XP2600+333FSB - Barton? Thornton? T'bred?

haysdb · Post by **haysdb** » Wed Jan 21, 2004 10:46 pm

I have a 2500+ Barton, which F@H "benchmarks" at 804. In the same make and model board, a 2600+ benchmarks at 915, i.e. a 4% faster (speed rating) cpu with 14% better folding performance.

I have been calling this 2600+ a "Barton", but the invoice just says "AMD Athlon XP2600+333FSB", and the substrate is a different color than the 2500+. I assumed it was a Barton because the die appeared to be the same shape and size.

Newegg lists two 2600+ processors:
256MB cache, 2.08 GHz, Thoroughbred core, $88
512MB cache, 1.9 GHz, Barton core, $100

Is the 2600+ T'bred really a Thoroughbred, or is it a Thornton, i.e. a Barton with half the L2 cache disabled?

Any way you slice it, the 2600+ T'bred walks over the 2500+ Barton. According to the F@H benchmark anyway. But I tend to place some stock in this since the conditions are absolutely identical for the two boards - both booting the exact same Linux core and running an identical F@H exe.

Edit: A 2500+ Barton operates at a frequency of 1.83GHz. A 2600+ T'bred operates at 2.08GHz, or 13.7% higher. Is it a coincidence that the 2600+ T'bred benchmarks 13.8% higher than the 2500+ Barton? Is this proof that the extra cache does not help folding performance?

Edit: Read on. This is not the end of the story!

David

unregistered · Post by **unregistered** » Thu Jan 22, 2004 5:52 am

Here is a thread that may interest you. I'm sure that you can ask and get good answers here too. http://forums.amd.com/index.php?showtopic=5797

mas92264 · Post by **mas92264** » Thu Jan 22, 2004 7:53 am

A Barton core is clearly a different size than a T'bred core. Barton's are "longer." The pics of the 3200/400 Barton at Newegg are correct. The substrate can be either brown or green. I've got Bartons of both colors. I'm pretty sure that this means nothing.

AXDA2600DKV3D is a Tbred, where KV3D means 1.65V, 85 Centigrade, 256k cache, 333 fsb, respectively.

AXDA2600DKV4D is a Barton, where KV4D means same as above, except that 4 = 512K cache.

AXDA means Tbred/Barton Athlon XP for Desktop.

Just using common sense, it seems that Thortons are Barton "floor sweepings."

mas92264 · Post by **mas92264** » Thu Jan 22, 2004 8:01 am

Also, CPUID will identify running cpu's.

Model 10 = Barton/Thorton

Model 8 = Tbred

Model 6 = Palamino.

"Search" for cpuid at www.amd.com

Mutt_n_head · Post by **Mutt_n_head** » Thu Jan 22, 2004 9:28 am

I wouldn't ignore the difference in overclocking potential between a Barton 2500+ and the TBred 2600+, I would guess the Barton overclocks better with that core.

Just a thought.

sbabb · Post by **sbabb** » Thu Jan 22, 2004 10:42 am

I've seen F@H give benchmark scores that vary by several percent on the same machine. Do other people get varying scores?

haysdb · Post by **haysdb** » Thu Jan 22, 2004 11:03 am

sbabb wrote:I've seen F@H give benchmark scores that vary by several percent on the same machine. Do other people get varying scores?

On my server, scores are all over the map. But on the blades, the scores never vary by more than 2, e.g. a low of 914 to a high of 916, or a low of 804 and a high of 805.

Mutt_n_head, I wish I could say that overclocking was a factor, but so far none of the microATX boards I have tried have offered much in the way of overclocking features. If what I have found is correct, a Barton 2500+ would have to be overclocked by 14% to match a 2600+ T'bred.

mas92264, I would need to install Windows on a blade in order to use CPUID. Removing the heatsink would be easier.

David

mas92264 · Post by **mas92264** » Thu Jan 22, 2004 11:12 am

Ok, I give up. Where are you getting these benchmark scores?

Lockheed · Post by **Lockheed** » Thu Jan 22, 2004 2:33 pm

Post removed

sbabb · Post by **sbabb** » Thu Jan 22, 2004 3:29 pm

mas92264 wrote:Ok, I give up. Where are you getting these benchmark scores?

From the queue.dat file.

If you're using Electron Microscope, put your mouse over the green PC icon and a tooltip should pop up telling you that your core (ARRRGH! 1st Tinker in months on this PC!) is good. If you see a number below that (you might not, depending on a number of things) put your mouse over that number. If you don't see a number, just move your mouse down a bit below the PC icon. A tooltip should pop up that says "Number of work units in que." Click there and it will pop up a window with the info from the queue.dat file. The left and right arrows let you scroll through last 10 WUs. The benchmark score is next to last at the bottom.

On this PC (XP2000 laptop) I have benchmarks of 5208, 5244, 5176, 4008, and a bunch of 2636es. I guess the benchmark varies depending on what else I'm doing on the machine at the moment it benchmarks.

