Author: Robert Hyatt
Date: 17:14:09 12/10/02
Go up one level in this thread
On December 10, 2002 at 19:36:15, Vincent Diepeveen wrote: >On December 10, 2002 at 12:26:55, Robert Hyatt wrote: > >>On December 10, 2002 at 11:03:13, Vincent Diepeveen wrote: >> >> >>> >>>please read the data posting done here about crafty with mt=2 at a 3Ghz P4 >>>with SMT turned on. >> >>That is _one_ test run, on a _very_ small set of positions, with a version of >>Crafty >>that is _known_ to have a problem with spinlocks/spinwaits on an SMT machine. >>So what is your point? > >So how can we take the data from Nalimov serious knowing coming year we >can't buy that hardware he uses for his SMT runs at the 2.8 Ghz Xeons? Can you explain what you are talking about? You can contact www.dell.com, order a poweredge 2600 with dual 2.8 xeon cpus, and they will ship it the _next_ day. I have one in my office. I ordered it Monday of last week, it was in my office on Wednesday. Once again, another statement that is nonsense... > >This P4 3.0Ghz run is an *actual* run. The chance this run is off from the >average case by 20% is like nearly 0%. My 2.8 xeon data is an _actual_ run... You don't even have one but can make absolute pronouncements as to its performance. :) Typical... > >In diep i hardly *ever* get such big differences. Yes 10% sometimes, >very sometimes. Usually always the same search times though. Nodes >can change. Search times hardly! Crappy program... what more can I say... > >I remember a case where i was upset about a 6 seconds difference at >a 100 seconds run. 6%. > >I wanted to explain it how and why. I could never repeat that run. > >Never. > >But saying that the numbers are not accurate beause of 1 accurate run >is not accurate enough for you, that is not very scientific. It is _very_ scientific. You are the _only_ person that claims to have no variability in your parallel search times. The _only_ person on the planet. That either makes you something special. Something stupid. Or just something that posts fraudulant stuff... > >Instead that is completely denying the truth. One of us is doing that, for sure. Of course, I am sure that myself, Bruce and everyone _else_ with a parallel search, each and every one of us is incompetent... > >I would want to say that nalimov is working >for wintel, that we shoudl skip all his tests unless we put a >sticker on it 'not available in any shop coming year'. The 2.8 xeon is available right now. Just call Dell and ask, rather than making statements that make you look like an idiot.. > >That means all the tests you quote from the not yet available 2.8Ghz Xeons >with SMT (and i doubt we will ever see them on the market, because only >Xeon MPs are having SMT and P4s >= 3.0 Ghz; see the intel official >statements at their OWN homepage; they would be shouting from the ROOF >of course loud so that everyone could hear it if their 2.8Ghz stuff >would be already having a well working form of SMT) you should not post >either unless you post with them: "this is not getting sold in any shop". > Mine was sold to me by Dell. It is a dual 2.8 xeon. It has hyper-threading. It works just fine. It can be ordered today and it will be there Thursday. Regardless of all that hand-waving nonsense you spout... >Apart from that. I can't remember that from any solution time you give >from crafty you repeat the test twice. So if you solve some position >that is not needed, but for a speedup measurement it is? > You know that is simply a false statement. As I have reported _many_ times when I report speedup numbers I report an _average_ over several runs. Are you really that incapable of remembering important details, or are you just being obtuse to look stupid? >Are you willing to declare here that not a single solution time of >crafty we should believe because in reality it will be far off from >what you post? Didn't say that at all, quite the contrary in fact... > >Best regards, >Vincent > >> >>> >>>> >>>> >>>>>so is no good of a standard >>>>>here. Also it is doing 2 probes in 2 different hashtables which i cannot >>>>>do even in DIEP (too slow for me) i do 8 probes in 1 hashtable sequential >>>>>(so a good bandwidth is helping diep more than it is crafty for example). >>>>> >>>>>My own testing at the machines you mentionned, with exception of the >>>>>3.0Ghz P4, i found that for the newer generation P4s the speed >>>>>difference is only 1.5 for 133Mhz versus 133Mhz bus. >>>> >>>>What in the world does that mean??? >>> >>>Ah bad memory of you and not 100% clear sentence of me: >> >>No bad memory of me. You said "difference of only 1.5 for 133mhz vs 133mhz" >>so there is _no_ memory issue there. There _is_ a writing issue. _yours_. If >>I >>read correctly, 133 == 133, yet you say 133 is 1.5X faster than 133. And that >>is >>a "bad memory" on my part? Thought so... :) >> >> >>> >>>1 Ghz K7 = 1.7Ghz P4 (old core with 100Mhz quad pumped rdram versus 133Mhz >>> memory of k7) >>> >>>new core (ddr ram especially both 133Mhz now) >>> >>>1 Ghz K7 = 1.5ghz P4 but now you must see it as 1.6Ghz K7 == 2.4Ghz P4 >>>for DIEP that is. >>> >>>For WARP it was better than that. >>> >>> >>>No as you can see in the posted data, crafty had a 13-16% speedup at >>>a single cpu P4 at 3 Ghz with MT=2. >> >>On that one set of positions. I posted _other_ data as well... >> >>You need to stop quoting things as "absolute fact" when there is so much >>variability. Particularly things like "needs less nodes with mt=2" as that is >>total nonsense for the average case and only happens in unique cases. >> >>>>>>Vincent
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.