Will 4x4 b able to beat it?
This is an interesting question if you think about it for a sec. To answer this first one shoud answer this question : "Will Intel's quad core be FSB bottelnecked or will they manage to squize something out of the old FSB yet again ?"
The memory presure of four cores accessing the memory through one point only might be a little to much this time. On the ohther side AMD already has better memory tranfer rates usiong AMD2 than Intel. Using two processors with two memory HT links and on die memory controlers 4x4 will see an inpromvement in memory bandwidth tha's for sure. But higher badwidth doesn't guarantee higher performance especialy on high memory latency. To have higher performance with using high latency memory you either need a dam good prefech mecanism that doesn't miss or a rather large cache to hide the latency.
As you see I havent actualy aswered your question, but what I think it will happen is that performance will be higly benchmark sensitive in the future.
everyone mentions the FSB is getting old for intel and they dont really understand the real bottle neck. AMD uses the FSB also. go read about it and stop posting this stuff. infact now that you can see intels bench scores using their fsb
http://www.xtremesystems.org/forums/showthread.php?p=1527913
how does this compare to amd's 4x4 benchmarks? please explain these bench mark scores to me and how the fsb bottlenecked these scores.
if you really look at how amd and intel use the FSB you will see they hit the same exact bottleneck but just have a different path to get to the bottleneck
FSB DOES offer LOWER memory thoughoutput comapred to AMD memory aceess infrastructure. The problem is that current AMD processors can't hide DDR2 latency (lack of and advanced memory prefetc + rearanged access pattern) and also can't chew up all that data because of lower pipeline parallelism comapred to Conroe. So basicaly AMD right now can't take full advantage of an superior architecture and it's real world performance is below FSB's performance because of processor architecture only. At these speeds and high cost of non sequencial memory access you realy don't want any cache misses, and this is conroe's strong part for DDR2 memory. These are the problems for wich the K8 architecture wasn't projected. All those issues will be solved using K8L architecture, and more.
I do understand that Conroe is an amazing architecture to be able to squize every bit of performance using an extended array of tricks and techologies.
All of this comes from Intel's early adoption of DDR2 wich gave them some extra R&D time to solve all of the problems related to high bandwidth and also high latency memory.[/code]
so if as you state fsb does lower memory latency and amd doesnt do prefetch and AMD right now can't take full advantage of an superior architecture and the reason amd is behind FSB right now is because of proc architecture.
1. what is superior architecture about amd?
2. what is amd changing in the newer processors to address these issues?
A. making a prefecther? B. changing the proc architecture? C. more pipeline parallelism?
3. I would not claim that early adoption of DDR2 is the reason conroe dominates. you said it yourself FSB is slower than amd's superior technology.
4. you never figured out or answered my question how can a "Bottlenecked FSB out perform amd's "Superior architecture"
the answer is right infront of you. if i see that you try to answer it then i will tell you what it is
1. AMD architecture is superior because it alows even with curent k8 processor a higher memory throughoutput an lover latencies du to the integrated memory controler. But this isn't enough to offer superior performace comapred to Conroe because this simply is a too big o bite to swalow for K8. K8 isn't memory bottlenecked, nether C2D isn't memory
bottlenecked.
But in the case of quad core the perf increase from doa core is 54% as shown from aforementioned tests on extremesystems. The dual core increase from sigle core was 80%. This is where the FSB bottleneck begins to show up.
2. A, C and a little of B
3. That's exactly the reason C2D dominates it was build to be ablea to chew up high bandwidth with low latency. All the architectural improvements in C2D were for this purpose.
4. AMD K8 even if they had used FSB wouln't had been bottlenecked by it until quad core arival. The problem is that K8 is from DDR era while Core 2 Duo is a DDR2 era prpocessor this is were the performance diference comes from.
How can you explain C2 Duo's performance increase over Netburst having the same bandwidth available ?
imagine all the cores on your system using the same cache and the cache is huge. then add the ability to do more than one instruction per core (and there are 4 of them). if you look at the bench data it is pretty much tied to the number of cores. although C2D did this and made it power efficient. now do you see the bottle neck was never the FSB?
also think about this
with an amd proc if a needed page of memory is not in the proc cache where does the memory controller get the page from? in this situation wouldnt you want a larger proc cache? so even though the imc in amd is good shouldnt you always want to get your data from the proc cache? intel has eliminated the advantage amd used to have by getting a great prefetch with a large proc cache. amd simply needs to work on this
Some benckmarks simply don't put any memory preasure on the processor, and if you run the same aplication on all 4 cores you will definatly have alot of cache hits and the bottleneck is practicaly hidden.
As always these tests are made to show the procesor in a very good light and hide any posible bottlenecks.
But when you run difent memory hungry aplications on each core, I can bet my money (all of them) that Intel's quad core will be higly bottlenecked by FSB, and AMDs 4x4 or quad core wont be!
Intel's prefetch technologies will counter that quite well, you obviously have little knowledge on how software is being executed on the Core 2's wait till the recompiled software starts to show up.