64-bit Toys for Athlon-64?

gbulmash asks: "I'm looking to see just how much performance I can squeeze out of a new Athlon-64 system. This isn't for benchmarking, but more like you got a new car and you're looking for a long, straight road where you can push the needles into the red before letting off the gas (and then maybe a twisty mountain road to test cornering). Can the all-knowing Slashdot readers recommend some AMD-64 enabled/optimized distros and packages that will let us new proud papas of AMD-64 systems fully open up the throttle on these bad girls and see what they're made of?"

Comments

Well,

[Sevn]

You can build a Gentoo system completely optimized for AMD64 with help from this page:

http://dev.gentoo.org/~brad_mssw/amd64-tech-notes. html

T-Minus 10 second and counting til someone starts bashing Gentoo and recommending Debian.

Re:Well,

[Anonymous Coward]

gaytoo is for the G4Y!!!! bee a man and run DEBIAN you pushyfag! DEBIAN IS DE R0XX3RS!!

Re:Well,

[Anonymous Coward]

gentoo has a shockingly high amount of neonazi users. i'm actually writing my thesis on the disproportionate amount of neonazis who use gentoo compared to other operating systems... in summation-- dont be like nazis, use debian.

Re:Well,

[Anonymous Coward]

U shud write your papier on why D3B14N R000L3Z de m4crovuz! Gaytoo not for n4z1s. Sum n4z3rs are c00lz. Gaytoo iz f0 D3 GHEY!!! D3B14N F00EVAR!!

Re:Well,

[Anonymous Coward]

I am SO fucking sick of you debian zealot kiddies posting your stupid shit every time someone mentions Gentoo. Can you just get over yourselves already? Gentoo is better. Just because you can't figure it out because you are so stupid you can only use debian isn't our problem.

Re:Well,

[Anonymous Coward]

Oh yeah. Reading a reply like this from a typical debian user definitely makes me think highly of it. You and your thesis writing butt buddy should just get over yourselves, stop the gay bashing, and try Gentoo.

Re:Well,

[bobbozzo]

Also, I think there is a AMD 64-bit version of Suse available, and a beta of RedHat also.

Re:Well,

[Creepy Crawler]

>>>"T-Minus 10 second and counting til someone starts bashing Gentoo and recommending Debian."

Wrong. Debian is made to be a easy-to-maintain stable linux distribution. Because of being made for servers, it also uses tried and tested (read: outdated) packages optimised for the whopping 386.

Gentoo is for die-hard users who demand everything current and fully optimised for their system. Because it's their system, they can sacrifice some stability.

Sinply it's a trade-off for "stable, unoptimised, and oldness" OR "instability, very optimised, and new".

And I boldly say, if you want to fully use a Opteron, use Gentoo.

Re:Well,

[You're All Wrong]

"""
T-Minus 10 second and counting til someone starts bashing Gentoo and recommending Debian.
"""

??? I'm a died-in-the-wool Debian user, and yet I've recommended Gentoo to many more people recently than I've recommended Debian.

Do you really see that much Gentoo bashing? Are you sure they're real debian users and not just MS plants? Either way, they're idiots, Gentoo's perfect for this situation.

Foot to the metal!

YAW.

Re:Well,

[Anonymous Coward]

I am SO fucking sick of you gentoo zealot kiddies posting your stupid shit every time someone mentions... well, anything. Can you just get over yourselves already? Arguing about which LINUX distribution is better is like arguing over which type of obstructive bowel disease you'd rather get. Just because you can't figure it out because you are so stupid you can only use LINUX isn't our problem.

Re:Well,

[bhtooefr]

Well, the poster of the article DID ask for what distro was best on an Athlon 64.

OpenSSL

[m0rph3us0]

Compile an optimized OpenSSL for the Athlon-64. Sure it might not look pretty but I bet it can do RSA really fast.

Re:OpenSSL

[nelsonal]

distributed.net? OGRs were particularly Mac (altivec engine) friendly as I recall, a friend's 300Mhz ish imac would regularly post equivelent scores to up to 1 ghz k7 and P3s. I would assume that the 64 bit processor would be better at these too, the engine might need some optimization work done first though.

no don't

[lubricated]

By pushing a brand new car to the redline you risk voiding your warranty and causing serious problems later in the life of the car. Cars need to be broken in gently.

opcodes

[physicsboy500]

amazingly enough, the athlon 64 does not seem to have a vastly large number increase in opcodes. I believe they added another set to that long list of opcode types... MMX, MMX2, SSE, etc but the only real preformance increase comes from allowing it to access memory a double word at a time. the problem with that is it only really increases the preformance of floats and arrays if properly optimized.

