What is Digg?Sponsored by HowLifeWorks
New Food Sprinkle Convinces the Brain to Stop Over-Eating view!
howlifeworks.com - How sophisticated scents are helping people control their appetite and lose weight
55 Comments
- judicar, on 10/12/2007, -4/+32Captain Obvious : "I bet the 8-core Mac Pro is faster!"
*Captain Obvious flys away* - lieutenantmudd, on 10/12/2007, -6/+28And that program would be...
Pardon my skepticism, it's just that I don't believe you. - caliform, on 10/12/2007, -1/+22Now! With an exclamation mark behind all sentences because it's so fast! Wow!
- mikev, on 10/12/2007, -1/+19better than a "AMAZING PERFORMANCE SPEED. MUST SEE BENCHMARKS!!"
:P - adinb, on 10/12/2007, -4/+16@nreynolds:
*snarky*
If you don't even have the capacity to understand what "compute intensive" means, then please get out of the apple section. Yep, just deselect it in your Digg preferences.
*/snarky*
A "Compute intensive" task is a task that's usually bound by processor or "compute" resources. In parallel computing it's important to identify the largest bottlenecks holding back performance of a task -- if its disk or communications bound, then we'd say that the task was disk/memory/communications intensive.
And its very easy to use up all the resources of a 8-core machine in a highly parallel task. In my CS undergrad days at UCF we got to play with all kinds of parallel architectures, from a SIMD toroidal mesh MASPAR (8K processors, good for communications intensive fine grain parallel tasks, like correcting Hubble images before we got the new lens on it) to a Hypercube, BBN Butterfly, HP SMP machine, and a PVM "machine" (of sparcs) with its own dedicated "high speed" backbone. Point being is that there are plenty of tasks for an octocore that could see 100% speedup from a 4 core machine...the task just needs to be large grain/compute bound instead of communications bound.
BTW, MasPars **rocked**. I loved being able to init a 90x90 array in one op and then being able to do all kinds of matrix math in a single step. It actually made shortest path problems *fun*. ::wistful sigh:: - Fedge, on 10/12/2007, -5/+16@nreynolds
compute (verb): to determine especially by mathematical means
You know, like an application that does a lot of computations. It's not a big stretch there guys...
Matlab, renderman, video-encoder... use you imagination. - JavertHolmes, on 10/12/2007, -1/+10I find my urethra just as lifesaving.
Clearly my original post was offensive to the low energy efficiency community. My apologies go out to them. - waterdrop, on 10/12/2007, -3/+11Maybe because Intel's Core 2 Duo series is faster than anything AMD offers currently.
- meatmcguffin, on 10/12/2007, -3/+9Surely that should be the other way around?
- Goosemaster, on 10/12/2007, -1/+5Basically this thing kicks ass for easily-threadable tasks....
I can see why people are still praising AMD's Hypertransport now..... - geminitojanus, on 10/12/2007, -0/+4"not accroding to this test"
1: those tests aren't showing top speed, they're showing performance per dollar.
2: those tests were published /before/ Intel made their own price cut to counteract AMD's price cut, effectively making AMD's chips look $40-80 more favorable. The 6400, a Conroe (4MB L2) core instead of the Allendale (2MB L2) 6300 would be a more fair pairing (and the two chips should perform almost identically, except the Intel chip is running at 600+MHz less than the AMD processor, and putting off about half the heat).
Also, don't you think it's a bit sad that AMD's 2.8GHz X2 actually /lost/ or scored evenly in a few of those tests to a 1.8GHz Allendale? - neonenergy, on 10/12/2007, -1/+5Uh longbow, core 2 duo really is faster than AMD's chips...
- Francky, on 10/12/2007, -0/+4I have no Idea why streak was digged down so much, as I have come to the same conclusion here too.
Running 3D applications that uses 100% of the processors when rendering obviously run twice as fast. - inactive, on 10/12/2007, -2/+5@joshuapechter
Double the cores does not mean it is necessarily double the speed. The program must be inherently concurrent (multi-threaded) to take advantage of multiple processors. Additional overhead may be required to manage the threads, depending on how independent they are. And of course all this depends on the OS threading model and whether other applications are using some of the cores. So for an application to come close to doubling speed on 8 core compared to 4 core, it must be multi-threaded (more than 8 concurrently), and extremely computationally intense, enough such that processing speed is the bottleneck, not hard drive reads or something else, as the original poster was saying.
This is what I understand from my concurrency class, let me know if anything is incorrect. - sintaxi, on 10/12/2007, -2/+5Its a good Idea to benchmark using common games. I'm sure we could go back two years and find machines that were bechmarked with those games. This makes for good comparison (hence "benchmarking").
