Everyone favorite rag has supposedly pinned AMD down on the existence of a CrossFire-enabled chipset due by Feb.

I guess by next month we'll know if the chipset has any tweaks to improve the NUMA perf (even though Vista has the implementation, a bad chipset will cause perf problems).

It is due to debut around the time that R600 is to pop up its head. ATi chipsets are known to be lowest power ones otu tof the thre bigs X86 manufs.

I guess it can't get worse.

I still say no hope for Quad FX until it offers something that Intel can't. It's only hope is octo-core (two quad cores CPUs). But even then, that'll be a niche market for AMD's people who need platformance for their mega-tasking.

Interesting find Baron, thanks.

I don't see how a chipset could improve NUMA performance in any way.
NUMA is completely bypassing the chipset, the memory controllers are in the CPUs, and they are directly connected by HT.
A chipset can help only in increasing bandwidth/latency toward the PCI-Express bus, or reduce power consumption, but i think the former is not a problem at all with 680a, and the latter could be reduced, but by a marginal factor in the overall power balance.
The only way to increase NUMA performance is a faster HT link, with lower protocol overhead, and more cache in the CPUs; none of this can be obtained with a new chipset.
...
Oh well, if i want to pull science fiction out of my arse, i can think the following:
- chipset with direct access to memory, large integrated cache (let's say 16MB or more) and prefetching logic to sit in between the 2 CPUs, hooked to the HT links and to the memory controllers of the CPUs, which are not able to access memory directly anymore.
The chipset would fool the CPUs into thinking they are directly talking to memory or the other CPU, while in fact they are not, they get served their data through their IMC and HT links from the chipset, which has access to a single memory pool.
This would be in fact some kind of north bridge.
Of course, such a solution is sheer foolish, it completely defies the concept of Direct Memory architecture, would cost a lot and be extremely difficult to design.
Not on earth.

I disagree. For people who want their preferred brand to be faster than the compeition that maybe important but I personally only want it to be faster than the AMD chip I have.

QFX is that and more.

Let's say that you buy a car that has a bad carburetor. You race it and lose, but then find out the carburetor was bad. Does that mean the car was bad?

You fix the carburetor and though it doesn't make it go faster than the defined max speed, it also won't go slower.

I guess we would have to get an Opteron 2218 and run single threaded apps on it with XP Pro, X64 and Server 2003.

If the Opteron scores higher at the same clockspeed then the chipset is

HOLDING THE PLATFORM BACK.


That was my point. Not that a chipset can increase speed.

Uhh, yeah, i got your point.
However, nVidia should have screwed up big time (like a major bug) to induce a penalty to CPU performance in a chipset which is essentially just a tunnel between the HT link and the PCI-Express bus.
The case of traditional chipsets such as Intel ones is different, since there the north bridge integrates the memory controller, hence it plays a major role in CPU performance.

If I remeber correctly lots of folks had to flash the Asus P5W(?) BIOS just to put Core 2 on it, so it is possible that a glitch liek that slipped through QA at Asus.

Also as everyone says it's a brand new mobo with a brand new design, so there maybe issues with the first retail rev. Poeple here even said to wait for a new revision of Core 2 when it was first released.

Crap happens and like I said the only real way to prove it one way or the other is to compare it to Opteron at the same clock with different OS' and singlethreaded apps.

It is easier to explain this to a brick. It has more chances to undersand anything.

The funny thing is that I can't stand you but I don't stalk you on the forum. if you have no opinion regarding the topic, go to a post where you do.

This would make sense for some performance spanking reviews of the R600. A good chipset with what we hope is a killer card in the R600 will divert some of the specific attention away from the 4x4 and move it to the performance of the whole system and it's potential. Which is only a good thing for AMD.

You cant improve something that ISNT THERE.

And that was my point. The platform is a good idea but the issues revealed by LegitReviews cast a shadow over the launch. Of course if it was Intel, they would get the benefit of the doubt.

Anyway, that is also my hope; that a new chipset(lower power) and GPU (lower power 80nm R600 even with 500M transistors).

Perhaps nVidia will pick up spedd and go to 65nm quickly that will definitely drop power a lot.

Fortunately for me I will be getting this for the two sockets and NOT the 4 PCIe and the apps that matter are EXTREMELY multi-threaded.

It will do well for me even if it's not faster than C2Q.

The inquirer is predicting 720million...

Not for the high end, just craploads more transistors which means more heat.

Sure it is. It hardly ever loses to C2D X6800 and E-versions are long gone.

I am interested in the scaling shown in multithreaded game tests. Even if it achieves less than this theoretical number, it will still take basically every other chip out and beat the crap out of it.

This is of course excluding C2Q.

Why are you all so down on this? Two sockets are supposed to use more power than one. Except for singlethreaded games FX62 is totally improved upon. I would say that if the NUMA of Vista does it's job correctly, all game data will be on one socket's RAM banks.

Either way, Visual Studio will fly.

The inquirer is predicting 720million...

Not for the high end, just craploads more transistors which means more heat.

TSMC and UMC are already working on 45nm and there was a report that nVidia is looking to shrink G80 soon. Supposedly because of the complexity they didn't want to do the shrink at the same time. I would think that by April we will see a 65nm announcement for GPUs.

That would definitely give nVidia back the advantage since it is said that R600 will be faster than G80.

TSMC and they're not making 65nm products atm.