Re:opcodes

[reynaert]

The most benefit will most likely come from the 8 extra general purpose registers the amd64 cpus have. The x86 has always been a register-starved architecture, and adding a few will result in a lot less register saving/restoring overhead. That alone will seriously increase performance.

Re:opcodes

[Anonymous Coward]

IIRC, they added SSE2.

What tools can be used to bech Linux period?

[schotty]

Other than compiling an app and timing the time spent, what are good tools in general that can be used?

Re:What tools can be used to bech Linux period?

[Carnildo]

The standard Linux benchmark is to time how long it takes to compile the kernel. Another benchmark I know of for Linux is the POV-Ray benchmark scene.

Try this

[kinnell]

int main(int argc, char *argv)
{
double i=0;
while ( i++ != 0xFFFFFFFF);
return 0;
}

Re:Try this

[Anonymous Coward]

You are fired.

Re:Try this

[KarmaPolice]

Well, 0xFFFFFFFF is the largest 32bit number you can write with an unsigned integer - or -1 for a signed 32bit number. Weren't we looking for things to do with 64 bit processors?

So I believe the code should be:
int main(int argc, char *argv)
{
double i=0;
while ( i++ != 0xFFFFFFFFFFFFFFFF);
return 0;
}

Re:Try this

[bluGill]

Note that you have to turn optimications off, or declare i as volitale or any half way decient compiler will optimise it out and turn your program into int main(int argc, char * argv) { return 0; } which is not what you want.

I would also recomend that you change that double to a long long or something (been too long since I've done 64 bit, so I don't recall the exact syntax), double is plenty of space to store [a rounded representation of] that number on most platforms such as x86, and most FPUs are good enough that there will be no difference. (A double fits in a x86 register on the FPU)

Not much

[MerlynEmrys67]

Ok, there are a few applications that 64 bits really help - some public key cryptography, and any application that has a process space over 4 GB (ok really 3 - but that is another story)

actually I don't know this... does Athlon 64 have a 64 bit x 64 bit = 128 bit result operation ?

The other applications that do really well with more memory are database type operations, where you are caching 16 GB of tables in memory so you don't have to go to disk (note, yes this means you need about 20 GB of RAM on the system)

Have fun

Re:Not much

[Glonoinha]

Big monster databases, assuming you loaded it with much memory as you could afford. You start doing work on a 5G database on a machine with 6G or so of RAM and can actually use it all ... and leave the Wintel boxes behind eating your dust.

Re:Not much

[burns210]

check if distributed.net has an optimized athlon64 client, or help in writing one, and then compare yourself to other's stats.

Re:Not much

[MerlynEmrys67]

Hate to say it, but a 5 G database isn't considered monster by any strech of the imagination... It will also run fine on an even slightly tuned 32 bit system...

What you want to look at is something well north of 100 G, oh and of that you want a 20-30 G working set... Now you are talking about something that can use 64 bits

Re:Not much

[Glonoinha]

A 5 gig database isn't a monster if we are talking Oracle or MS/SQL, but a 5 gig Foxpro database with associated index and memo fields is a monster.

But I agree with you ... along the lines of 'if you are gonna dream, dream big' I see where you are going : if you are going to throttle your box with a large database (in order to make it look good against a 32 bit machine,) go for broke.

