Could one core use all of the Q6600's 8mb cache if an application could take advantage of it?

[Hajpoj]

I wanted to share an early hypothesis as well. Theoretically, a quad-core should be superior in all single-threaded apps solely due to more cache available to that working core. Say if you disabled 2 of the q6600s cores: clock-for-clock, an 8mb dual core should perform better than a 6mb dual core(e8400) right?

 

[Martimus]

No. Only half of the L2 and a quarter of the L1 cache are available to each core on an Intel quad core, because of the MCM process. AMD cores would have access to all of the L3 cache, but only 1/4 of the L2 and L1 cache, because of the way that they are designed.

 

[RedShirt]

On AMD's Quads the L3 cache works this way. Intel's Quads actually are 2 dual core chips, so, any one core could only access 4mb of cache. Any communication between the two dual core dies has to occur over the bus, which is much slower than just using cache.

Fortunately for Intel, their chip design is just so much better than AMDs that even with this limitation the Intel quads kill AMDs.

 

[Cogman]

Hence AMD's hype about having a "True Quad Core" True, their core should be able to share data pretty easily, but their overall design is inferior compared to intels .

Isn't Nehelam going to be a "True Quad Core" and a "fake" octa core?

 

[Idontcare]

Originally posted by: Cogman
Isn't Nehelam going to be a "True Quad Core" and a "fake" octa core?

You could phrase it that way.

Quad-core nehalem with SMT will enable 8 threads of processing power too.

To me this (SMT enabled quads) is the single-biggest threat facing AMD in 2009 because they simply have nothing (that they've published) in the works to combat such massively threaded beasts per socket.

 

[Martimus]

Originally posted by: Idontcare

Originally posted by: Cogman
Isn't Nehelam going to be a "True Quad Core" and a "fake" octa core?

You could phrase it that way.

Quad-core nehalem with SMT will enable 8 threads of processing power too.

To me this (SMT enabled quads) is the single-biggest threat facing AMD in 2009 because they simply have nothing (that they've published) in the works to combat such massively threaded beasts per socket.

If AMD can get their own form of SMT with the Bulldozer, then I don't think it will be a problem. Until that time, I doubt SMT will make much of a positive difference in most situations compared with a quad that does not have it enabled.

 

[bunnyfubbles]

Originally posted by: Martimus

Originally posted by: Idontcare

Originally posted by: Cogman
Isn't Nehelam going to be a "True Quad Core" and a "fake" octa core?

You could phrase it that way.

Quad-core nehalem with SMT will enable 8 threads of processing power too.

To me this (SMT enabled quads) is the single-biggest threat facing AMD in 2009 because they simply have nothing (that they've published) in the works to combat such massively threaded beasts per socket.

If AMD can get their own form of SMT with the Bulldozer, then I don't think it will be a problem. Until that time, I doubt SMT will make much of a positive difference in most situations compared with a quad that does not have it enabled.

The extra logical threads actually have to provide a tangible performance boost as well. HT on the P4 wasn't anything too amazing except in relatively niche scenarios. It isn't like Intel hasn't yet designed a processor with specs that make it sound better than it really is...with the P4 it was hyper inflated MHz that didn't scale with real world performance, and really was only there so that CPUs would sell themselves. With Nehalem it very well could be a move of, "LOOK AT HOW MANY 'CORES' THIS HAS!!!1!1!1!!" Not saying that's definitely how its going to go down, but we'll definitely have to see how the parts actually perform before we put another nail in AMD's coffin .

 

[Nathelion]

Hyper-threading did actually work though. In multi-threaded scenarios, performance boosts of 10-25% were often observed. The reason everyone laughs about it is because it wasn't nearly as good as having a whole extra core in there (up to 100% performance boost), and Intel was trying to make it out as if a P4 with HT was just as good as an AMD dual core. Which is, of course, bogus. Still, 10% is nothing to sneeze at, much less 25%.

 

[Martimus]

Originally posted by: Nathelion
Hyper-threading did actually work though. In multi-threaded scenarios, performance boosts of 10-25% were often observed. The reason everyone laughs about it is because it wasn't nearly as good as having a whole extra core in there (up to 100% performance boost), and Intel was trying to make it out as if a P4 with HT was just as good as an AMD dual core. Which is, of course, bogus. Still, 10% is nothing to sneeze at, much less 25%.

That extra 10-25% extra performance comes at the cost of quite a few extra transistors, which AMD may feel are better used for something else. I'm not saying that it is useless, just that there may be a more useful thing to do with that extra circuitry when you already have 4 entire cores to do threaded processing.