I have read the article and the performance difference between Dual and Single channel memory is... almost none?
So, would be possible to AMD develope a dual core Athlon 64 for ppl who want to keep their socket 754 systems (like me)? Or there's another problem such eletric current and etc?
Since AGP is still enough for a mid-range VGA and support DX10, a Dual Core CPU for this platform would be great, wouldn't?
Ok, it lacks from PCI-E and etc, but HT800 is very fast, fast enough to keep a dual core CPU, isn't?

Anything's possible, really. I really doubt a product like that will come out, though. socket 754 is "dead" from the corporate view, and that's a pretty specific product; the market demand won't be great for it. not much incentive to dump money into R&D to make something that won't sell in great volume or be cutting edge. would be nice though, huh? i've got a friend who has a 754 that could use a cpu upgrade.

Possible, yes. Viable no.
In the processor world, the most money is made in the high end chips. When AMD went to socket AM2 754/939 were forgotten, as the cost of R&D for the old platform was restrictive.
So no, there won't be a 754 dual core. Sorry.

I knew the market demand wouldn't be enough to make AMD invest in a s754 dual core processor. But that would be nice =D
I just wondered... AMD simply moved from one socket to another in a couple of moths. From the Intel side, almost every decent LGA775 mobo can handle a budget dual core (I think). The only socket from AMD that doen't support Dual Core processors is the 754.

Man... now I realized something: it's simply impossible to run Crysis in a S754 system, if the system requirements are really "requirements". Or better, with almost any single core.
Single core: 2.8Ghz (XP) 3.2Ghz (Vista)
Dual Core: 2..0 or 2.2 Ghz, I don't remember (XP)
Well... man... that's a game or a monster?
I think they took the Pentium D as refference. Otherwise, I couldn't run Crysis in a C2D E6300 or similar, that's known as a very decent processor.

AMD and it's IMC make it impossible to be dual channel and still s754, while Intel s478 has dual channel and even DDR2 capabilities, but still only a CoreDuo from the mobile market that is very inappropriate in performance/price from a completely new system is available for it. And I am not even sure of the CoreDuo thing as I haven't seen one with my eyes, just advertisements and unreliable sources. Worth a question though: Has anybody seen a s478 with 2 cores?

Intel's S478 is to AMD's S462 as Intel's LGA775 is to AMD's S754/939/940/AM2, from a technological point of view. I know that S754 haven't dual channel support, but in this article, THG almost proves that there's just a little real world performance gain using dual channel memory, almost insignificant. Single channel memory is enough (at least, DDR2). For DDR, even with lower clock speed, it's lower latencies would be enough to supplu this difference, I think. When AMD moved from DDR to DDR2, there were no performance gains coz DDR2 latncies were 2x the DDR latencies.
Wow... what if you double the speed and souble the latency? You get nothing!

I'm not trying to make AMD creat a S754 dual core CPU. Well... I'm happy with my Single-core setup, and I can't upgrade it very much. I bought it when S754 was a great platform. So, would be great if they take two Venice and put them togheter to make a simple dual core for us, users of S754 platform. I don't know how to say that in english, of course. If I knew, I would simply do that.
But see...

Intel: S478 > LGA775
AMD: S462 > S754/S940 > S939 > AM2
And now, just because AMD got crazy, I'm stuck in a single core configuration, in a platform that would be totally acceptable as a Dual Core platform.
It only lacks of PCI-E support.

