Author: Vincent Diepeveen
Date: 12:00:00 05/28/02
Go up one level in this thread
On May 28, 2002 at 11:24:58, Slater Wold wrote:
please checkout specbench.org for their benchmark on
this motherboard. it's like 20% slower than a few epox
single cpu mainboards with the same processor!!!
>One of the features of the Asus mobo that I like the best, is the ability to
>turn off 1 CPU, to make it a single CPU system. This also turns off a LOT of
>the SMP crap that would slow a single CPU system down. Not everything, but a
>LOT.
>
>In my post, I clearly stated that my benchmarks and Sandra results were BETTER
>than any I have seen posted at Toms Hardware, Anand, or any other website.
>Therefore, that leads me to believe that my system is running fairly good.
>Perhaps I am mistaken.
well but it is comparing apples with pies. If you take default sandra
bench then it's using optimizations which favour intel hardware.
If you would recompile it for AMD you will see AMD is doing pretty well
on it.
Theoretical the bandwidth of RDRAM is much higher than from DDR ram,
and on paper the intel cpu's are better in bandwidth.
However, some practical experiments from AMD, which were repeated by
several experts with same outcome, they showed that the maximum bandwidth
they could use was in fact higher with AMD-ddr ram than with P4 + RDRAM.
Pretty amazing to me, but they were not beginners who did it.
I have to admit that in past times i also made huge mistakes comparing
the different processors. For example at some Sun processors i tended
to use the gcc compiler to compile my applications with.
Now that was *not* smart.
Also using gcc at the 21164-633Mhz didn't do much good for DIEP either.
The native compiler appeared to be lightyears better.
IN time we all learn... ...but my feeling says that for computerchess
in future times AMD will remain pretty fast.
Intel is not focussing upon making a good processor, instead they
focus more upon 2 different things
a) a 'cheap' p4 processor which is not so fast but is very
high clocked and which is having fast FSB and such things,
small but very fast L1 cache, fast L2 cache etc. In short
favouring streaming data programs a lot.
b) a very expensive highend processor which focusses upon specint
and especially specfp. The specfp estimations from several
hardware experts of the mckinley are *real* high. Way above
what the cheating Sun-art, IBM and alpha teams *ever* got.
Even if only half of all this is true, then for each Mhz the
McKinley will be a major killer, ALSO for specint programs
such as crafty.
On the other hand AMD is focussing upon puttin gcheap processors
on the market where they save money on FSB (or the follow up of it)
and save money on faster L1 and L2 caches, but the performance of
the processor itself is completely killing away all competitors at
the PC market, which for DIEP is of course the only interesting
thing.
Also i expect that end of this year we get great windows compilers
on the market for the AMD processors.
>
>On May 28, 2002 at 10:22:07, Vincent Diepeveen wrote:
>
>>On May 28, 2002 at 10:05:33, Slater Wold wrote:
>>
>>Slater i agree with you that the Asus motherboards are great
>>motherboards. All kind of features etcetera.
>>
>>However if i compare my asus dual P2/P3 motherboard with a stupid
>>incompatible dual supermicro motherboard which has a stupid bios,
>>then the supermicro P2/P3 motherboard provides up to 20% faster memory
>>access tan the dual P2/P3 asus motherboard did.
>>
>>Note that duals are slower than the fastest single anyway.
>>
>>In case of testing a program single cpu, the dual machines are
>>a very bad testing environment, because they do all kind of stuff
>>to prevent parallel problems. This means in general that testing
>>a program single cpu at a dual system is not a good idea, especially
>>with K7 parallel chipset, which is known slow.
>>
>>On the other hand, a friend of mine, Ron Langeveld, he has a single
>>cpu mainboard with 2-2-2 muskin ram and a 1.73Ghz K7, and it kicks
>>the hell out of the tests you show, even with the same executable!
>>
>>Regrettably some programs which hardly get profiled, like crafty,
>>they depend a lot upon memory speed.
>>
>>At bob's quad which are only 700Mhz processors and where memory
>>goes in PARALLEL, this is simply no problem. Bob doesn't have a K7,
>>and doesn't like AMD much, otherwise i'm sure he would have bought
>>a dual K7 already some time ago and would have found the bottleneck
>>soon.