Scott

mas92264 · Post by **mas92264** » Thu Jan 22, 2004 3:29 pm

Lockheed,

Thanks for the info. I went over to amd and searched for cupid that would run under linux, and, no joy.

mas92264 (3rd highest weekly SPCR folder!)

Well, right now anyway.

haysdb · Post by **haysdb** » Thu Jan 22, 2004 4:36 pm

mas92264 wrote:Ok, I give up. Where are you getting these benchmark scores?

I'm getting mine from the FAHlog.txt files. I believe the benchmark score is not displayed unless -verbosity 9 is specified. The numbers I get under Linux don't equate with those I get under Windows. The Linux scores are smaller by factor of 7 or 8.

mas92264 wrote:I went over to amd and searched for cupid that would run under linux, and, no joy.

A link in his post points to a .tar.gz file, which is a Linux file. I assume he added the link after you posted this reply?

David

mas92264 · Post by **mas92264** » Thu Jan 22, 2004 5:10 pm

The Linux scores are smaller by factor of 7 or 8.

Ok, I see what you mean. Tbred 2600, 512 ram, XP Pro:

[00:54:05] + Benchmarking ...
[00:54:07] The benchmark result is 6800

haysdb · Post by **haysdb** » Thu Jan 22, 2004 6:59 pm

Lockheed wrote:There's no need to install windows just to run CPUID. Try http://www.ka9q.net/code/cpuid/

It took me less than 10 minutes to muddle through how to install it.

gunzip, tar, make, baddabing. The output is not pretty, but I now know the cpu in my server is

Model 6 (Palomino)
XP2000+

It says nothing about SSE, but it does have SSE.

David

ColdFlame · Post by **ColdFlame** » Thu Jan 22, 2004 9:44 pm

Palomino has SSE, at least mine does and I can't imagine them producing some with and some without

Lockheed · Post by **Lockheed** » Fri Jan 23, 2004 2:42 am

Post removed

Tobias · Post by **Tobias** » Fri Jan 23, 2004 5:05 am

What is the difference between a thorton and a barton?

mormakil · Post by **mormakil** » Fri Jan 23, 2004 5:11 am

I think a Thorton is basicaly a Barton with half the L2 cache disabled (it has 256kb instead of 512kb). I think they just needed more "TBreds" and used some "wrongs" Bartons or perfectly good Bartons to be able to sold them as "Tbreds". You can even make some trick and enable that disabled cache if it does works good.

kai · Post by **kai** » Fri Jan 23, 2004 10:56 am

you don't need to download anything to find out cpu info on linux, just type the following at the command prompt:

Code: Select all

# cat /proc/cpuinfo

this will list the processor, family, and all flags (sse, etc)

The following is my output:

Code: Select all

$ cat /proc/cpuinfo
processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 8
model name      : AMD Athlon(TM) XP 1800+
stepping        : 1
cpu MHz         : 1800.924
cache size      : 256 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mmxext 3dnowext 3dnow
bogomips        : 3547.13

haysdb · Post by **haysdb** » Fri Jan 23, 2004 12:03 pm

mormakil wrote:I think a Thorton is basicaly a Barton with half the L2 cache disabled (it has 256kb instead of 512kb). I think they just needed more "TBreds" and used some "wrongs" Bartons or perfectly good Bartons to be able to sold them as "Tbreds".

It's common practice to produce multiple parts from one die. VIA, for example, produce multiple chipsets from the same die, depending on whether they need an AMD or an Intel chip, S-ATA, and so on. The economies of making just one chip rather than several, more than make up for the incremental extra cost per chip.

What's interesting to me is that a 2800+ Barton becomes a 2600+ Thornton (2.08 GHz) and costs (at the moment) ~$50 less. Since the extra cache doesn't help folding performance [Edit: this was an assertion which has been shown to be wrong], that makes the Thornton the Folding value champion in my book.

David

haysdb · Post by **haysdb** » Fri Jan 23, 2004 12:04 pm

kai wrote:you don't need to download anything to find out cpu info on linux, just type the following at the command prompt:
Code: Select all
# cat /proc/cpuinfo
this will list the processor, family, and all flags (sse, etc)

Thank you kai!

David

ColdFlame · Post by **ColdFlame** » Fri Jan 23, 2004 1:42 pm

My Palomino 1.67 Ghz is about the same as my Duron 2.0 GHz so I think in case of Duron vs AthlonXP cache makes a difference. Maybe 256K vs 512K is less difference.

kai · Post by **kai** » Fri Jan 23, 2004 2:03 pm

Thank you kai!

David

You're welcome!

haysdb · Post by **haysdb** » Mon Jan 26, 2004 3:23 am

I have some additional confirmation that clock speed rules when it comes to folding performance, that cache doesn't play a large role. The benchmark scores that Folding@Home generates on my "blades" are remarkably consistant. Now, how well they correlate to real world performance in open to debate, but the scores are repeatable and consistant, so I think it makes them at least useful if not necessarily "the last word."

