Anyone with a Phenom X4 could do these tests... and more. The reason: in Phenom all the cores have same priorities and same connections to the outside world (unlike Core 2 Quad, for example). Method 1: start the benchmark, prepare, set affinity in Task Manager to cores 0, 1 and 2; do the benchmark. Method 2: start small program consisting of an empty loop, set its affinity in Task Manager to core 3 (or any other) and priority to realtime or high; do the benchmark.

THG did some testing of one dual-core Opteron and one single-core Opteron, both in a dual-socket mainboard some time ago. AFAIR, besides obvious lack of symmetry and problems with different instruction sets everything went OK.

The only problem is when an application thinks that the only possible core number is 2^n. If the application properly counts the available cores (and it's not that difficult), and properly divides the work to available computational resources (and this one is not trivial), no problems should arise.

As with all models, the price will tell if Phenoms X2, X3 and/or X4 are good buys. I have low-end Athlon X2 in my less-used desktop, so upgrading to any of these would be a fine choice, and I will probably try and get an X4 before Xmas.

As for the performance, it's almost certain that X3 performance will be better than 3/4 of X4 performance. One reason is Amdahl's Law and the other is the L3 cache, as it is shared between the active cores. That's why Core 2 has its smaller scalability from one core to two on Duo's (shared cache), and quite good scalability from two cores to four (from Duo to Quad, separate caches).

My opinion on Phenom X[2|3|4]'s performance is that AMD needs to make an option in BIOS to enable/disable current behaviour of checking L3 _before_ starting memory access. This is preferred way of doing things in most servers and HPC machines, but in many cases the low latency of concurrent starting both (and just discarding the memory result on L3 hit) could be much faster and would give back the great latency Athlons had. Intel did the same thing with HT and advanced prefetch options; while both were great in some cases, they were slowing systems down in others. Making them an option at the discretion of user was great thing to do.

What sucks is the fact that the 65nm Phenom X2 is just 10% faster than the 65nm X2. Its not all, if you go over the reviews comparing the 65nm X2 with the 90nm parts, you would find that 90nm X2's were 10% faster than the 65nm X2's on average. This simply means that the great Phenoms are just as fast as the 90nm X2's.
What a strategy!! Drag the better CPU's away from the market, substitute them with slower 65nm parts, wait for a year till people forget the 90nm parts (public memory is short you know) and then release the new core killer that's (oh yes) just as fast as the older 90nm cpu's.
Hey- Hector and Dirk! Guys, you work for AMD, not Intel, remember?

Anyone else wonder if there will be issues with 3-way SMP because of the OS? I mean, 3-way SMP hasn't existed up until now on x86 platforms AFAIK.

On another note, no wonder Phenom X4 has such a poor showing vs. Core 2 Quad. Look how close the Phenom X2 and Athlon X2 benchs are... wasn't Phenom have some serious IPC improvments, especially with SSE instructions? Those cinebench results don't look good. Maybe AMD can get it right with the next revision, but I won't hold my breath.

Um, holy crap but those benches actually make the Phenom X3 look damn good. Adding a third core show a better improvement, percentage wise, than adding a fourth core in all three of those benches.

Gord, imagine the irony of AMD spending all their time and money on quad core to end up having tri core being a more compelling product.

Edit: These benches officially do not warrant a "holy crap" response and my initial post was way out of line.

I was expecting an eastern techsite to review one first, but a German site? Could this be the beginning of something new?

i'm more interested in an X3 vs a Core 2 duo at the same clock speed. If AMD prices them right (i'm sure intel would respond w/ pricing) they might save a tad bit of their crushed face.

If AMD can fix their clock speed issues with their Barcelonas and Phenoms and whatever is coming down the line, and they do it soon - within the next three months say - then things might start looking better for them. They really do need to get 3 GHz or above CPUs being made.

Then they've got the 45nm hump to get over - I don't see AMD launching 45nm until late 2008 at this rate, unless their 65nm clockspeed issues are simply because they are concentrating mostly on their 45nm process in order to get it done earlier.

The tri-core concept is nice if they have a significant number of quad-cores they can rebrand down to tri-core due to having a single dead die. Alternatively maybe having a tri-core design is worthwhile in terms of die size savings, but I don't think they'll be going this route.

As long as they get clock speeds up around 3.0Ghz and keep prices very low these will be an OK upgrade for any AM2 motherboard owner. More performance cheap can never be frowned upon. Lets resign ourselves to the fact that AMD has about a year and another processor change before it has a chance to catch Intel. Meanwhile they can only compete with pricing.

Geez, following AMD lately is like reading about Britney Spears, i.e., what's the newest bad news?! Their 3800 cards have been a bit of a ray of hope, but now with the prices for the 3870 up to $269 at newegg and I guess elsewhere, their price competitiveness with the 8800GT has been lost. Plus the 3870s are as scarce as the 8800GTs. Happy holidays.