- streak, on 10/12/2007, -0/+3In servers, Core 2 has memory *access* problems that can arise even with 4 cores. In order to provide more DIMM slots to achieve larger memory configurations in servers, FB-DIMMs (fully buffered DIMMs) are required. FB-DIMMs are outfitted with an intermediary "AMB" (Advanced Memory Buffer) chip that sits between the memory bus and the memory chips. The AMB introduces additional wait state(s) to each memory access, and also causes server DIMMs to consume twice the power of the unbuffered DIMMs found in iMacs, MacBooks, and MacBook Pros.
- streak, on 10/12/2007, -1/+3@lieutenantmudd and others can continue to disbelieve me. I can't say I blame you all. Because the media only has eyes for flashy benchmarks that they think they and the public will understand. But there is indeed at least one important application area, to which I refer, which is not as flashy as multimedia and gaming, where the 8 cores do shine over 4, running virtually twice as fast, using multithreaded applications that work efficiently on "embarassingly parallel" problems. Not on all problem sizes in this application area, but on all of the most commonly faced ones. Sorry I can't be more specific. That doesn't change the fact that I am one very happy camper with an 8 core (and multiple 4 cores). :-)))))
- adinb, on 10/12/2007, -1/+3Why bother with games at all? If they're running single threaded, you're not going to see much speedup, no matter how many cores you have. The additional cores will just let you run photoshop and FF at the same time as your game.
When games have been modded to take advantage of multiple processors, *then* I'll be interested to see how they benchmark, but only mildly. Additional graphics cards will have *much* more impact on game performance for the near future. - thehouse, on 10/12/2007, -0/+2I tend to agree with the threading of applications statement... I suspect that determining how wide to spread treads is a difficult, especially when you want to deliver reasonable performance for several architectures. Now, if the application was built specifically for the given configuration...
- AllenHSmilden, on 10/12/2007, -0/+2streak is obviously referring to mame with its brand new multi-threading support. :)
- blackwhitecolor, on 10/12/2007, -0/+1This isn't exactly new news. The first review of the Mac Pro 8-core went up a fortnight ago: http://www.digitmag.co.uk/reviews/index.cfm?ReviewID=738 and it compares the 8-core Mac not only to the 4-core, but an 8-core PC too. The review only uses the CS2 version of After Effects, but there's some mention of how AE CS3 works on the 8-core Mac in their beta preview of it: http://www.digitmag.co.uk/news/index.cfm?NewsID=7792
- sonofagunn, on 10/12/2007, -1/+2I know why Streak is dugg down - because he is being secretive about what program he's using. But he has a good point. Truly parallel problems should theoretically run almost twice as fast - but very few problems are 100% parallel.
On a related note, go to www.wikipedia.com and look up "Amdahl's Law" for the math behind how much faster problems can be solved in parallel by adding more processors. - inactive, on 10/12/2007, -2/+3wait until leopard
- Felshadow, on 10/12/2007, -0/+18 - 4 = 4 more cores.
and i quote "Dug down for utter fail." - ArmchairAthlete, on 10/12/2007, -0/+1@joshua
You can add as many processors as you want and it won't help a single-threaded test. A good test (though not an easy one to do) would be something like at least 8 processes that want time on the CPU(s). Or maybe some kind of instruction level parallelism. I'm guessing a lot of these individual tests aren't telling the entire story about what it's like when many processes at once are going.
Only parts of any given program/application are going to be parellelized.
Learn about parallel processing:
http://www-static.cc.gatech.edu/classes/AY2007/cs2200_spring/MPslides/CS2200_14_ParallelProcessing.ppt
http://www-static.cc.gatech.edu/classes/AY2007/cs2200_spring/textbook/Chapter12Rev08.pdf - vertinox, on 10/12/2007, -0/+1Most 3d graphics rendering have a neat feature that leads people to call them...
http://en.wikipedia.org/wiki/Embarrassingly_parallel
Hence, 4 to 8 cores will increase power by double. Applications that don't parallel well don't. - M05H3, on 10/12/2007, -0/+1For the first time ever, hardware is ahead of software.
- JavertHolmes, on 10/12/2007, -3/+4What's the performance increase:wattage increase ratio? A number greater than 1 would be interesting.
- JaYBrooks, on 10/12/2007, -0/+1Don't the programs have to have instructions to use all of the cores or processors? It would seem like the machine is to new to really give it a good test.... wouldn't it?
- emceepecks, on 10/12/2007, -0/+1RAMBUS was terrible, and your point to support the idea of bringing it back is also terrible
- inactive, on 10/12/2007, -0/+1This still won't allow iTunes to tell you where the hell it is looking for a song file when it allegedly goes missing.
Oh wait, ***** software on eight cores is still ***** software. - peterinjapan, on 10/12/2007, -0/+0I have an Octocore and a Mac Pro 3.0 quad, and there's *not much* difference between them in speed for most tasks. However, since the 8-core Mac is running a bunch of processes in the background (including SpamSieve), it's nice to have the extra muscle. The other day, just to see what would happen, I converted some ripped DVDs to DIVX using D-Vision 3 while doing my important work for the day. With the "use dual processors" box checked in preferences, it cleanly output two files in tandem while I worked. There was no slowdown or change in the machine. Best of all, there was *no* fan noise or increase in heat. The machine is nearly perfectly silent compared to my Quad G5.
