Did anyone see that article on Toms about the 275 with the 248? (3-core)

[Order]

http://www.tomshardware.com/cpu/20051128/index.html

I was surprised that it worked at all, let alone rather well.

 

[pawstar]

AWSOME However, it would have been far more advantageous to see a 2.8GHz single core with the dual 2.2 GHz core - that way single threaded applications are running fast, and those applications that can utilize multiple cores also run fast. I was thinking about doing something like this for my next rig, but decided to stick to two single cores in my plan for stability sake. But if such a thing would work DROOL

 

[Dan_D]

Yeah it's pretty cool. It would be great if they had done it with two 90nm CPU's. That would have likely worked alot better. All the apps that gave errors might not have.

The way I'd set it up is to use a Opteron 254 as the primary processor, and use a 275 or 280 as the secondary CPU. That way I'll bennefit from the increased clock speed of the 254, and have the extra cores for anything that takes advantage of them.

 

[Order]

I'm pretty sure that the faster CPU would clock down to match the slower ones.

 

[Dan_D]

I'm pretty sure that the faster CPU would clock down to match the slower ones.

Nope. From the article I didn't get that impression. When you are dealing with seperate CPU's, it's not going to work like SLi or anything like that. The clock speed isn't adjustable in that way.

 

[NulloModo]

But the CPUs they used were all the same speed. I'm not positive, but isn't there only one system/memory bus? I know AMD systems technically don't have FSBs anymore... or do they?

 

[Steel Chicken]

But the CPUs they used were all the same speed. I'm not positive, but isn't there only one system/memory bus? I know AMD systems technically don't have FSBs anymore... or do they?

they could run the bus at the same speed for all the cpus, but different cpus could have different multipliers. they each contain their own memory controller.

 

[Order]

I'm 90% positive that you can't have mixed clocks.

 

[Dan_D]

I'm 90% positive that you can't have mixed clocks.

On a Xeon you absolutely can not. The second CPU won't even initialize. You'll get a mismatch error and that's that. On the Opterons it may work differently.

I am just impressed that Toms was able to get a 130nm and a 90nm CPU to work more or less stable at the same time. That impresses me more than the fact that it was one dual core and one single core chip.

 

[Order]

Yeah, I'm glad Tom's is doing something unique and will give its detractors something positive to erode their negativity.

 

[APOLLO]

I am just impressed that Toms was able to get a 130nm and a 90nm CPU to work more or less stable at the same time. That impresses me more than the fact that it was one dual core and one single core chip.

It just demonstrates the overall versatility and stability of the AMD architecture. Too bad Tom's didn't try assymetric clock combinations as well. If this was a completely viable alternative, it would be a great upgarde path for users who can't afford to fork out the cash for two dual-cores. The intermediate solution demonstrated superiority in many of the benchmarks that mattered.

I'm still trying to fully understand why the tri-core setup beat the quad-core in a lot of the results. Sketchy thread support for quads? If that's the case, a lot of software seems to plateau at three cores, and beyond is actually detrimental for performance. However, I imagine for ultimate multtasking, the more cores the merrier.