Well, didn’t Intel keep the 80% market share selling P4 and PD crap.
Didn’t people bought crap even knowing that it was crap. (Intel fanbolism)

So let people buy crap, however this crap is better crap than the Intel past crap.
In other words this is good crap, not bad crap like Intel.

Crap.

This AMD setup is crap. It's too expensive and has too much heat problems. I must say I am very disappointed at AMD. Being an AMD user since the 586, I have never seen such a sad attempt at competing with Intel.

Well, doesn't seem any more crap to me than Smithfields Intel touted as Holly grail just a year ago...

Also This is a Two-Socket PLATFORM! Meaning any one with a basic knowledge must not that it will suck power like no tomorow.

At the same time:
1/2pc . FX-72 + MB == $750
1pc . . FX-62 + MB == $1000

Gues what has better price/perf. ratio... Not to mention the I/O godness this ASUS thingie carries on board.

or anything with "Pentium" in the name ;)

This system tested was the 2x2. The 4x4 is dual quad.(err hence the name 4x4) That's why this system is called Quad FX. (ie. 4 in total in a 2+2 configuration.)

Facts are a good starting point.

The name "Quad FX" previously referred to as 4x4 represents four processing cores and four graphics cards. If you were correct, then AMD would not be able to release the Quad FX platform until late 2007 at the earliest. Although the system was only tested with one graphics card in this case, all of the tests relate to the performance of the underlying platform and not to the theoretical possibility of installing four graphics cards.

Facts are a good starting point.

XP 64 is not going to make the interleave problem go away. Only Vista has a NUMA aware scheduler, which will schedule processes to run on a the core which is closest to the memory where the process is stored.

Ultimately, AMDs approach is more scalable that Intel, because memory throughput increases with each CPU you add. You can see how bad the Intel FSB must be, by the fact that in the four-application test, that the QuadFX starts to perform heavily, and this is with the rather useless interleaved memory, and non-NUMA aware OS.

All this entire review shows is:
* A single process running on a single CPU with a single memory bus, is faster than a single thread running on a CPU with memory interleaved between fast local memory, and slower memory via another CPU somewhere is. Frankly this is not new.
* Two CPUs use more power than one.
* A single shared front-side bus sucks.

There should have been a test with Linux, which has a NUMA aware scheduler, and and putting the motherboard into full "NUMA mode". Or with Vista.

Only NUMA can scale. Front-side bus technology is dead.

Yes It's really surprising that the guys at xbitlabs have not taken the obvious approach: using a NUMA OS to test a NUMA platform. It is such a pity to use an outdated OS when you have one at hand that, daily, perfectly handle NUMA on computer up to 512 X2 cores. It would have turn this article into something much more interesting. Especially for HPC users like me...

This shit is bull. Damn AMD, you're looking like Intel did during the Netburst diaster.

Well, I do not remember P4 being an exclusive high-end product...
Or You wanna suggest QuadFX is a mainstream product ???

Let me laugh :-)

Bad comment. Commentator did not understand the principle or performance characteristics of ccNUMA architectures, and also did not read the review properly.

Please educate us stupid ones where in the review is stated whether Interleave was enabled(omptimal for non-NUMA-aware OS as XP) and where are tests on Vista o Linux???

I did not mean to imply that anyone was stupid. However, it is self-evident that different memory architectures are optimal for different applications, and that even though one can alter scheduler behaviour, this does not alter the underlying relative locality of memory access and thus may still be strongly non-optimal for a given application. That is, locality of processing does not necessarily imply locality of memory access, especially for (current) games--for which this platform is, after all, marketed--which are often fairly poorly multithreaded.

It is clear that the architecture performs well (and will continue to perform well) on server-type applications; however as I have commented in the main forum already, the wisdom of touting this as a solution for present-day games is perhaps questionable. Remember, the tagline for the platform was not "server hardware at a workstation price," even if that may have been more appropriate, and AMD has and will continue to have little or no control over the implementation of threading by games developers. With time, the application programmers will learn to take advantage of the hardware, but for now, scheduling tricks cannot do much to assist performance here. Certainly one could run two or more current games simultaneously while achieving better performance than the equivalent workload on the Intel platform, but that is not, you will agree, a very serious or practical application.