- Coniferous, on 10/12/2007, -7/+7@joshuapechter
I don't think its the memory thats the bottleneck, Most memory upgrades give extremely minimal gains. This wouldn't be true if it was the real bottleneck.
I personally believe that there just aren't enough well written multi threaded applications. Its not easy to do, so i don't blame developers. - snuf42, on 10/12/2007, -1/+1@coniferous
it isn't about how much memory you have or the rated speed of the chips it's about the memory bus (front-side bus) on the motherboard. There is a lot of speculation that the Core 2 chips will have problems at higher core numbers because they share the main data bus to memory on the motherboard. If you can't feed the chips data out of system memory because the bus is saturated you won't be able to get maximum performance from all cores. In contrast AMD chips do not use a shared bus with a on motherboard memory controller. Most likely Intel will move away from the shared bus in the future to eliminate this bottleneck. - TexanPsycho, on 10/12/2007, -1/+1CS3 sounds awesome.
- longbow486, on 10/12/2007, -3/+3you see, people are so quick to blame the software, when its truly the hardware at fault. intel is still useing their decade's old FSB archtecture. its time to move on and get something thats better, hmm lets give an example of.. oh and on Die memory controler. you get the full clock speed of the CPU, practically no latency, because the memory is talking directly to the CPU not; cpu talks to North chip, north chip talks to RAM, ram talks to NC, and NC talks to CPU. I have been saying this for years, it dont mean ***** when your CPU is 500x faster then the rest if you dont have a fast or wide enough buss to have it talk to the other devices.
- hoyanf, on 10/12/2007, -1/+0It's far from impressive to me as the speed is still inadequate and the bottelneck is still there not mentioning the probs with FBDIMM... Once was tested by anandtech that best utilization is limited to 4 slots of FBDIMM for performance and it'll go worse with 8 slot utilized... I would prefer if Intel would go back to old RAMBUS with on par bus speed than making new memory technology where it dont suffice... GranTurismo able to fully utilize full race length playback with car switch on the fly with RAMBUS, why cant INTEL...
- Kommy, on 10/12/2007, -2/+1OH WOW! YOU MEAN A COMPUTER WITH A SECOND PROCESSOR IS ACTUALLY FASTER?! WHAT MIND BLOWING RESULTS! Dug down for utter fail.
- joltjake, on 10/12/2007, -4/+3wow, so you're telling me that a mac with twice as many cores at twice the price performs better. OMG! I'll alert the media!
- Urusai, on 10/12/2007, -4/+2It's a Mac, they benchmark on the tiny handful of games available for Mac that might conceivably stress the system. No, Marathon doesn't count.
- Coniferous, on 10/12/2007, -3/+1@joshuapechter
I don't think its the memory thats the bottleneck, Most memory speed upgrades give extremely minimal gains. This wouldn't be true if it was the real bottleneck.
I personally believe that there just aren't enough well written multi threaded applications. Its not easy to do, so i don't blame developers. - inactive, on 10/12/2007, -5/+3It would be good for httpd.
- V1ncent, on 10/12/2007, -3/+1Most programs are not set up to take advantage of multiple cores. The multi-core processor is a new marketing scam to justify higher prices for chips that do not have a significant benefit to consumers. Like 64bit chips was earlier, now multi-core is the new scam.
- mehip, on 10/12/2007, -4/+1It all looked good to me until I got to the gaming section. Why are they benchmarking using "old" games like Doom 3 and Quake 4? I'd suspect any SLI system built in the last year would have pretty good fps numbers with these games.
- longbow486, on 10/12/2007, -4/+0@ Neonenergy
not accroding to this test
http://www.legitreviews.com/article/490/1/ - mehip, on 10/12/2007, -5/+1My point is, why not benchmark with modern games? Doom3 and Quake4 are hardly representative of what is currently on the market.
- FullHazard, on 10/12/2007, -7/+1Wow!
In related news, 8 is bigger than 4!
Later model processors are faster!
Things that cost more work better!
That's what I call SCIENCE, and i'm glad for the graphs, or I wouldn't believe it! - TheCash, on 10/12/2007, -7/+1Wait... it has twice as many cores, but only runs programs 25 percent faster and costs twice as much?
What kind of crazy business model does this make sense in? I think Apple has jumped the shark. - nreynolds, on 10/12/2007, -8/+1that colon there just saved your life, buddy.
- longbow486, on 10/12/2007, -8/+1mm yes, much faster when your memory bandwith is down to a blazing 350mb/s per core
-
Show 51 - 55 of 55 discussions
© Digg Inc. 2009
— Content created and posted by Digg users is