Sign in with
Sign up | Sign in
Your question

AMD64 up to 10x faster as P4 !

Last response: in CPUs
Share
Anonymous
a b à CPUs
February 4, 2004 11:43:50 AM

Okay, the title may be just a tiny bit misleading, but it is still true for at least one app; have a look here:

<A HREF="http://www.sun.com/amd/30579_AMD_Processor_Evaluation_G..." target="_new">http://www.sun.com/amd/30579_AMD_Processor_Evaluation_G...;/A>

Obviously most of the 64 bit benches are encryption stuff, one of the only apps that really benefit immensly performance wise from 64 bit integer math (intel showed the same apps to prove Merced was a screamer). But DivX also shows a nice gain (+18% !), as does ZIP (+100%).

But now everyone say with me:"64 is NOT about (much) better performance, it is NOT about supporting more than 4GB Ram, it is about having a flat memory address space, breaking the 2 GB virtual memory brick wall we are all heading straight into."

= The views stated herein are my personal views, and not necessarily the views of my wife. =

More about : amd64 10x faster

February 4, 2004 3:22:24 PM

just ask you sometime how it can be 10 faster

Just to show bad
February 4, 2004 3:45:08 PM

Well... I have seen that before in ace's forums.

However, it's not truly representative. I can bet that Itanium 2 can destroy - and I mean literally destroy - AMD in this field: encryption. It's quite possible that even the 1.4Ghz, 1.5MB cached Deerfield will take the lead here...

Just to bitch around a bit: 10x??? This is misleading. AMD64 can be up to 3 times faster, not 10! I take it you were probably joking with 10x.

However, we cannot forget that encrypting is <i>the</i> single 64-bit integer task - it should show 64 bit tech at its very strong point. It is not representative of overall performance gains from going from 32 to 64 bit at all, and it would be very simplistic to assume so. And personally, as interesting as encryption techniques may be, I don't care about them. :frown:

ZIP performance is nice to know, though. But quite frankly, I really want to know how truly high-powered apps behave under 64 bit. Like rendering, gaming and encoding, or even 3d modeling. I take it that many benefits are to be gained indeed, mostly for 3d modeling, rendering or programs that would enjoy the "absence" of virtual memory. Gaming... well, I'm not technically sure about one thing exactly that would make gaming faster under 64-bit...

BB, could you please explain to me what exactly is it that makes 64 bit faster under certain conditions?... This might sound silly, but...
Related resources
February 4, 2004 7:46:36 PM

Quote:
However, it's not truly representative. I can bet that Itanium 2 can destroy - and I mean literally destroy - AMD in this field: encryption. It's quite possible that even the 1.4Ghz, 1.5MB cached Deerfield will take the lead here...

SPECweb99_ssl disagrees.

<A HREF="http://www.spec.org/web99ssl/results/res2003q3/web99ssl..." target="_new">Click:</A> HP Integrity rx5670--2x1.5GHz Itanium 2/6MB, 16GB RAM
<A HREF="http://www.spec.org/web99ssl/results/res2003q4/web99ssl..." target="_new">Click:</A> Rioworks HDAMA/Opteron 248--2xOpteron 248, 16GB PC2100 <font color=white>.oO(PC2100? WTF?!?)</font color=white>

Granted, all things aren't quite equal, but it's liable to be as close as we'll get without a real head-to-head benchmark.

So why does Itanium2 get beaten? Probably because that 64-bit integer math that's so important to SSL just plain sucks on IA64. It's fairly common knowledge that IA64 doesn't do integer math very well, 64-bit or otherwise. The thing IA64 really excels at is floating-point, which really has nothing to do with the part's 64-bitness.

I find it somehow ironic that the "true 64-bitness" touted on IA64 apparently doesn't stack up to the "hacked 64-bitness" of AMD64. I think we'd better just consider IA64 as more of a floating-point powerhouse rather than a 64-bit powerhouse.

<i><Lionel Hutz> I'll be defending...The SCO Group!!!??? Even if I lose, I'll be famous!</i><P ID="edit"><FONT SIZE=-1><EM>Edited by kelledin on 02/04/04 03:50 PM.</EM></FONT></P>
Anonymous
a b à CPUs
February 4, 2004 8:19:05 PM

>However, it's not truly representative

Off course not. Its marketing.

> I can bet that Itanium 2 can destroy - and I mean
>literally destroy - AMD in this field: encryption

I've looked for comparable benchmarks, but came up empty. So while it could be true, I'd reserve judgement until we have some hard numbers.

>Just to bitch around a bit: 10x??? This is misleading.
>AMD64 can be up to 3 times faster, not 10! I take it you
>were probably joking with 10x.

No, 10x faster.. have a look at the PDF, last page "optimized RCA decrypt etc". 1020% the performance of a P4 3.2C. In 32 bit, it is "only" 224% faster. Off course this is not represantive..

