Need the Skinny on AMD Athlon 64 CPU Flavors

MERRICK

Howdy all!

regarding the AMD Athlon 64 CPU Can anybody tell me:

What is the difference between AMD CO version and CG Version CPUs?

What is the difference between AMD ClawHammer Core and NewCastle Core CPUs?

Does a 1MB L2 Cache give advantage over a 512 L2 Cache under 98SE?

Where can I find full comparison list of all Athlon 64 CPU's?

Thanks!!

SimGuy

CO -vs- CG:

CO is the first generation on-die memory controller and is much more picky about what memory modules can be utilized with the CPU. First generation ClawHammer's all were CO stepping and utilized the older memory controller. CO ClawHammer's don't overclock real well.

CG is the second generation on-die memory controller and is much more compatible with different memory types. First debuted on the Newcastle CPU's, the CG revision on-die memory controller has found its' way into the ClawHammer cores. All A64 3700's are CG-revision. CG ClawHammer's & NewCastle's overclock quite well.

ClawHammer:
Come in CO & CG steppings
1 MB L2 Cache.
Socket 754 only.
Available in: 3200, 3400 & 3700+ speed ratings (2.0, 2.2 & 2.4 GHz)
Being phased out & replaced by Newcastle core.

NewCastle Core:
Come in CG steppings only
512 KB L2 Cache
Socket 754 & Socket 939 CPU's
Available In 754: 2800, 3000 & 3200 (1.8, 2.0 & 2.2 GHz)
Available In 939: 3500, 3800 (2.2 GHz, 2.4 GHz)

1 MB of cache approximately equals the 200 MHz raw speed deficit between the ClawHammer & NewCastle CPU's. In mathematically-intensive applications (scientific research, SuperPi, 3DMark, etc.), at idential speeds, the ClawHammer performs faster than the NewCastle core's.

In gaming, the ClawHammer CPU's are ahead of the NewCastle CPU's at the same clock speed.

Overclocking wise, NewCastle CPU's overclock to higher raw MHz speed than comparable ClawHammer's (except if you get a CG Claw).

Full Comparison List:
http://fab51.com/cpu/guide/opn-64-e.html

//Edit: Wow. Today is A64's 1st birthday (it was launched Sept 23, 2003). Look how far AMD has come with the technology in only 12 months...

MERRICK

Hi SimGuy! Nice to see you after all this time.
Your answers are really helpful thanks so much! Hmm now let me see if I have my info correct for what I want to do:

Abit KV8 Pro MB

Crucial DDR 400 RAM (512MB chip, 256MB chip)

Win98SE*

Main Task: Calculating 64bit software digital signal processing on 32bit 96KHz (DVD Audio) wav files. I'd presume very math intensive.

(Based on pricing considerations)

AMD Athlon 64 3400+ (Will not be overclocked)
NewCastle Core (CG Stepping) 512K L2 2.4GHz
vs.
AMD Athlon 64 3400+ (Will not be overclocked)
ClawHammer Core (C0 Stepping) 1MB L2 3400+ 2.2GHz

I'd be inclined to say newCastle 2.4GHz over ClawHammer 2.2GHz. Unless I'm missing something? Any red flags here?


____________________________________________
*Why 98SE:
I have a self customised version of of Win98SE with many WinME system files and a Win95 explorer shell for standalone audio workstations. It is extremely fast and light (40MB install!). Though I use many 64bit plugins they are not written for a 64bit processor(yet).

primesuspect

You will not be able to do ANY 64 bit processing nor be able to access any of the 64bit features when using Windows 98. So, it's essentially pointless unless you are planning on creating a customized DAW-distro of WinXP 64bit edition or move this whole enterprise to a 64-bit build of linux.

MERRICK

primesuspect wrote:

You will not be able to do ANY 64 bit processing nor be able to access any of the 64bit features when using Windows 98. So, it's essentially pointless unless you are planning on creating a customized DAW-distro of WinXP 64bit edition or move this whole enterprise to a 64-bit build of linux.

Hi Prime!

To the best of my knowledge there is an exception. Certain digital signal processing (DSP) software plugin manufatuers have created plugins to to calculate the DSP (e.g. reverb, EQ, compression) in 64 bit float by crunching the numbers 2x through a 32 bit application. These manufatuers had these plugins on the market as far back as Windows First Edition. There are so many, here are two of the originals:

www.sonictimeworks.com
www.izotope.com

I understand that this is an exception to the rule. (I've never seen 64bit float mentioned in the same breath as 9.x anywhere else).

In the Audio Software world all the main Native DAW players (Cubase, Nuendo, Audition, Cakewalk etc.) are still pure 32bit platforms with not the slightest rumor of a 64 bit version, though I'm sure they're busy at work on it now.

primesuspect

Right, but my point is that having a 64bit processor will not make those calculations any faster than a 32bit equivalent - even if the DSP is doing fake-o 64 bit processing.

To enable the processor's 64 bit registers, you need to have a pure 64bit environment - the OS being the foundation, then the apps, and then the hardware drivers. all have to be 64bit.

So my point is, I wouldn't invest too much time in spec'ing out the 64bit platform because it won't do anything that a 32bit platform can't do.

MERRICK

Yes understood.
Okay let me take it from this angle (remeber I'm pure newbie please set me straight):

I started with the Motherboard and Chose Abit/AMD/VIA due to my consumer satisfaction.

I presume that SATA 150MB/s RAID, ATA 133 IDE are what they are regardless of O/S...

I presume 2 GHz FSB and RAM Speed are what they are regardless of O/S...

I presume 2.4GHz CPU speed is what it is regardless of O/S...

This blows away my KT7-RAID rig, right?

I'd love to get all this with a 32bit CPU but What chice do I have?