The "additional confirmation" is a T'bred 2400+ which just got added to the mix today. The "wildcard" is that this is yet another model of motherboard, this time an nForce2 board instead of the VIA KM400 in all my other boards. I will have to do a cpu swap with one of my other boards to determine how the nForce board compares to the KM400 boards with the same cpu. Here are the results so far:

Code: Select all

Biostar M7VIZ KM400   2500+ Barton 805
ASUS    A7V8X KM400   2500+ Barton 805
Biostar M7NCG nForce2 2400+ T'bred 881
ASUS    A7V8X KM400   2600+ T'bred 915

The performance of the 2500+ Barton is identical in two boards, lending credibility to the benchmark score of 805.
The same model board (ASUS) is running a 2500+ Barton and a 2600+ T'bred, lending credibility that the 2600+ T'bred is legitimately 13.7% faster than a 2500+ Barton. [Edit: ...on this one benchmark]

All boards are running at stock speed. Not so much as 1MHz of overclock on any of them.

How does the nForce board compare to the KM400 board with the same processor? I will have to do a CPU swap to find out. I think I will swap with my Abit board, which is underperforming with a 2600+ Barton (8% slower than the ASUS board with the same cpu).

David

haysdb · Post by **haysdb** » Wed Jan 28, 2004 1:33 am

mas92264 wrote:A Barton core is clearly a different size than a T'bred core. Barton's are "longer." The pics of the 3200/400 Barton at Newegg are correct. The substrate can be either brown or green. I've got Bartons of both colors. I'm pretty sure that this means nothing.

AXDA2600DKV3D is a Tbred, where KV3D means 1.65V, 85 Centigrade, 256k cache, 333 fsb, respectively.

AXDA2600DKV4D is a Barton, where KV4D means same as above, except that 4 = 512K cache.

AXDA means Tbred/Barton Athlon XP for Desktop.

I am looking at my 2500+ Barton and 2600+ T'bred side by side. Aside from the different substrate colors, they appear to be identical. The die are identical in size. Here are the part numbers:

Barton: AXDA2500DKV4D
T'bred: AXDA2600DK4VD

Perhaps the 2600+ really IS a Barton, but what I paid for, and what my invoice said, was just XP2600+, and the cost was $3 less than the 2500+.

David

haysdb · Post by **haysdb** » Wed Jan 28, 2004 3:47 am

Here is a summary of how three Athlon processors compare.

Motherboard: Biostar M7NCG 400 nForce2 microATX
Memory: 2x256MB (dual channel)
OS: Linux
F@H: 4.0

Code: Select all

                L2    FSB  Freq       Score
2500+ Barton   512MB  333  1.83        807
2400+ T'bred   256MB  266  2.00 +9.3%  878 +8.8%
2600+ Thornton 256MB  333  2.08 +4.0%  914 +4.1%

Edit: "conclusions" removed since they have proven to be wrong. All this shows is how these three processors run one particular benchmark, which may or may not bear any relationship with reality.

David

I have not taken time yet to experiment with the overclocking potential of the 2500+ Barton. That's on my list.

mas92264 · Post by **mas92264** » Wed Jan 28, 2004 5:28 am

Thoroughbred on the left and Barton on the right here.

mas92264 · Post by **mas92264** » Wed Jan 28, 2004 5:50 am

I assume that the "score" is the one generated by fah and posted in the log file. I'm going to guess that it (fah) sends some code to the cpu, the cpu whips it around some, then fah counts the number of whips and generates a corresonding number or score.

Therefore, if the benchmark code can fit in the 256k L2 cache, cache size makes zero difference and clock speed rules.

The only reason that I'm blabbering on about this is that my entirely unscientific study of the same protein main # (the rest of the protein id was different) on 4 or so computers, showed the bartons to be clearly superior.

This weekend, I'll try to capture one protein and run it on a Tbred and a Barton that I have at home.

haysdb · Post by **haysdb** » Wed Jan 28, 2004 12:13 pm

mas92264, it's true, this is just one single benchmark, and we don't even know how well it correlates with real world folding performance. It does seem to correspond fairly well with the PPD and PPW times reported by LogStat, but I haven't really analysed that very well.

I need to take the next step and compare the processors with the same proteins. EMIII can generate the numbers, which I can then pull into Excel and create a pivot table to compare the processors by Protein.

BTW, I have a Palomino, a "real" T'bred (a 2400+), a Barton, and the 2600+ which AMD calls a T'bred, but looks like a Barton, and is sometimes referred to as a Thornton

, so I know what they all look like now.

David

mas92264 · Post by **mas92264** » Wed Jan 28, 2004 12:38 pm

More on processing power:

A couple of weeks ago, I changed a 2400 Thorton box to a 2500 Barton, in mid wu. I expected a significant decrease in frame time.

Didn't happen. The frame rate was nearly identical, i.e. within 3 - 4 seconds.

However, this is 2.0 ghz vs 1.83.

Just trying to add a little more light on the subject.

M