[quotemsg=1753574,6,58658]Intel's S478 is to AMD's S462 as Intel's LGA775 is to AMD's S754/939/940/AM2, from a technological point of view.[/quote]

Uh, perhaps a little more clarification:

Socket 462 ~ Socket 370/423/478
Dual Socket 462 ~ Dual Socket 370/Dual Socket 603
Socket 563 ~ Socket 370/478
Socket 754 (desktop) ~ Socket 478
Socket 754 (mobile) ~ Socket 479
Single Socket 940 ~ Socket 478/775
Dual/Quad/8 Socket 940 ~ Socket 603/604
Socket 939 ~ Socket 775
Socket AM2 ~ Socket 775
Dual Socket F ~ Dual socket 771
Quad/8 socket F ~ Quad+ socket 604/771 (Tigerton)
Socket S1 ~ Socket M/P
Socket AM2+ ~ Socket 775/upcoming socket 715
Socket 1207+ ~ Socket 771/Socket 1366 (Nehalem DP/MP socket, aka Socket B)

Note that they tested Intel CPUs with DDR2-800 rather than AMD CPUs with DDR-400. DDR2-800 is a little less than twice as fast as DDR-400 and the huge caches and prefetching of the Core 2 CPUs mitigate the reduced RAM bandwidth. Basically, the article was to give a memory management guide to 4 MB or 2x4 MB Core 2 owners. It didn't really see what would cause the performance to really drop off nor look at any AMD CPUs. Here's what they should have done:

1. Used DDR2-667, DDR2-533, and DDR2-400 RAM in addition to the DDR2-800 and seen how much bandwidth the CPUs really needed before their performance started to drop off precipitously.

2. Tested the 2 MB versions of the Core 2 Duo and the 1 MB Pentium Dual Cores alongside the 4 MB versions to judge the performance increase of the cache with reduced memory bandwidth, where it would arguably be more important. Either AT or THG did this with dual-channel DDR2-800 and the chips all clocked at 1.8 GHz. They found performance already drops 5-7% when going from 4 MB L2 -> 2 MB L2 and around 20% when going from 4 MB -> 1 MB. I'd like to see it with single-channel memory, especially DDR2-667, 533, and 400.

3. Tested AMD processors with varying cache sizes: the 2x256 KB L2 Windsor X2 3600+, a 2x512 KB chip, and a 2x1 MB L2 Windsor. And test them with DDR2-667, 533, and 400 as well as 800.

That would give a good clue as to how RAM bandwidth and cache size affect performance. Simply yanking a stick out of a dual-channel setup on one kind of CPU is informative only for that speed of RAM and CPU they were using.

That's not possible to do. You would need to make an MCM and that would require giving each die its own access to the memory as well as having an HT link between the two dies, as well as having some way to control that. You'd need a socket with dual-channel RAM connections (one for each die's IMC) and also with the pins and a chipset that handles the cache-coherent HT needed to link the two dies. You could probably do it with a socket 940 or 1207 Opteron board as the pins and chipsets support the ccHT links and there are two memory channels wired to the socket. Socket 754 lacks the extra pins for the ccHT links and the chipsets wouldn't know what to do with it anyway. So you'd need a monolithic die with all core-core and core-IMC traffic handled on-die. Probably the easiest thing for AMD to do if they wanted to make a dual-core 754 chip would be to plop a Manchester/Toledo X2 die on the 754 package and only wire up one channel to the IMC.

A lack of PCIe is pretty critical in today's day and age. AGP is basically dead and PCI is heading that way. With parts being the prices they are today, you would do well to get a new board, CPU, and RAM rather than try to upgrade a 754/AGP system any further.

Sorry, I made a mistake
I wasn't reffering to wht Intel did with Pentium D. I was reffering to make an AMD style dual core using what would be two venice cores. I don't know any Athlon 64 X2 core names, so I tried to say in another way. The correct would be "put two venice in a single die", as they do, or be, to make a dual core die that works at 2.2Ghz and has 2x512KB of L2 cache, HT800, IMC, 90nm, SSE3 etc, just like a Venice, but dual core ._.

Sorry

Please, do not understand me as a critic, I'm more like a "wonderer".

Yes, man, you're right. Well, HT is not FSB. Maybe, since Intel is still using FSB, the performance hit wasn't so big as it would be in AMD processor. What the syntetics benchmarks show for me is: we need better processors, not faster memory modules and buses, for a while.