>>
>>133Mhz FSB is simply dead slow and a deliberate marketing choice
>>from AMD in order to let their new cpu's look even faster.
>>
>>I read now that the hammer is going to be 30% faster with memory
>>latency, that's going to kick butt of course for crafty.
>>
>>Direct 10% speedup for free, which currently means a lot for
>>specbenches.
>>
>>Apart from that we'll see i guess at the end of the year a new
>>release from visual c++ which will hopefully perform up to 50%
>>better for AMD processors when talking about speed.
>>
>>For diep, the FSB speed is not such a major issue as i get less
>>nodes a second. So in short for every million cpu instructions which
>>diep executes, it is doing MORE with the processor than other
>>programs. Crafty needs way more memory lookups in the same million
>>cpu instructions.
>>
>>That means bigger dependancy upon the FSB speed.
>>
>>You can blame bob for this, you can blame AMD for having a small FSB,
>>which logically means that streaming data is always faster on intel
>>(like 3d video rendering, of course AFTER installing a decent
>>graphics card with the latest drivers)
>>
>>Personally i care not so much for this difference in busspeed, but
>>it is obviously a 'we can produce it cheaper now and look even
>>faster next time we release a cpu' decision from AMD.
>>
>>Fact that at specint2000 the extra 256KB L2 cache of the newer northwood
>>is pretty important for many programs, that tells more about programs
>>being too much dependant upon main memory, than it says something about
>>the P4.
>>
>>>On May 28, 2002 at 08:26:16, Aaron Gordon wrote:
>>>
>>>>Slate. I was only 'hostile' due to poor testing methods. When you start
>>>>comparing things that aren't comparable (for example, P4-SSE2 FPU vs AthlonXP
>>>>straight fpu, no SSE/3DNow) then you are providing deceiving results. Would it
>>>>be fair for me to compare a K6-2/300MHz's FPU using 3DNow! to a Pentium3 500 w/o
>>>>SSE? For you.. maybe. I would never do such a thing however. It misleads people
>>>>who don't know any better and is downright bad testing. The same goes for
>>>>Quake3. You and I both know a Tbird 800MHz on a good board providing proper
>>>>memory bandwidth can get more fps than 139. The same thing goes for
>>>>encoding/decoding. Without a good motherboard the CPU or anything else can't do
>>>>much good unless the benchmark/test is 100% cpu biased. Encoding/decoding gets a
>>>>nice boost from faster memory. Whats bad about it is the 'good' boards I am
>>>>talking about are only $80-100. Why not grab one and retest? I can call you and
>>>>tell how what to setup in the bios, which via 4in1's to use, which detonator
>>>>drivers to use & etc. Then you will see a monster come alive..
>>>
>>>#1.) I am using the best dual board out there. The Asus AMD Dual board. There
>>>is NOT a better motherboard (for duals). Period. (In the BIOS, I have the
>>>ability to disable 1 CPU, which I did for these tests.)
>>>
>>>#2.) I am using some of the best memory money can buy. Samsung PC2100
>>>registered sticks.
>>>
>>>#3.) I was using ALL the updated drivers for EVERYTHING. From the chipset to
>>>the damn USB driver. I spent almost 3 hours alone downloading all the newest
>>>drivers for both computers.
>>>
>>>#4.) The BIOS settings on the AMD are just as you have described.
>>>
>>>
>>>The settings in this test were PERFECT Aaron. I am not asking you to believe
>>>me, but I am telling you, IT IS SO. I made sure MYSELF.
>>>
>>>Go to Tom's Hardware, or Anand, and compare my Sandra results to theirs for an
>>>AMD 1.73Ghz. Mine are actually faster. Come on man, I am not an idiot. These
>>>systems were setup fine. The 139 fps for the AMD surprised me too, it's a shame
>>>the GF4 wouldn't work in the P4, I get a 2x+ result with it using the AMD. But
>>>XP didn't want anything to do with it, so I was forced to use the GF1.
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.