Do I have other options that you see?

Hellllp!!

primesuspect

Right, it sounds like a great system. I'm only saying that if money is a concern, you could build a very high performance 32 bit system using an athlon XP+ or a sempron, and save yourself money over the a64 platform.

However, personally, I would still go with the a64 platform, but don't think that you are going to be doing 64bit calculations. That "64bit DSP" stuff is marketing mumbojumbo.. It's still being done on a 32bit processor with a 32bit OS.

MERRICK

primesuspect wrote:

Right, it sounds like a great system. I'm only saying that if money is a concern, you could build a very high performance 32 bit system using an athlon XP+ or a sempron, and save yourself money over the a64 platform.

Oh Man I think it's back to the drawing board. I think I should revise this whole thing! I'm starting over 'cause the money is looking too high!

primesuspect wrote:

but don't think that you are going to be doing 64bit calculations. That "64bit DSP" stuff is marketing mumbojumbo.. It's still being done on a 32bit processor with a 32bit OS.

I'm not saying you are wrong I don't know. However in the six years that companies have been marketing 64bit DSP no one has ever challenged the claim. Is it unfathomable for a program to calculate how much/how long the "echo" will be on a snare drum that is 24 bits of audio. Normally the audio is 24 bits with 8 bits of calculations which "cycles" through the CPU 1x, or the program calculates the audio 24 bits with 40 bits and "cycles" through the CPU 2x. I've read a few papers on DSP calculations comparing different fixed and float resolutions (24,32,48,56, 64) using chips (SHARC, Motorola)
and Native CPU. I'd have to say that based on the various references I have seen, this is not mumbo jumbo.

Excellent article/resource:

Computer Do The Math by Oliver Masciarotte Mix magazine Feb 2003

http://www.seneschal.net/papers/bitstream/bstream035.htm

"Motorola 56k family of DSPs, 24 bit fixed point machines with 56 bit accumulators. A common choice for floating point DSP is Analog Devices’ SHARC family, which is a 32 or 40 bit floating point device with a 32 bit mantissa accumulator.

The 56k and SHARC are two common hardware examples but host–based, software–only DAWs largely use the CPU’s built–in, fixed or floating point processing. Since personal computers are general purpose devices, they can perform most any arithmetic operation they are called upon to do, though it may not happen as quickly as a purpose–built, hardware device."

I know one thing, My "64bit" programs take longer to process slow the machine down in real time, freeze the machine in real time or take a long time to process in offline mixdown mode. A faster CPU is definitly what I need.

primesuspect

Maybe we are talking about two different things. If the DSP utilizes the hardware DSP on the pro audio card that you are using, that could very well be 64bit. However, a 32bit OS is only capable of 32bit calculations, no matter what. I'm no engineer, so when you start talking about "cycling through" the CPU more than one time, I really have no idea if that's possible or what. It doesn't sound plausible to me, but then I really don't know. Someone with some CPU engineering experience will have to step in on this one.

MERRICK

As I re-read what I wrote I could clarify some things.
I am not saying that a 64 bit processor could in any way aid in these current plugins in any way especially speed.

I can say that I have done extensive comparison in listening to identical settings for 32 bit and 64 bit plugins. In general 64bit sounds better with a slight overlap between some companies (Waves 32bit EQ vs. Anwida 64bit EQ for example) which I attribute to just better code writing in the particular 32 bit application but overall the 64 bit applications sound better. with iZotope being the best (They give option to cut the frequency into 16,000+ bands as compared to 2048 by most others. It's mad I tell you!)

As I am not an authority I decided to do more research on this paradox:

This paper is much more explanitory though still over my head:
http://www.seneschal.net/papers/bitstream/bstream036.htm

As you say "Someone with some CPU engineering experience will have to step in on this one."

But that aside I appreciate your forthright assesment of what I'm doing and I have a feeling you saved me some dough.

primesuspect

Okay, I think you and I are both clear --- clearly lost

Just kidding. I think I understand what you are saying now. I think you should save some money and buy a hot-ass 32bit system instead of needlessly getting a 64bit processor at this point. If you have cash to burn, by all means go 64bit

MERRICK

Just to show you what a dumb ass I am, I knew I was eating 32bits from the start on the a64. Sure the FSB looked sweet, but remember I said I was out of touch for a while. Well the last big CPU I recall was 1.3GHz~1.4GHz. So I come back thinking I recently heard that 3.xGHz was coming to market So I go to Abit, and first mistake is I'm on Pentium page and think that Abit dropped AMD until I realize that there's an AMD page. I figure I'll drool at the top of the line 3GHz boards and then go down to a respectable 2.xGHz board. But the top board/cpu is 2.4GHz yikes! I can't believe CPU's have advanced so slowly? I presume that the 64bit 2.4GHz was the dope. I figured I was stuck with 32 extra bits that I have to pay for to get the 2.2GHz CPU.

That's all because I was impatent and I need a new DAW fast since the old one is really hurting.

So after you slapped me around I said let me go back one more time and make sure, and now I'm looking at 32bit 98SE heaven (KV7, SATA RAID 133 ATA, DDR 400, Barton XP 2.2 GHz 400FSB and for the right price).

Hey man you know I'm long winded, gotta catch up for time lost

Be cool!

mmonnin

Actually the floating point unit in your CPU is 80 bits wide. It calculates 80bits at a time even if its 1.1+2.2. They will both be encoded with 80bits. (2.2 actually cant be converted into a finite number if bits. X.2 as a decimal can never be expressed exactly in binary.)

I just started going over the floating point CPU so I am not sure if there is a way to get it to calculate several RAW 32-bit numbers into 1 or more 80-bit numbers. I would think there would be a way otherwise it would be a waste basically.