Mattie Pattie, Laddie
Look, if a 1.4GHz Niagara suffers from a small cache, how come it is several times faster than a 5GHz IBM Power6 CPU on certain benchmarks? The Power6 has larger cache and 3 times the clock frequency and STILL it looses big to a Niagara. Can you explain this fact? No? Maybe Niagara doesnt suffer from a small cache - as it is faster than the slow Power6?
The Niagara doesnt suffer, as it is several times faster than CPUs with a large cache. The larger cache CPU suffers, as it is slower. Try to dis-explain that! :oP
".....No, I believe the Power CPU is actually a fast server CPU, from the empirical evidence of seeing it run real world applications....."
A server CPU should be suited to handle many clients. And a legacy CPU that depends on a large cache and high clock speeds - will never be able to handle many clients well. Because the cache will never be able to fit in several thousands clients workload into it's cache. The result is that the cache will thrash, swapping data in and out all the time, never reusing the data.
Only a CPU that does not depend upon the ability to fit in the work load into it's cache (like the Niagara) will be able to handle many clients well. And THAT is the reason one Niagara box is twice as fast as three IBM P570 servers together, on SAP benchmarks. This result tells me that Power6 actually sucks on server client workload. How can you need 6 of the P570 to match ONE Niagara box? How slow can the Power6 CPU be? It's a catastroph.
".....As I explained before, you didn't specify what load, so in theory a CPU with a large cache such as Power6 or Itanium2 could do just that. Unrealistic, maybe, but still possible...."
Are you ignorant? How can you believe that a large cache can fit in a true server - client workload??? And then you have the OS to fit in also! The kernel will occupy large portions of the cache. You should go home and do some studying. Begone. Go home! You are not allowed to utter a syllabus until you know what's the purpose of a cache! (It is funny you accused me of being to theoretical)
".......And totally irrellevant to the current thread, which is about the new evidence that Rock has been quietly axed....."
But you never answered to me lastly. "Do you really believe that a server CPU can fit in many thousands clients workload into it's cache?" You just came with some bad explanations. The point is Mattie Pattie, YOU ARE WRONG. I suggest you think over if you might be wrong on other things, as well.
Regarding SUN dropping Rock, yes I suspect that is true. So what? There IS a high performant SPARC cpu in the Fujitsu Venus CPU. That octocore SPARC will reach 128 GFlops. If you need single threaded performance (servers are mostly not dependent on single threaded performance) you use Venus. For massive throughput, you use Niagara. IBM can only offer single threaded performance.