Slashdot Mirror
Intel vs. AMD - Today's Generation Compared
Bender writes "The Tech Report compares 15 Core 2 and Athlon 64 processors from Intel and AMD — from sub-$200 to a cool grand, from slower dual cores to fast quad cores — in 32 & 64-bit apps in Windows Vista, including the new, multithreaded RTS game Supreme Commander. 'The release of Windows Vista and a round of price cuts by AMD prompted us to hatch a devious plan involving Vista, a new test suite full of multithreaded and 64-bit applications, fifteen different CPU configurations, and countless hours of lab testing. That plan has come to fruition in the form of a broad-based comparison of the latest processors from AMD and Intel... from the lowly Athlon 64 X2 4400+ and Core 2 Duo E6300 to the astounding Athlon 64 FX-74 and Core 2 Extreme QX6700.' Folding@Home in Linux, power use, and energy efficiency are tested, too."
Despite the FAH PS3 client has been out under 24h the PS3 client performance is overtaking all the CPU/GPU FAH clients combined!= osstats
http://fah-web.stanford.edu/cgi-bin/main.py?qtype
http://folding.stanford.edu/FAQ-PS3.html
I disagree with their definition of "low end." Maybe low end as far as what they tested, but there are a lot of non-X2 Athlon 64s and Pentium/Celeron Ds being sold. At the true low end, AMD is still more than competitive. It's only when you near the most-horsepower-per-dollar peak that Intel really pulls away (and that's where they seem to start measuring here). It's worth noting that I have no dying love of AMD. I have two AMD processors and one Intel processor running in my current personal machines and plan to get a Core 2 as soon as the next significant price drop occurs.
Nice reading.
But of course conclusions are not that surprising. AMD is 10+ times smaller than Intel (judged by capitalization). Intel has many fabs - while AMD is constantly struggling expanding its production capacities.
Yet, AMD (with Athlon 64) had managed to pull quite a match against Intel. Kudos to AMD: without you Intel's CPUs for sure would have costed $2500 a piece.
All hope abandon ye who enter here.
Most of these benchmarks are targeted towards unified caches.. (Intel)
Meanwhile real world apps favor separate caches per core.
(Where one user app isn't flushing cache entries of another app executing on different core.)
If they wanted to make it fair..
They should execute n-copies of each benchmark compiled separately using different module names. (no unified cache sharing.)
Next item.. Graphics & games. What are they really measuring?
The ability of some device driver writer to take advantage of some esoteric CPU optimization?
Last item they disabled Cool and Quiet on over clocked AMD configuration s it should have never been published.. I.E. They're simulating certain AMD configurations and aren't testing the real thing..
What gets to me is the way that most reviewers compare power usage by simply comparing the listed thermal envelopes. AMD lists the maximum power used, whereas intel lists the typical power used. Furthermore, for laptops and other machines where heat is your big concern, you do care a lot about the loaded maximum power used. However, for most desktops in which heat is not really an issue, you're more concerned about the cost of the electricity you burn. In that case it's almost more relevant to measure the idle power usage, since most desktops sit around doing nothing most of the time. Any good review should actually measure the system power between wall socket and PSU, otherwise it's not really infromative to the actual concerns of the user.
TR's benches clearly show that someone working primarily with POV-Ray would get better performance for $599 with AMD than for $999
Ok, let's be realistic here. Does anyone use POV ray for anything other than processor benchmarks? I have yet to see one real production, student or otherwise, rendered with POVRay. Let's see benchmarks with PRMan and Mental Ray. Those are production renderers people are actually using.
Why, so you can wait around longer for stuff to happen? Unless Apple does a major overhaul of their kernel (ie: get rid of that primitive microkernel), Windows (and Linux even more so) will beat OS X hands down in any multithreaded application benchmark. Sad for the zealots, but true nonetheless.
We've found after extensive testing that the Xeon line, the 5000 series, and specifically the 5150@2.66, are several percentage points slower than the Core 2 line of processors. The Core 2 Duo 2.66 is faster than the 5150 2.66 processor. So buying the Xeon processors apparently only gets you SMP capability for the higher price(?)
What effect the L2 cache on the Intel chips have on the numbers.
III.IIVIVIXIIVIVIIIVVIIIIXVIIIXIIIIIIIIVIIIIVVIII
Now, that said you need a good caching strategy - there is a LOT of effort working on caching the correct memory to have it available quickly to the processor. I don't know where you get that one active process will flush the cache at the expense of another application that also needs a heavily used memory location... YMMV
I have mod points and I am not afraid to use them
My comment addresses yours, but wanders, so apologies in advance.
I use POVRay for explaining engineering concepts to co-workers, in papers for external viewing, and for use by the marketing folks. Word on the street is that its renders find their way into publications such as Nature and Science.
You talk of "production," which sounds like "movie," but it isn't a chump app just because movies don't use it for their render engine. It's a free, not-too-restricted-source-code app that yields stunning results. (Side-note: PRMan is ~$1k, and I couldn't find a price on Mental Ray, but my attitude towards Autodesk is that of the indentured slave to his master. But I digress.)
It is the single most time-consuming task I do, doing a render, and about everything else runs "fast enough" for me on my three-year-old mid-grade P4 system. Sometimes I wish for greater performance with an Octave/Matlab script when I am playing around. But that's hard to benchmark & compare on systems and only rarely does it take the same order of magnitude of time that POVRay can take.
So in response to another, I am one (and likely not the only) person who would lean towards a system based upon its POVRay performance. I have just been overjoyed that this has started to be used for benchmarks. I personally find the frames-per-second on Doom4 benchmarks useless, but what it comes down to is one thing:
The more apps people benchmark (accurately), the more people benefit and can make informed decisions that address their specific situations.
So I would like to see the apps you mention benchmarked, too. CompUSA never let me install photoshop and run my personal tests on it back when Photoshop was important to me. And now the greatest computer selection is from online retailers; how will you compare the value of a computer for you? $3000 for a Core2Duo Extreme Quad SSE2 from Dell, $2800 for an AMD Dual Quad HyperTransport blah, blah, from HP, and a bare-bones system from somebody else? What is each worth to you? And performance, obviously, gets far more complicated when you move beyond the processor isolated in a system. Is it worth the extra $100 for me to get another 1GB of memory? Will I ever really know how that will effect my apps, or does it just come down to whether or not I am willing to hand over another $100 just in case it might help?
I guess that is it: What are people's expectations for spending their money? Those who look at benchmarks might just find the "best" and drop their cash (or credit) on that one. Those who don't look at benchmarks might hit a price point, and just grab the best-looking system or find something from a specific brand. People like myself who want to optimize on a personally-important criteria are mostly left to guess and always be uncomfortable with any choice they make. And this situation might only change when spending the $1000 (or whatever) is worth caring a great deal about.
Apparently you've never put together a low-power system, because you don't just try random combinations and see what the end result is... you look at individual components and find low-power combinations that work well together.