Oh yea, and just like we saw when we went from 8 bit to 16 bit computing, and again when we went from 16 bit to 32 bit computing, going from 32 bit to 64 bit is going to afford us three things :
1. The first generation out of the box is going to probably run slower than the fastest release of the older chip. This because the compilers don't take the new stuff into account and the OS doesn't use the new stuff in efficient ways yet. Also because the old chip is generally running at a faster clock speed than the prototype new chips. P60 showed us that when the 486 chips were running 133 or so.
2. You get more memory to address, but until you actually buy and add the additional memory that potential is untapped. 8088 boxes peaked at 1M, 286 boxes peaked at 16M (so did the 386sx, IIRC), Pentium class boxes peaked at 4G, and the new 64 bit machines are going to peak at some unGodly amount of RAM that this time I guarantee nobody will ever be able to afford to totally fill one up and even if they did they wouldn't be able to use it all (sound familiar?)
3. Generally the next generation of chips scale higher than the previous generation of chip. 8088 : 12MHz. 286 : 25MHz or so. 386 : 40MHz or so. 486 : 133MHz. 586 : 233MHz. 686 : 450MHz or so. 786 : 1,400MHz. 886 (P4) : 3.2GHz. These new Opterons, and the next generation P5 (64bit?) chips : who knows. Granted a 1.4GHz PIII is faster than a 1.4GHz P4, but the P4 runs all the way to 3.2GHz or more.

Re:Not much

[randombit]

actually I don't know this... does Athlon 64 have a 64 bit x 64 bit = 128 bit result operation ?

It does (it's just like the 32-bit x86 multiply except it puts the 128-bit result into rax:rdx, instead of the 64-bit result into eax:edx). And so, yes, you can do sick fast PK crypto on it. Half the number of words => 1/4 as much work, cause lots of the underlying algorithms (like multiplication) are O(n^2), or sometimes O(n^~1.6).

Another note:

[Sevn]

You might also want to see if you can get your hands on the 64bit XP beta. I found some information here:

Microsoft Announces Beta Version of Windows XP 64-Bit Edition For 64-Bit Extended Systems

It looks like you can download the beta if you are an MSDN member only.

Uhhhh.....yeah

[gazbo]

Nice coding, dude, except that it will run forever. When you're using a test you use a DOUBLE EQUALS sign, we learnt this weeks ago in class and its pretty basic - your code will just assign 0xFFFFFFFF to the variable over and over.

I'm a bit confused why you'd want to assign a string to a double in the first place, but I guess in C everything is a number at the end of the day so it's still valid.

Re:Uhhhh.....yeah

[PaladinAlpha]

I applaud your interest in C, but stay awake in class.

The double equals sign is in fact a conditional operator, and the single equals sign is in fact an assignment operator, but the != is by no means a single equals sign. It is correct as stated, read simply "not equals".
Also, he is not assigning a string. He is assigning a hex value 0xFFFFFFFF. By prepending a value with 0x you are stating that the attached sequence should be interpreted as a series of hex characters -- 0-9 and A-F. If you haven't learned about hexadecimal yet, you soon will. Perservere.

Re:Uhhhh.....yeah

[Rakefighter]

Also, he is not assigning a string. He is assigning a hex value 0xFFFFFFFF...

Actually, he's not assigning anything...he's comparing the number 4294967295 to the value of i. Thing is, his code is a sort of a trick. The first time into the while loop, the test will pass (because 1 is not equal to 4294967295), the statement below it will be executed, and the program will exit (return 0). With proper brackets (because slashdot is stupid and won't let you indent), the code would look like the code below. The brackets after the while statement are not necessary if only one line of code follows (as in this example), and are assumed in the parent post's code.

int main(int argc, char *argv)
{
double i=0;
while ( i++ != 0xFFFFFFFF);
{
return 0;
}
}

Re:Uhhhh.....yeah

[skwirlmaster]

actually your wrong too...

if it was a:
while(i++ != 0xFFFFFFFF)
{
return 0
}
it would run once and then exit...
but because of the semi-colon right after the comparison, nothing is executed in the loop, it just loops, then once i is sufficiently large... it will return 0... although if it was void main, the return wouldn't be needed at all... :)

Re:Uhhhh.....yeah

[Anonymous Coward]

You need more class obviously.

Unless this is a troll, or you have literally had 2 weeks of C, you just embarassed yourself...

Re:Uhhhh.....yeah

[Rakefighter]

ARRGH! You're right! I totally missed that semi-colon... Please disregard my earlier post...

Re:Uhhhh.....yeah

[francium de neobie]

well actually, actually... you are wrong too... it won't compile at all ;)

Re:Uhhhh.....yeah

[shaitand]

actually your wrong, the original code will compile fine in most compilers, since the braces which are missing are implicit, it's simply a quickie while loop that increases the value in a double, the bracing in the original code is exactly as it should be.

it initializes the variable, declares a while loop which only accomplishes the autoincrement, looping until it reaches the proper value. Then when finished with the while loop it exits with a return value of zero, the return statement is not part of nor has any relation whatsoever to the while loop.

