Click to See Complete Forum and Search --> : cpu clockspeeds
mart_man00
06-08-2003, 12:52 PM
how much is to much? programs can never come up to quick but when is that extra speed never used?
if i wanted a box to do some serious compiling regulary and quickly what should i get? i was looking into dual xeons but with some of the specs im wondering if it makes a differnce. and im keeping smp either way!!!
i know there are benchmarks, but id like to know why.
ricstr
06-08-2003, 01:43 PM
If i were to go SMP today i would be tempted to buy 2 * Athlon 2500+ Bartons and do the L5 Mod on them, and an Asus or Tyan board.
mart_man00
06-08-2003, 01:50 PM
why not intel? hypthreading and a faster bus.
can anyone answer my first question?
mart_man00
06-08-2003, 01:51 PM
o yeah, and not so many fans that your computer sounds like a airplane taking off! i like that one, any truth to it?
ricstr
06-08-2003, 02:22 PM
Yes it is true, i have a Thermaltake Volacno 7 in my box thats up in the loft and i can hear it down staris, but the temps are 38c :) i had it with a fan on it that i couldnt hear at the other side of the room but the temps were about 56c. I personaly like AMD to Intel because they dont restrict you by locking the multipliers.
MorphiusFaydal
06-08-2003, 02:35 PM
if you're willing to spend some money you buy a pair of AMD Opterons. you get the bonus of a 64-bit system, which is nice. if you want a 32-bit, id go for 2 AMD Athlon MP 2800+, the 2500 Bartons are only available in Athlon XP, which isnt designed for multi-proccesor. i couldnt even begin to say anything about mobos, cause i have never had a reason to look at them
just my $0.02
mart_man00
06-08-2003, 02:47 PM
so is there a limit where high clocks speed wont do any more?
mart_man00
06-08-2003, 02:56 PM
what i was thinking of was this
motherboard (http://secure.newegg.com/app/specification.asp?item=13-129-127)
cpu (http://www.newegg.com/app/viewproduct.asp?description=19-117-019)
its really pricey but i think it will be less or the same as my desktop.
ricstr
06-08-2003, 03:02 PM
Bartons are only available in Athlon XP, which isnt designed for multi-proccesor.
If you have a look over at http://www.2cpu.com/ they have stuff about the differences of Athlon XP an MP and how to use XPs on a SMP setup.
I heard the opterons wernt avaliable untill september.
so is there a limit where high clocks speed wont do any more?
Nope there is no limit as far as i know.
High clock speeds will speed up processor intensive tasks like encoding video/audio compiling etc. but in general for every day tasks such as web browsing, checking email there wont be much difference as these tasks will be processed so fast.
bwkaz
06-08-2003, 04:41 PM
Originally posted by ricstr
I heard the opterons wernt avaliable untill september. That's the AMD64 chips (the UP version, basically). The Opterons (the SMP version chips) are available now. :)
I'd go with AMD too, and part of the reason is that Intel doesn't have faster bus speeds. They used to claim they did, when they used Rambus memory, but that was a crock anyway, because the processor would only ever use one of the two channels that Rambus offered. So the quadrupling of the real bus speed that they did (e.g., if the real bus was 200MHz, they claimed it was 800), wasn't reflected AT ALL in real world performance.
Now, most Intel motherboards use DDR instead of Rambus (and as a bonus, DDR is much less expensive as far as RAM goes), so the memory bus speeds are the same. The processor bus is, too, pretty much.
The other reason I use AMD (Athlon XPs) is that the chips are (or at least, were) cheaper for the same "effective" clock speed. Of course, the Athlon is running at a lower clock (my XP1800 is running at 1533MHz), but because of the much, much lower pipeline stall penalty, it ends up outperforming the P4 2GHz in a decent number of things.
The pipeline stall is a function of how deep the pipeline is (how many separate instructions are executing at any one time, basically). It takes more than one clock cycle to execute one instruction, but each instruction's tasks can be split up into a certain number of stages. Each stage executes in one clock cycle, on one piece of the pipeline hardware. When the clock ticks, the results of that stage are saved into intermediate registers, then they get used by the next pipeline stage during the next cycle. This ends up doing (about) one instruction per clock cycle.
It all gets screwed up, though, if the processor mispredicts a branch instruction (because it prefetches a lot of instructions, normally). When a branch gets mispredicted, every instruction between the branch and the instruction that is going into the pipeline needs to be squashed (because they shouldn't be executing because of the branch). In a deep pipeline, each stage does less work, so these instructions are taking up more stages than in a shallow pipeline. So more stages get squashed, and more clock cycles are wasted completing the new instructions.
And AMD locks their multipliers too. They just let you get around it by putting the circuitry that does that on the chip itself, rather than in the silicon core, so you can use conductive paint (if you're REALLY careful) to unlock them.
deanrantala
06-08-2003, 05:42 PM
I've done mods on XP's Myself and it IS a vers cost saving deal. Also, if you are not afraid to spend money, you con go with the opteron or (as an AMD fan, I hate to admitt) a slightly better full out 64 bit deal: Intel Itanium2.
mart_man00
06-08-2003, 06:16 PM
id like to throw in one more factor, the compiler.
intel has a compiler that is supposed to be great. i hear it makes use of almost everything, if it will compile. i know that some gentoo guys are working on patching the code for icc, but what about amds? any optimizations for them? does it matter? i mean how much does amd beat intels by?
thanks for the info guys.
bwkaz
06-08-2003, 06:20 PM
Recent versions of gcc have -march=pentium4 and -march=athlon-xp optimization flags. Don't know how well they work compared to each other (since I don't have a P4 -- I do know that athlon-xp works pretty well), but I do know that -march=i686 on a P3 is much faster than -march=i586. Of course, that doesn't have any bearing on this discussion... ;)
ricstr
06-08-2003, 06:33 PM
And AMD locks their multipliers too. They just let you get around it by putting the circuitry that does that on the chip itself, rather than in the silicon core, so you can use conductive paint (if you're REALLY careful) to unlock them.
I have used Athlon XP1700+, XP2200+ Thouroughbred and XP2500+ Barton they all seemed to be unlocked.
Posibly the motherboards I were using.
bwkaz
06-08-2003, 09:42 PM
Were you changing the multiplier, or were you changing the front-side bus?
If you were changing numbers like 9, 9.5, 10, 11, etc., then you were changing the multipliers.
If the numbers you were changing were 100, 120, 133, 143, 150, or whatnot, then you were changing the front-side bus speed.
ricstr
06-09-2003, 04:24 AM
I have changed both multiplier and front-side bus without unlocking the chips, I using Asus A7VXXX boards.