>However, we cannot forget that encrypting is the single
>64-bit integer task - it should show 64 bit tech at its
>very strong point. It is not representative of overall
>performance gains from going from 32 to 64 bit at all, and
>it would be very simplistic to assume so.

That is what I said, never claimed anything else.

>BB, could you please explain to me what exactly is it that
>makes 64 bit faster under certain conditions?... This might
>sound silly, but.

Oh dear :| could you then explain me the meaning of life ? :) 

Well, let's see. first of all, I want to stress that performance as such is *not* the biggest advantage of 64 bit computing. Its the flat memory addressing, more than 2 GB addres space for your apps, no more fragmented virtual memory leaving you with just a couple hundreds of megabytes of continues memory.

That being said, a 64 bit cpu can be faster in some cases, especially the K8. First off all, there is like you mentioned, apps that require math on integers bigger than 32 bit. A 64 bit cpu can crunch these in theory up to 4x faster (in the case of 64 bit multiply which can done at once, instead of in 4 steps like on a 32 bit cpu). However, these apps are rare: mostly encryption, mathematical/scientific apps or simulations. not something you'd run on a daily basis (unless perhaps those Keygen cracks :) .

Then there is the memory addressing issue; a 64 bit app under a 64 bit OS can access all of its memory directly, whereas on x86, a process is limited to 2 or 3 GB period. Hard to put a performance estimate on this, but if an app requires more than what is available, it should launch serveral process and the OS is forced to use PAE and bankswitching crap that is slow as hell and a PITA to develo for. I can't give hard numbers on this, but I would guess memory access could be as much as 5x slower using PAE and passing data from one process to the other. Not to mention, certain apps just can't be programmed around this at all. This will be most apparent in server apps (ERP, Databases, etc,..) and is the reason every other server ISA except x86 has moved to 64 bit ages ago. On the desktop this will be less apparant, since no desktop apps that I am aware off support PAE or /3GB in the first place. The difference will not be performance, but rather: it will work instread of not work. (think photoshop using huge images, and other memory hogs)

Then there are some specific advantages to AMD64. First of all, AMD extended the register set from 8 to 16 (of those, 5 I think are true general purpose registers, so the real increase is from 5 to 15). depending on the code, I would guess this might give you a ~5-10% performance boost.

AMD also extended SSE2 with 8 additional 128 bit FP registers. SSE2 performance therefore should be noticeably better under AMD64 in 64 bit long mode than in legacy 32 bit mode. I don't dare give an estimate though.. but this could have a noticeable impact on SSE2 dependant code (games, 3D rendering, encoding,..).

Lastly, I expect a speedup because of K8 specific compilation optimizations. Currently, most software is optimized for a wide range of cpu's, going from P2/3 to P4, Athlon, A64, etc.. Code compiled for AMD64 should not enable switches to remain compatible with (or optimized for) older cpu's, since K8 is the only AMD64 cpu anyway. I would bet currently most new applications are compiled with the P4 as performance target, not the K7 (which is logical given the marketshare). For 64 bit code, this won't be true anymore, performance optimization target will be the K8, and this alone might lead to more significant speedups than everything mentioned above, even though it has nothing to do with "64 bit" as such. Its just the benefit of the market leader (in this case, the prettty small 64 bit x86 market).

But AMD64 also has its drawbacks, not everything will speed up. Two things: when you use a 64 bit OS, and if you run an app that doesnt need 64-bit data you can use the old 32-bit instructions just fine and only use 64-bit ones as necessary. However, a pointer is always going to be 64-bits in AMD64 long mode, so some applications might take a slight performance hit because of the extra strain on the memory bus caused by loading 64-bits for every pointer instead of 32. I would expect the extra registers to offset this possible disadvantage, but quick & dirty recompiles could actually give you a minor performance hit, instead of an advantage.

A second disadvantage is the fact that ICC (intel compiler) doesnt support AMD64 (yet ?). ICC is probably the compiler that generates the fastest x86 code. It is not widely used except for "benchmark software" and of course SPEC, but some cpu intensive apps (think 3D render cores, divx codecs, maybe even GPU drivers,..etc) might well be compiled with ICC because of the performance. Porting to AMD64 will require using other compilers (Microsoft, GCC, Portland, ..) which might not give you as fast binaries. Especially SPEC (where ICC shines) scores will be affected, but some often used benchmark programs might take a hit as well. Your average game OTOH, is not likely to have been compiled with ICC.

Now, you where wondering, what will this give me for "my apps" ? the answer is: I don't know. I am not expecting leaps and bounds, even though I just saw some results under XP 64 that where just incredible (like more than double the performance using DivX). Overall, I am prudent, and do not expect more than ~10%. In same cases, I even expect a small loss of performance, while a few other specific programs might give spectacular increases. For gaming, I doubt we will see spectacular increases in FPS, but I firmly expect future games to enable you to select higher details, bigger maps, etc on a 64 bit CPU/OS. Compare it to DX8 verus DX9; not really faster as such, but more eye candy and more flexibility for developpers.