Re:Uhhhh.....yeah

[Paul Jakma]

actually... the first comparison would be of:

0 != 0xFFFFFFFF

The variable is evaluated /before/ the post-increment operator is applied. The value of i inside the loop first time round OTOH /would/ be 1.

Also, just for the fun of it, change the declaration of i from 'double i=0' to 'int i=-1' and see what happens (at least on x86, alpha, usparc, ppc), find out why, and dont ever forget :)

Re:Uhhhh.....yeah

[Glonoinha]

The first thing I thought of was FFFFFFFF being a negative number, but it wasn't until much later that I saw he wasn't doing a less than, he was doing a not equals.

Then I realized I was up pretty late tonight, and went to sleep.

Re:Uhhhh.....yeah

[francium de neobie]

no, man, your code really won't compile

> while(i++ != 0xFFFFFFFF)
> {
> return 0
> }

You've left out the semicolon after "return 0". Well just a minor mistake.

Re:Uhhhh.....yeah

[skwirlmaster]

Well the structure was the point :)
Slashdot comment boxes need 2 things
1) The ability to tab
2) A compile button :)

Re:Uhhhh.....yeah

[Trejkaz]

If it compiles it won't run. char *argv? Please.

Re:Uhhhh.....yeah

[shaitand]

This was the original code, and yes it compiles and executes properly.

int main(int argc, char *argv)
{
double i=0;
while ( i++ != 0xFFFFFFFF);
return 0;
}

Crazy?

[Anonymous Coward]

Why would you test the performance/handling limits of a new car in such a way. I understand the need to discern the envelope of your new car, but that is just dry/wet braking and evasive cornering. You can do both on a deserted well lit parking lot at low speed. But you can get an idea of these figures from something like Consumer Reports. Redlining the car on a rural road is just asking to hit a deer and speeding on a mountain road is going to get you to cross the yellow line into oncoming traffic which you cannot see. Or better sill, off the side of the road. I guess you wanted to see the maximum speed of your car so how does 180 off the side of a cliff sound. Please grow up.

Re:Crazy?

[Anonymous Coward]

Dork.

I bet your car *IS* your father's Oldsmobile.

See, there's this thing called a 'sports car', which is odd because you don't have to be any kind of athlete to buy one... but anyway, people buy them to drive 'em in a FUN way, and because some skeezy chicks are impressed by 'em... but anyway, these are cars that are designed and intended to be driven hard.

If it were true that 'speed kills', the countryside would be littered with the bodies of dead fighter pilots.

don't make everything x86-64 native

[Splork]

the benefit of the Athlon-64 / Opteron is that it is a blazingly fast x86 (32bit) CPU as well as having 64-bit stuff.

64bit code will be larger and gcc is currently much worse at generating code for x86-64. Only compile things that actually show a benefit from it as 64bit native if speed is what you desire.

An easy easy answer.

[tvadakia]

http://www.x86-64.org/ There's a lot of info there. Have fun.

Re:An easy easy answer.

[DA-MAN]

Yeah, aside from press releases....Try to find something useful that runs on x86_64 hardware. Aside from RH Beta, Gentoo, & SuSE, there isn't much out there for these opterons. Good thing they perform damn well in 32 bit.

You should already have something to test it with.

[jkc120]

Ok, so maybe this is a bit troll-ish, but if you've spent the money to buy one of these things, wouldn't you have a target _application_ to run on said platform?

I mean, the only thing you've proven is you can purchase something hot off the assembly line so that it can sit there and do 0 work for you. Don't you have an application or server or _something_ that you intend to use this machine for?

If not, I hope you have some kind of justification for this box for your management in their budget ;)

Oh yeah, I forgot.

[tvadakia]

Try this as well. Fun stuff.

ftp://ftp.suse.com/pub/suse/i386/x86_64

Great distro too.

Re:Oh yeah, I forgot.

[DA-MAN]

No Go!

ftp://ftp.suse.com/pub/suse/x86_64

That is the correct URL

Another distro to try

[SoCalChris]

Is here, although I doubt it is the kind of distro you're looking for :)

NAY!

[nlh]

Some would say otherwise.

