core), and "Solo", "Duo", "Quad" (number of cores). More than that though, and you're only confusing your customers.
This is the list of the CPUs in the article:
Core 2 Extreme QX9650
Core 2 Quad Q9550
Core 2 Quad Q9450
Core 2 Duo E8500
Core 2 Duo E8400
Core 2 Duo E8300
Core 2 Duo E8200/E8190
Core 2 Extreme QX6850
Core 2 Extreme QX6800
Core 2 Extreme QX6700
Core 2 Quad Q6700
Core 2 Quad Q6600
Core 2 Duo E6850
Core 2 Duo E6750
Core 2 Duo E6600
Core 2 Duo E6550/E6540
Core 2 Duo E6420
Core 2 Duo E6320
Seriously, someone in the marketing department needs a swift kick in the ass.
I'm also curious to see a greater variety of benchmarks, for both 32 and 64bit code... And explanations of wether the code uses SSE functions or not etc...
Different processors are faster for different tasks, and i would like to get the best price/performance relative to what i'm doing...
If i'm running purely 64bit code for instance, AMD cpus do quite well, but if i'm running heavy SSE based code Intel chips tend to perform a lot better.
http://spamdecoy.net - free throwaway anonymous email - avoid spam!
* If you want two cores, you buy the E8400.
* If you want four cores you buy the Q9450.
* If you want two cores and can't afford the E8400, you buy a cheaper CPU in the series.
* If you want four corse and can't afford the Q9450 you buy a cheaper CPU in the series.
* If you don't know what you want, you buy a C2D in the cost-bracket that you feel affordable.
Wow, could that site possibly have any more ads on it?
What ads?
A fool and his lamb are worth two in the bush.
The best bang-for-the-buck in medium to high end Intel gaming rigs is currently an OC'ed E8400 (or stock E8500/E8600 if you have the cash) on a reputable X38 or X48 mobo (you'll want to use a 1600mhz fsb), with some good dual-channel DDR3 ram (PC3-12800), and a single nvidia 8800 gtx.
Can anyone provide some elucidation for this topic?
jdb2
US Government regulations require that Intel publish performance numbers for all of their CPUs. See the following links for the relative performance of all of Intel's CPUs. Make your own graphs if you need a pretty picture.
Intel microprocessor export compliance metrics:
http://www.cpubenchmark.net/
Almost every CPU compared. Now was that so hard?
Without seeing the source for both machines, it'd be very hard to say exactly why this is. But, from the "sky-level" view that we have, there are a number of rather obvious factors as to why the Core 2 architecture is faster than Barcelona.
The Core 2 is effectively wider than Barcelona. The Barcelona is very much like the K8 in the front end, with only 3 "complex", full-time instruction decoders. The Core 2 has three "simple" decoders, and one "complex" decoder, with macro-op fusion (possibly decoding 5 x86 instructions with Penryn). The Barcelona is only capable of dispatching 3 uops to the next stage of the pipeline, whereas the Core 2 can possibly dispatch 4 (5 if you count uop fusion).
The only other significant performance difference between the architectures is the cache architecture, where Intel has been running circles around AMD for years now (and there's plenty of material on why Intel's cache is so much faster than AMD's).
There are a lot of micro-performance optimization differences between the two architectures, but quite simply, Intel's stripped the x86 architecture down and made their chips run extremely fast on the most basic instructions, damning the performance on the complex instructions. AMD's performed significantly better in the past as they did a very good job across all of the x86 instructions. But now, as compilers have become tuned to better generate code that avoids more expensive instructions, you start to see that performance advantage AMD had over Intel for years evaporate. This, coupled with Intel's work to make the front end of their chips wider to accommodate the extra width in the backend they've had for years, and it becomes extremely apparent why Core 2 is so much faster.
As much as people want discard the idea of "they're all RISC behind the scenes", this is an excellent example of just how true it is; as Intel's chips express more and more RISC-like behavior, they're becoming faster and faster. Almost every performance win can be directly attributed to the work to better saturate the simple instruction decoders and free up the complex decoder. Meanwhile, as AMD's chips become more monolithic (doing the same exact thing as they did before, plus picking up whatever Intel technology they can add), they're losing territory. AMD could add a 4th complex decoder and make up a huge amount of ground, but it would be a significant undertaking (and by then, Intel will already be bringing Sandy Bridge to the market with 256-bit SSE, etc. etc.)
I don't really see the point in this sort of study.
The games mentioned, particularly Crysis, are going to get limited by graphics cards rather than processors anyway beyond the very lowest level processors and the most simplistic in-game graphics settings.
Some of the processors overclock much better than others, and for anyone who's actually reading an article like this and going to do anything with the information, this is likely to be very relevant.
FWIW, it's only a few months since I put together my latest rig. At the time, the widely agreed sweet spots in price/performance were the newly released E6850 and the Q6600. The E6850 runs at 3GHz but has only two cores; the Q6600 runs at only 2.4GHz but has four cores and double the cache. Unless you've got a demanding application that can take proper advantage of all the cores (and I've got a load of benchmarks that tell me neither of the games mentioned in TFA do so in practice, despite some of the publicity surrounding their releases) that made the E6850 seem like the better buy in the short term, but probably the Q6600 the more future-proof choice if you expect your PC to last long enough for demanding applications like games to get multicore right. But then you realise that actually the Q6600 is well-known for being highly overclockable, and numerous places will supply components or build you a whole machine to match a Q6600 running at 3GHz or more. That made the Q6600 a complete no-brainer.
