1. 4x4 will be almost practicle as s940 Opteron 2xx. The 4x4 is capable to work only with FX series and they are expencive. The mainboard for the 4x4 will be much more expencive and at least 4 modules of RAM are required. 4x4 will perform competetive to Kentsfield system, but will not beat it and will cost more.
2. The FSB will bottleneck Kentsfield, but not that bad. Even with the bottlenecked FSB, it still outperforms same clocked quadcore AMD K8 system.
3. 4x4 will scale more linear in performance than Kentsfield for multithreaded apps with the NUMA model. But it is not quite good as native quadcore where all cores are connected to the same crossbar. The adventage of two dualcore K8 CPUs are the two ODMCs, improving the total RAM bandwidth.
Kentsfield is glued using "the old" Pentium-D technique of glueing singlecores.
4. No, Deerhound will perform better.
4x4 is not restricted to FX. It is not possible to restrict it. FX has one HT link, X2 has one HT link. For any chip they just need to turn on an extra link.
There have been closed system demos where it was said to be 80% faster than one dual core.
Kentsfield will be fast but it is said it will use 125W+ so 4x4 can compete power wise. K8L will kill them both though, BUT FX is supposed to get the K8L core at the end of next year, maybe sooner.
Ok.. Baron.. you need to do some reading outside of the inquirer and Sharikou PHd.
Tech Report: AMD Senior VP on 4x4
Tech Report: AMD FX 4x4 CPUs to sell under a grand
Regardless, the idea behind 4x4 is a good one: to give enthusiasts a crack at getting to larger numbers of CPU cores sooner by fostering the development of dual-socket motherboards with tweakability and a strong enthusiast feature set. One of our major concerns about these plans, however, was AMD's stated intention to tie the 4x4 initiative to pricey Athlon 64 FX processors. Even after AMD's recent and dramatic price cuts, requiring a pair of Athlon 64 FX processors would make the price of entry for 4x4 over $1600—for the CPUs alone. That would make 4x4 a high-priced, low-volume stunt—an "image product" with a very limited customer base—rather than a compelling reason for PC hobbyists to turn their attention from Intel's excellent new Core 2 processors.
Fortunately, AMD has listened to feedback from the enthusiast community on this issue, and has pledged to sell a range of two-CPU bundles for 4x4 with price tags that extend to "well under $1000." Beyond that, many of the details of AMD's 4x4 plans are either undetermined or not yet ready for prime time. All of these processors will carry the Athlon 64 FX brand name, but AMD was not willing to discuss details like clock frequencies or L2 cache sizes for these new FX chips. We asked about the prospects for 4x4 capable versions of AMD's Energy Efficent Athlon 64 X2 CPUs, and AMD's reps said only that they'd consider the possibility.
In addition to aggressively priced two-processor bundles, the company is mulling over how to handle customers who may want to buy one processor for a 4x4 system up front and purchase a second one later as an upgrade.
All during an AMD demonstration of 4x4 technology.. something Sharikou PHd and The Inquirer never saw. Tech Report did and AMD answered many of there questions during a Q & A period after the demo.
Again.. stop posting FUD. So as of now, you can only get 4x4 with newer CPU's (90nm AM2's won't work as like with any SMP, the processors need to be of the same design often same core revision). Also these CPU's will come in a bundle and carry the AthlonFX name. So anyone else telling you different is spreading AMD Fanboi FUD.
STFU. I said exactly what the link said. I read the demo article with Rahul Sood and it was said that it used ENGINEERING SAMPLES with HT links turned on.
What they brand them has nothing to do with the fact that they will all (X2 OR FX) need to have links turned on. But with the Newegg prices for Opteron 270, they can just disable the ECC pin on some of those while turning on links in some X2\FX.
But they have said (maybe you should visit voooopc.blogspot.com) that they (AMD) don't expect this to be a volume proposition. If they can get full systems down to under $1500 with onboard video - definitely a possibility - I will send emails to every dev I know. Kentsfield will be good but it won't have quad channel RAM. Rahul has quoted that the system will use separate RAM banks for each proc.
Maybe you should read the VoodooPC blog, dufus.
I don't need to read someone's blog moron..lol. I don't need to read other people's opinions. I prefer and am able to interpret the data on my own. And I said nothing to counter any of these new arguments you're putting forth.
Yes.. look at the puuurty picture. It tells you everything you need to now about 4x4's memory subsystem as well as it's HT Links. Notice how each processor has independent access to the RAM. Yes, that's right.. Each processor has it's own Dual channel dedicated bandwidth. (I've said nothing to the contrary).
This is good and bad. It's good because it gives each Processor dedicated bandwidth and as a whole platform it gives it Quad Channel memory. It's bad because you now need 4 sticks of RAM to get Quad Channel.
Another reason why it's not so swell is because The K8 design does not reap huge benefits from a higher memory bandwidth. It's more prone to greater benefits in lower latencies. Mind you, more bandwidth enables the AM2 level processors to match the performance level of socket 939 processors this has more to do with more bandwidth compensating for the higher latency of DDR-II memory over DDR memory. the scaling stops at Socket 939 performance levels.
So, yes.. quadfather's higher memory bandwidth will enable it to win a few benchmarks such as Sisoft Sandra mem test and Everest mem test... and possibly a few other synthetic tests. But on the whole Real world performance will not increase (gaming primarily will not see an increase).
You might be able to re-iterate what others say. But formulating your own conclusions is something totally different. This is why I don't fall prey to Sharikou PHd or other FUD.
There's no denying that AMD's architecture has greater room for improvements and when you get down to it is far more advanced on the bus/memory side of things. But they lack something. AMD's Cache technology is like something out of the 90's It's outdated and they need a new Cache technology. They also need more efficiency per clock and a better process (SOI is not as good it seems as Intel's newer processes and doesn't scale as easily as AMD is already running into problems).
Also there's the threat of CSI and an IMC coming in part in mid-2007 with Tigerton (the CSI) and then released on Itanium and Xeon in 2008 we'll see the IMC and CSI working together.
So Intel have bought themselves a year. And with K8L desktop parts only launching in the 3rd quarter of 2007 (Opteron server design in the half way mark of 2007).. it's pretty close to Intel's release.
This is why things will remain quite competitive. AMD and Intel fanboi's alike are wrong.. neither of the two will be the dominant player. Expect Intel to lose more market share but still retain around 60-70% while AMD will be going back into the red because they will be losing design wins to Intel's Woodcrest and Core 2 Duo Processors as well as the HUGE, AGRESSIVE price cuts they've undertaken. Also they're acquiring ATi and we all know it's going to hit there revenue as they need to borrow part of the funds to make the acquisition.
Facebook
Twitter
Instagram