Granted, it's not gospel, but sound in theory.

Bikers have a word for this

[gmhowell]

you're looking for a long, straight road where you can push the needles into the red before letting off the gas (and then maybe a twisty mountain road to test cornering)

No idea about auto drivers, but bikers have a word for this: squid. It's about as relevant as graphics card benchmarks.

See, anyone worth a salt tailors the race to their bike. I ride a touring bike. I'll race any super sport ten laps around Laguna Seca. But we start and end the race in Maryland. My father has a super-motardish bike. He'll race against any sport bike. But the road has to have potholes, lots of switchbacks, preferably some dirt road thrown in, etc.

Now, in an attempt to avoid being completely and 100% off topic, have you given any thought to custom compiling gentoo? Start with a stage 1 install, and compile everything from the kernel to glibc to the bash prompt. It shouldn't take long at all on your machine. I was able to do a KDE desktop on a P4 in just a few days.

Crunch some numbers!

[fatphil]

I organise a small number-crunching project (computational number-theory),
and if you'd like to race my code on your Athlon-64 verses everyone else's
PII/III/4s and Athlons, I can let you have the source, and maybe reserve a
chunk for you (it's not one of these flashy client-server setups). It assume
the usual GNU compiler toolchain, so any linux distro or similar would be
ideal. (Gentoo - get everything optimised for your system?)

Phil

Re:Crunch some numbers!

[JDWTopGuy]

Why don't you post the code somewhere publicly? Will it compile for a PPC?

Re:no don't, but hey, wait a min

[xgamer04]

That's kind of funny you would mention that, as when I bought my new car a few years ago they told me to drive it hard some to break it in. The whole "gentle break-in" thing is a myth.

hmmm

[josepha48]

(LINUX) cd /usr/src/linux && make -j .... see how long that takes ;-)... or (FreeBSD) cd /usr/src && make buildworld

SUSE for AMD64

[jfunk]

SUSE Linux Enterprise Server 8 for AMD64 has been around for a long time. That's why it was used in all of those Opteron benchmarks after it was released.

For something a little more affordable, SUSE 9.0 for AMD64 will be released tomorrow. (Along with the IA32 version I preordered)

For a no-cost alternative, you can download all 9 ISOs for the SUSE 8.2 for AMD64 beta here.

Mod Parent Down, Please

[Anonymous Coward]

This is just further proof of the corelation between neonazi IT professionals and Gentoo users. One quick look at this guy's history shows he is nothing more than a troll. *sigh*

How good is GCC with athlon64?

[Screaming Lunatic]

I run Gentoo at home and I'm loving it. But you won't be able to push the athlon64 to the max unless gcc can compile code that can take advantage of it.

Anyone have any experience/knowledge?

Re:How good is GCC with athlon64?

[Anonymous Coward]

"Anyone have any experience/knowledge?"

Well, if a link to goatse.cx and 6 replies stating "lunix si teh ghey" is what you're after, you've certainly come to the right place. If you want educated and professional opinion that's worth more than what SCO shares will be in 3 months time, then try elsewhere. Seriously.

Halp halp!

[Anonymous Coward]

Halp halp! I bought a shiny shiny new AMD-64 (INTEL SUX) and there's no software for it! *cry!*

-10: Flamebat && Troll

[dh003i]

Silly wabbit, don't you know that posting pro-MS comments on Slashdot is for kids?

Seriously, why anybody even bothers to post a recommendation of something MS on /. (aside from their MS Sidewinder or MS Wheel) is beyond me.

64 bit chess

[Anonymous Coward]

Some chess engines represent the chessboard using "bitboards". A bitboard is where you use every bit in a 64-bit unsigned int to represent the state of a square on the board. To represent the whole board you use multiple bitboards... for example:

bbWhitePieces = 11111111111111110000000...0b; bbRooks = 1000000100000000....0b; ...

To get the location of white rooks you would and the two bitboards above together. You'd have to store black pieces, knights, bishops etc... in their own boards. You get the idea.

So these programs tend to be a hell of a lot faster on 64 bit processors than on 32 bit machines because all this anding and oring can be done in one cycle rather than two.

One of the best free chess engines, crafty, is a bitboard program. So to answer your question - ftp a copy of crafty source code, build it with your handy dandy 64 bit amd64 compiler and run the benchmark command to see how fast it is.