And for what it's worth, looking at both Crysis and Supreme Commander running at 1920x1200 with high quality settings on an 8800GTX and the aforementioned Q6600@3GHz, the answer is "fast enough, because it's the graphics that limit things anyway, but when the processor matters a few more Hz won't help if you're ignoring most of your cores anyway".
If you disagree, post your argument. (-1, Overrated) isn't your personal censorship tool for views you don't like.
I haven't been known to be on the top of the CPU performance graph for years now, and this benchmark proves one thing to me: I'll rather go for a fast hard disk and fast internet connection before thinking about upgrading my CPU. Sure, there are things that are CPU bound. But only if I really see a need for performance for a very specific task, then I'll look for a quad core or high frequency processor. And if it is for a single task, I'll look around and ask a few friends who might have one already. Maybe for batch graphics processing or PAR2, but even then the HDD might max out instead of the CPU.
7-zip compressing. Multimedia encoding. VC++ has an ever so awesome /MP switch. Windows threads getting previews for all your files in explorer. Heck, even Excel 2007 will thread your calculations behind the scenes.
Games may not make significant use of multi-core yet, but some other real-world things do. Quad core definitely makes an impact on my daily usage.
Yeah, remember when we talked about the benchmarks being only marginally meaningful, back when there were lots of CPU architecture options for the desktop? 6809 vs. 8088, 68K vs. x86 (>286), and then all the various reductions (optimizations) to specific classes of operations (sparc, arm, etc.)?
All the stupid half-baked arguments about which was better overall when you are more interested in specific functions, all the production of half-useful benchmarks taken completely out of context to support the sales crew's arguments to buy their processor.
Apple to oranges.
So, what has apparently won for the desktop is the company that had the money and ambition to invest in emulating their old architecture at the highest speed. (And the chutzpah to steal engineers and tech to get there.)
"Hurray for us!!! We can run the old race course at 1.2 times as fast as brand X!!!!"
(And brand X is being tuned for a different race for some reason. Is it any wonder the industry seems to be going in circles?)
No. Seriously. Look at a dual G4 and a Core 2 duo at the same processor and cache speeds running similar code. For all the hype, there isn't nearly as much difference as you'd expect, considering the difference in expenditure in the development cycle. In fact, you^H^H^H I have to wonder whether iNTEL's ambitions haven't led it to doing the hare vs. tortoise game again, running fast in the wrong places and failing to run the race in the important places.
So, we all suckered in for it. Just like we suckered in for Microsoft's sales pitch in the early '90s.
And now we can't even intelligently compare apples to apples any more. (It would be oranges to oranges except that Jobs has bought into it, too.)
Computer memory is just fancy paper, CPUs just fancy pens with fancy erasers; the 'net is just a fancy backyard fence.
E8xxx are 45nm. The die is Wolfdale. E6xxx are 65nm. The die is Conroe/Allendale. Q9xxx are quad 45nm. The die is Yorkfield. Q6xxx are quad 65nm. The die is Kentsfield. Generally speaking, anything in the E8xxx line is better than any of the E6xxx line, and anything in the Q9xxx line is better than anything in the Q6xxx line. I'm not understanding what's hard to figure out about this. It's pretty clear nomenclature, especially compared to AMD's 3405940900+X2 silly naming system that's a relic of the megahertz wars, and ATI's similar crazy system. It seems if you go by relative performance in your number system, people complain, but if you go by new "families" like Nvidia, people complain too.
Wow. Let's run a bunch of single-threaded benchmarks to show how a faster CPU beats a slower CPU!
BRILLIANCE!
Why not run tests for F@H or Distributed.net and show what beats what?
Oh yes. The mythical "ordinary load".
This article may as well have just run 3dMark and called it a day, their results being almost as meaningful.
Chas - The one, the only.
THANK GOD!!!
Lots of bar graphs in the article, but no mention anywhere of the prices of the various Core 2 CPUs, which is surely important (especially if the chip has the word "Extreme" in its name)? It made the conclusion page fairly lame - most people would want to know which CPU offers the best bang for your buck, but we're given no clue about this metric and therefore are just as confused as to which Core 2 CPU to buy!
I do remember quite fondly when I purchased a 386 33 MHz, and it was quite obvious how much faster a 386 66 MHz chip would be. Then there was the 486 line, and the DX2/DX4 monikers to indicate internal clock double or quadruple. Not a big deal. Then it got into Pentiums and things really got confusing, because they released the Celeron, then Pentium M, then a whole bunch of other chips since then that don't seem to give you a good understanding of exactly how much horsepower you're purchasing.
I just purchased a new desktop PC and wow, the number of options is bewildering even to me, and I have a degree in computer engineering!
"I have never let my schooling interfere with my education." - Mark Twain
Super PI measures real world performance?
I think their benchmarks give too much weight to the quad core processors. It's still the case that most applications that people use don't really use more than two cores at once.