Hope this answers your question.. now, you tell me, what is the purpose of life ?
:D 

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 4, 2004 10:05:17 PM

Ooooooo man u are making me want to install 64 bit linux on my friend's comp and run one of my simulations! This could get very very sick!

The one and only "Monstrous BULLgarian!"
February 4, 2004 11:10:21 PM

When you run in what is called “Long Mode” or what a 64 bit OS would run in, you have 2 modes, “64 bit mode” and “Compatibility mode” A 32 bit program would run in compatibility mode and have an address pointer size of 32. Other than thunking necessary for 64 bit OS calls, it would have the exact memory footprint as a normal 32 bit program.

Dichromatic for your viewing plesure...
February 5, 2004 12:13:01 AM

Thanks, BB. I knew about a number of those things; however, hearing them again made me organize my thoughts. And this transition looks promising.

Intel better follow suit indeed. :mad:  And alacrity would be most appreciated.

As for the meaning of life...

I'll let you know when I find out myself. But searching for a meaning is already a good step.

What is a meaning, anyway? Doesn't it take understanding to attribute meaning? If there is no understanding without life, then how can there be a meaning? There would be no need to be one; think about it - the mere fact that we exist would be the sole cause of our looking for the meaning, and hence would deny much of what we'd call a meaning. There is none, in the greater sense, I suppose.

Question is, what is the meaning of life <i>for each one of us?</i>

(Just having fun here) :cool: Cooked this up myself... :eek: 

But don't ask me this tomorrow again - I'll probably answer something completely different. It's one of those moment things.
February 5, 2004 12:34:42 AM

Dude.....the meaning of life is like....pi....warm apple pi....

The one and only "Monstrous BULLgarian!"
February 5, 2004 2:00:30 AM

"Meaning of Life" is the most mediocre Monty Python move.

Dichromatic for your viewing plesure...
February 5, 2004 3:07:21 AM

what if the meaning of life was to find that the meaning of life is finding the meaning of life????? That's an infinite loops or some shite. At this moment shiny objects appear very attractive....oooo

The one and only "Monstrous BULLgarian!"
February 5, 2004 3:28:36 AM

bbaeyens keep up the good work,

I have been catching up for the last few hours. I find your posts interesting in that I'm learning (or at least willing to admit i'm always learning). Not that you can't be wrong but I like how you have the balls to say what you believe and it looks like your not just spewing crap.

you call it like it is and don't let anyone push you around.

Well done. Keep it up I like to learn.

If I glanced at a spilt box of tooth picks on the floor, could I tell you how many are in the pile. Not a chance, But then again I don't have to buy my underware at Kmart.
Anonymous
a b à CPUs
February 5, 2004 5:15:50 AM

>When you run in what is called “Long Mode” or what a 64 bit
>OS would run in, you have 2 modes, “64 bit mode” and
>“Compatibility mode” A 32 bit program would run in
>compatibility mode and have an address pointer size of 32.
>Other than thunking necessary for 64 bit OS calls, it would
>have the exact memory footprint as a normal 32 bit program.

True (I guess) for 32 bit code running in long mode, but 64 bit code will be zero filled, when using 32 bit instructions, and therefore use 64 bit pointers instead of 32. Not sure how much of a performance penalty that would give though.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 5, 2004 5:20:36 AM

The meaning of life is ofcourse 42, the problem then is what is the true question.
Anonymous
a b à CPUs
February 5, 2004 5:46:29 AM

I saw it, but I am not believeing it just yet. AMD itselve only shows a 18% performance increase in DIvX in their handpicked 64 bit showcase, so i am a bit reluctant to take these results for what they are...

= The views stated herein are my personal views, and not necessarily the views of my wife. =
Anonymous
a b à CPUs
February 5, 2004 6:00:50 AM

thanks mate

= The views stated herein are my personal views, and not necessarily the views of my wife. =
February 5, 2004 6:11:01 AM

It is PCmark04 afterall... :\
They have some pretty good ideas as to why it performs THAT well but we shall see soon enough with some real encodes.
February 5, 2004 8:43:38 AM

Quote:

The meaning of life is ofcourse 42, the problem then is what is the true question.

For those of you who want to know how that answer came about, check out Douglas Adams' "Hitchhiker's guide to the galaxy" (a trilogy in four parts). And no, it will not help you in finding out what that true question is, but that will become apparent at the end of the series.

As to all of you nimble knuckle posters and up, I guess you already found out the meaning of your life and I hope you are enjoying it to the fullest. At least I'm having a great time reading it all.



BigMac

<A HREF="http://www.p3int.com/product_center_NWO_The_Story.asp" target="_new">New World Order</A>
Anonymous
a b à CPUs
February 5, 2004 10:47:42 PM

>As to all of you nimble knuckle posters and up, I guess you
>already found out the meaning of your life and I hope you
>are enjoying it to the fullest.

:D  subtle, but oh so true ;) 

= The views stated herein are my personal views, and not necessarily the views of my wife. =
!