Chess?

[WolfWithoutAClause]

Chess programs often express many aspects of a chess board in binary, with one bit per square. For example the attack pattern of a queen in the top left hand corner is:

$7fC0A09088848281 (I think I have that right, if you break that up into one byte per row on the chess board you can check it).

which neatly fits in one 64 bit word. It turns out that 64 bit processors are great for chess, significantly faster!

Re:You should already have something to test it wi

[JDWTopGuy]

dnetc

Re:no don't, but hey, wait a min

[Anonymous Coward]

Actually, it wasn't originally. It had to do with the way engines were built (tolerances) and the compression ratio of the engine.

It is true, that this is largely no longer the case. It is now much more true "drive it like you will drive it." You can actually ruin the engine now by driving it nice and gentle for 30K miles and then driving it like you will.

So, old cars: break them in gently!
New cars: drive it like you always will!

Chess on 64bit?

[igny]

There are 32 pieces on board (max), each requires coordinates on 8*8 board, which can be represented by 8 bits. Overall, a position requires up to 8*32=256 bits. How can it fit into 1 64-bit register?

Re:Chess on 64bit?

[Anonymous Coward]

Because each piece gets its own board...

So white rooks get....
010000....10000 ( 64 bits )
And black Knights get
001000....00010

And so you have 12 64 bit bit-boards for each type-color of piece. 2 x ( rook, bishop, knight, king, queen, pawns )
You have a board for each color and type of piece.

This makes it easy to see if a move causes a piece to take another piece. Just move your knights on the black knight bit board and AND it with the other White boards. If you have a one in a resulting board, it means your knight tooks something out.

Re:Chess on 64bit?

[acidrain69]

I think someone noted that you store a new array for each type of piece, and an 8*8 board, @ 1 bit per position is 64 bits. I don't quite understand it myself. I suppose you would have each 8x8 board assume the piece is in the middle of the board and show all the possible moves from THAT position, and just move it around the board to see what is possible. ie, the queen's matrix would look sort of like an asterik (an 8 starred one) originating from the middle of the board, and if the queen is in a corner, you just move the star, but then you have the problem of extending the diagonal move to the other corner, because the diagonal wasn't that long on the original matrix. And you'd probably need a special matrix for pawns, because pawns move one way and capture another. And another one for rooks and kings for castling moves. Damn. Chess is just one big special-case nightmare pit.

Re:You should already have something to test it wi

[drinkypoo]

Many people -- unlike me -- can afford their very own opteron system. The major stumbling block today is the motherboard, which will run you a good five hundred bucks. You can get a tolerably speedy processor for a mere $250 or so.

You could get your processor...

[Anonymous Coward]

... and stick it up your ass.

Re:You should already have something to test it wi

[gbulmash]

This is a personal purchase.

I've got some Win 32 bit games, but the main reason for this is that I've never owned a bleeding-edge, top-of-the-line system before. Could never afford one before.

With upcoming life events, it's unlikely I'll be able to justify something this new/hot/expensive for the next 20 years. So my wife agreed not to complain if I bought a badass system as a b-day present for myself.

So, yeah, it's a bit unjustified, maybe an early "mid-life crisis" toy. But it's better than buying a sportscar, divorcing my wife, and cruising the local community college for a 19-year-old bimbette to make me feel young.

- G

Yeah...

[jo42]

Start working on porting GIMP, OpenOffice and some other apps to 64-bitness

Re:You should already have something to test it wi

[randombit]

The major stumbling block today is the motherboard, which will run you a good five hundred bucks.

Yeah, but why bother when you can get an Athlon64 board for = $100? (Unless you want, and can afford, to put 2-4 of those little beauties all together, in which case the answer is obvious).

Re:You should already have something to test it wi

[drinkypoo]

Opteron also has more cache and, IIRC, a more fully developed memory controller. So there is a reason besides SMP - though SMP is obviously the driving force behind Opteron sales over Athlon 64, at this point.

dude, we're talking computers here!

[mabhatter654]

the worst that can happen is smoke rolling out of the machine! And he wasn't talking about OC'n so that shouldn't happen either. He just wants to know where his hard earned cash went and feel "warm-n-fuzzy" inside!

Dude, it's late, put down the caffine and step away from the coffee pot!