Jesper Frimann
Even if I dont agree with you, I enjoy discussing with you. Because you are well educated and no troll. You back your claims up with links, and we can discuss around your links, and your explanations. That is academic. Matt Bryant, OTOH, just posts his opinions, which he has explained. He explained that "in his opinion, the Niagara suffers from a small cache". Well, it is a difference between facts and opinion. Never should you try to disguise your opinions as facts, as Matt Bryant does. You distinguish between your opinions and facts. I wish more people discussed like you do.
.
"...Sorry no points for quoting BMSEER, this is SUN FUD central..." You mean that because BMSEER posted this benchmark, the benchmark is not valid? Why is it not valid? Is is a fake? Has BMSEER made up the numbers? You mean the true numbers are maybe half of what is reported? Or? I dont understand. Even if IBM reported a benchmark, I would accept it. I dont expect IBM to lie about hard numbers?
So, what do you think the true Niagara numbers are for string pattern matching? BMSEER reports that one 1.4GHz Niagara achieves 24.6Gbit/sec and two 3.2GHz Cell achieves 3.8 GBit/sec. Do you mean in reality one Niagara reaches maybe... 1GBit/sec and the two Cell reaches 50Gbit/sec? Because BMSEER is from SUN, he is lying about the true numbers? Could you elaborate on his lies? Exactly which number is a fake?
.
"....Again this is a cherry picked benchmark...." So what? Every solution will have it's strength and weaknesses. If IBM picks a benchmark that shows Power6 strengths, I would be dumb if I dismissed it as fake? I mean, the numbers are real. The Power6 got that performance. Why would I dismiss the numbers? I am trying to show that Niagara is not as slow as FUDers say. There are things where the Niagara is fastest on the planet. Of course, the Niagara is not fastest on everything, it has it's weaknesses. But is also has it's strengths. But when people try to show the Niagara's strong sides, you just dismiss them. About Power6, I know it has it's strong sides, I dont dispute about that. But people FUDs about Niagara has no strong sides at all, "it is just a weak CPU in general". I say "Wrong! It has strong sides". But they are just dismissed. I dont dismiss the Power6 strong sides, but you dismiss the Niagara's strong sides? Fair eh?
.
"...What our good friend BMSEER has done is to point to the place where Cell doesn't perform well on pattern matching. Actually for small dictonaries a single cell, using it's SPE's will do around 41 Gbit, which is 1.7 times what the Niagara processor will do..." Ok, it looks like one 3.2GHz CELL achieves 70% better score on small dictionaries. If I were you, I would have said "no, it can not be true, because IBM published that paper". But I do not. I accept that number. Instead I think that small dictionaries shows that CELL doesnt scale to larger realistic loads when the performance drops to 1.9 GBit/sec, that is 96% drop of performance, which in my eyes is terrible. And besides, who is interested in small loads? I mean, all realistic true life server workloads are large loads. Who is interested in small things that no one can use? Even the Power6 would perform well, if the load was sufficiently small enough to fit into it's cache. But CELL performs good only on small loads, not interesting for real life. Whereas Niagara has again proven that it does not need a large cache to perform superior on large heavy server loads. Mattie Pattie Laddie, here you see again that despite having a small cache, the Niagara does not suffer. Instead it smokes all compettion on large realistic loads.
.
"...So again this is a case of not telling the whole story, which is that For small dictonaries Cell is faster for large ones Niagara is faster. And that there is quite a good reason for this, the two processors are targeted at different markeds/workloads..."
But who is interested in small loads? The difficult part is large loads, real life, realistic loads. And which solution should you choose then? For real life, realistic loads, you should choose Niagara. Crystal clear. Niagara wins again. If you have small loads, that no one runs in real life, then choose Cell. Or choose Niagara and you can use it for small loads AND large loads.
.
James Billingham
"...Sun ref. states CELL has 8 cores - which is as untrue as my statement, one PPE which is a cut down Power core and 8 SPEs which are by no means general purpose CPU cores and as such should not be compared to the Niagras 8..."
"...Having read both I'd have to conclude the Sun article is a peice of marketing FUD..."
So you mean that CELL does not rely on it's 8 cores to do work? And therefore you should only count CELL as one core, the powerpc core? Cool. So you mean that the powerpc core alone, by itself, achieves 41GBit/sec on small loads? Or is it so that the CELL uses all 9 cores to achieve 41GBit/sec? So when you talk about the CELL you describe it as "multi core CPU having 8 cores" or as "single core CPU having 1 core"? What I am trying to say that you can describe the CELL as single core if you wish. Me, will describe it as 8 core + 1 core = 9 core CPU, just as SUN does. And that is a more truer description of CELL. And that is not FUD, neither from me or from SUN. Even IBM describes the CELL as multi core CPU. I must say that you IBMers have a strange view of what FUD is. "SUN stating CELL is multi core CPU is FUD" - yeah right.
So you mean that when IBM claims one Mainframe can consolidate 1500 x86 servers are facts? It turns out that IBM assumes the x86 servers to all idle and the Mainframe loads at 100%!!! So I can claim my laptop can consolidate lots of servers, if they all do nothing - just as IBM does. This is just lies.
IBM also claims that Power6 has 250GB/sec bandwidth. Then IBM adds all bandwidth in the CPU! That is a lie. If there is a bottleneck on 10GB/sec, the chip will never be faster than 10GB/sec. You can not add all bandwidth, that will never be achieved.
Or when IBM claims "one Niagara core is slower than the Power6 core, ergo the Niagara CPU is slower than the Power6 CPU". That is also lies. Assume the worlds fastest CPU has 10.000 slow cores. Can you then say "one core is slower than Power6 core, ergo, the cpu is slower"??? No. That is just lies.
I must again say that IBM has a strange view on what FUD is.