Re: The worlds fastest cpu
I do agree that generation are difficult to define precisely due to the multitude of things that could change. If someone else wants to use a different definition, that's fine, but it's rather missing the point. If a minor change to the chip, say changing PCIe2 to PCIe3 is considered a generation, then simply upping the processor speed could be classed as a generation, which is silly. Something major has to change to be judged a generation and this often means fundamentally, the core. For instance, I wouldn't classify Power X to Power X+ as a generation. Although it has varied, that normally just involves something like a shrink and speed increase.
Although I do look at benchmarks to get some information, I have never said they show performance in a real world environment, whether Sparc, Power, Xeon or whatever. You can deduce some things from them, but when some of them effectively include the performance of the disk subsystem as well and much better results can be got by simply changing to flash storage, this renders them somewhat pointless in terms of processor meaning. So, no, my stance hasn't changed at all.
"You know that four 1.6GHz SPARC T2+ was as fast as fourteen (14) 5GHz POWER6 in official SIEBEL v8 benchmarks?"
Ha, ha. Anybody who has run both these platforms (and I have) knows that this shows benchmarks have no relevance to real world performance. From what you're posting, you don't seem to realise that benchmarks mean nothing in the real world, where real workload actually runs. The early Tx processors could be used, but you had to be incredibly careful of the world, or they were real dogs. Until the T4 and its critical threads implementation, anything requiring decent single thread performance ran appallingly. Of course, using critical threads had the effect of dramatically reducing the throughput of the servers, so the benchmark flies out the window again. I'm not saying these servers were rubbish, just that you had to be very careful when and how you used them and the benchmarks meant nothing in the real world. On exactly the right workload, throughput was great. On the wrong workload, they sucked.
"For instance, one single IBM P595 server with 32 cpus that took the old TPC-C benchmark, costed 35 million USD list price."
Now, this is a really embarrassing statement of yours and shows you don't understand the benchmarks. The cost is actually for the WHOLE setup. That includes disk subsytems etc.etc. Not just the server. In most modern benchmarks of this type, the vast majority of the cost is the disk subsystem, not the server.
"How much slower does POWER8 get when turning on full encryption?"
You didn't read a word I wrote. Both Power 8 and Sparc M7 have encryption accelerators. However, Oracle software is specifically coded not to use the Power 8 encryption accelerator, so of course, Power 8 is slower as it still has to do it in software. However, this is the softwares choice. If on x86 software is written to not use SSEx (take your pick which variant), it doesn't matter if the processor has SSEx built-in or not. It won't help anything.
You really need to stop believing benchmarks and actually try running some workloads across different architectures and manufacturers and then you'd realise the benchmarks are rubbish. It seems clear you either don't actually run workloads, or you only run on Oracle hardware as a matter of faith rather than rational thought process and facts.