Well, you can disable NUMA on a architecture that is NUMA.

Local memory is fast, and remote memory is slower.

Interleaving memory is just stupid. 50% of your memory accesses will be remote, and be slower. But while stupid, this is the only option when dealing with a non-NUMA OS.

Normally in a proper NUMA system, you would not interleave. For instance in a 2GB system, the first 1GB of RAM would mapped to the first CPU and the second GB would be mapped to the second CPU. The OS would normally understand to run process on a core which is local that memory, so as much as 100% of your RAM access are fast. Plus, this will leverage the multiple memory buses. Processes can still access memory on the other CPU, but they will just slow down a bit. Apparently, it will not slow down much, as QuadFX is close to Intel on the four application tests, running in NUM-crippled mode, as shown in the review.

NUMA is common is big servers. Such severs might have more than 32 CPUs, several GBs of memory per CPU. NUMA is the only way to scale, as you need to have a dedicated memory bus per CPU.

Doesn't matter much anyway, Quake 4 doesn't really scale much if at all beyond 2 cores currently.

Try running Linxu on it, with the interleaving disabled. And then run some highly threaded apps (well at least four threads).

If XP2 is NUMA aware in sense it really understends an optimizes for it, I would eat my chair.

Actually those FX62 vs. FX72 comparisons in the review are direct proofs of its non-awarenes...

If you mean that it is NUMA aware in a sense it can sense whether running on NUMA system, well based on it DOS 1.0 is NUMA aware with just a few codelines inserted.

And AMD quad will be a TRUE quad core.
Amd is late with 65nm that’s why the delay vs Intel.
They beat Intel with the dual core release.

Intel Double dual core isn’t quad core by my standards.
Packing trick, not for me.

That is what im saying, delay is gonna kill them. Yeah AMD is gonna make a quad core, but in the time intel is gonna do something to beat it, etc.. And i dont see how buying ati was a smart move. I am myself a amd fan, i always owned amd cpu for all my computer. But seeing intel product wich is outperform by far in any benchmark amd product and still using 90nm.. :( sad

It doesn't really matter for AMD.

Regardless of X-Bit Labs bogus claim that AMD only has the low-end market (I guess for workstations only), AMD Opterons are very strong on the server side. HyperTransport and NUMA have a real advantage over the antique single front-side bus that Intel is still using. And these servers would run a NUMA-aware OS with non-interleaves memory, so they actually run fast (unlike the NUMA-crippled motherboard and non-NUMA OS in this review).

In the four to eight CPU market, AMD is uncontested. Intel is trying to get four Xeon's running on a single front-side bus, and performance stinks. AMD gives each CPU a dedicated memory (NUMA). If you want facts, check AMD's sales. Check sales of Sun, for that matter. All Sun's x86 servers are AMD. They don't sell any Intel x86 stuff. Even Dell had to give in, and sell AMD servers, because their server sales have been sucking lately.

AMD will come back to the 1 to 2 CPU market space again, but they must use 65nm obviously. And pack in four cores. NUMA probably won't work on the low-end desktop (yes, gamers are still low-end), because you can't get any advantage unless you have lots of DIMMs and have a highly threaded workload.

One more thing also is, you should add an other test machine, ( Dual Xeon (DualCore) Dual Opteron(Dual Core) the about same price cpu and compare it to the New quad 4x4. I bet that is going to be the same performance or it can be interesting to see if it really worth to put money into the quad 4x4 or if it the same as a Server platform wich exist LONG time and they didnt invented nothing new.

1) Dual Opteron 2000 series is far more expensive at the same performance(frequency) and uses far more expensive(and slower) Registered memory ...
2) Dual Xeon 5100 series is as expensive as Opteron(at the same performance) and uses also far more expensive(and MUCH slower) Fully Buffered memory...
3) WS boards comparable to the QuadFather one go for $450+

Actually on that day AMD will have Agena to trump DualConroe CPU's and AgenaFX will simply leave them in the dust.
Not saying jour comments have merit, just thar even provided they would have, they would be baseless...