Hi, On Thu, Aug 15, 2002 at 11:55:53PM +0300, Oded Arbel wrote: > Yedidyah Bar-David wrote: > [snip] > > >2. This, of course, isn't very interesting - what's usually is is actual > >performance. And this, as we all know, depends heavily on the application. > >[SPEC'S CPU2000 benchmark snipped] > > > I'm too a bit surprised at these results. first - I try to follow Tom's > hardware when I have the time, and IIRC in AMD vs. Intel benchmarks, > Athlons have scored pretty tight with P4s of comperable "marketability" > (AMD like to use PR ratings - let them have their fun), some times > scoring higher on more benchmarks then Intel's. It is also known that > while floating point calculations have been a bane for AMD ever since.. > well - ever since NexGen, the new AMDs score higher in floating point > then Intels, while falling back in the integer calculations. > > I'm also missing the G3/4 PPCs in this benchmark - the G4s are currently > clocking way below 1GHz : about 600 to 800 MHz and offer competitive > performance with applications built to take advantage of the G3\4's > unique processing architecture and instruction set extensions.
Do you think they are significantly different from Power? I haven't followed closely, but I understand they share a lot of the design. I though the diff is mostly things like caches etc. > > >I must admit this last one was a big surprize for me. I anxiously wait > >for the Hammer to fight back. > > > I'm not very optimistic : the Hammer is supposed to be somehwat x86 > compatible, while the Itanium carry no such burdens - as a result we can > see it display very good results in comparison to rival chips twice its > clock cycle speed (see what a timely re-design can acheive). I read somewhere (Tom's? anadtech? don't remember) that a pre-production demo of Hammer at 800Mhz gave results similar to 1.6Ghz P4. They obviously invested a lot in the new design (as opposed to what I wrote in the previous mail, of course). Also, the new architecture of the Itanium will require very sofisticated compilers to achive reasonable use (in terms of number of simultaneous instructions dispatched), hopefully the Hammer will do that by itself. BTW, for anyone interested in such things, I suggest reading a comparison between f-cpu (a "freeware" CPU, whose design is underway, and I enjoy following even though I understand maybe 20% of the posts, not counting french-written ones) and Alpha, at <http://www.f-cpu.org/alpha_vs_f-cpu.txt> > > >So, in "work per cycle" the P IV is worst, but who cares? In > >"work per second per $" the fight is not over yet (I didn't check > >current prices) but is definitely not towards being efficient. > > > I'll advise consulting other benchmarks too, before making a decision as > to where to put your money. No doubt about that. My personal favourite (besides my own programs - the current is my hebrew OCR) is kernel compilation. > > >Conclusion? <snip> people love GHz. > > > That they do. and a good advertising campaign is always your best asset :-) > > -- > Oded > > ::.. > Famous Last Words 081-"OK, O Mighty Odin, as long as you're not gonna answer my >prayers, I'm gonna tell ya what I REALLY think of ya! > Didi ================================================================= To unsubscribe, send mail to [EMAIL PROTECTED] with the word "unsubscribe" in the message body, e.g., run the command echo unsubscribe | mail [EMAIL PROTECTED]
