AMD Barton or Thoroughbred

A forum just for SPCR's folding team... by request.

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

AMD Barton or Thoroughbred

Post by haysdb » Mon Dec 01, 2003 10:28 pm

The choices are confusing:
  • 2200+ Thoroughbred, 1.8 GHz, 256KB L2, 266MHz FSB, $67
  • 2400+ Thoroughbred, 2.0GHz, 256KB L2, 266MHz FSB, $78
  • 2400+ Thornton, 1.80 GHz, 256KB L2, 266MHz FSB, $75
  • 2500+ Barton, 1.83 GHz, 512KB L2, 333MHz FSB, $85
It doesn't make sense:
  • Thornton is a Barton with half of the L2 cache disabled, but
  • Barton is little more than a Thoroughbred with twice the L2.
In other words, T-bred = Thornton. Same amount of L2, same FSB.

It's supposedly the 512KB cache that "entitles" AMD to give the Bartons higher numbers than equivalently clocked T-breds, so I don't understand how the 1.8GHz Thornton can rightly be labeled a 2400+ when the 1.8GHz Thoroughbred is a 2200+.

OK, but the real question is, how much does the larger L2 cache benefit Folding performance? Is a 2400+ Barton a better folder than a 2400+ T-bred? The Barton has twice the L2, but the T-bred sports a faster core clock, at 2.0GHz vs. 1.83GHz. I guess I would expect them to be roughly equivalent.

There isn't a big price difference (between a 2400+ T-bred and a 2500+ Barton), so I suppose I will go for the Barton and the 333MHz FSB.

Comments?

David


BTE, I am researching this question, starting with Anandtech.

ColdFlame
Posts: 451
Joined: Wed May 21, 2003 9:39 pm
Location: Somewhere in Time

Post by ColdFlame » Mon Dec 01, 2003 11:23 pm

What I say might be misleading but:
1) in general, Barton > T-Bred, Palomino, etc.; Bartons also overclock way better, usually 2500 can reach 2.2 GHz
2) for folding, my experience is that Celeron folds quite nice and it is definitely lacking L2 cache. Which leads me to believe that L2 cache isn't that important for folding. I might be very wrong here. If I had identically clocked p4 and Celeron then I'd be able to say for sure. But my Celeron is really nice, especially at 2.9 GHz!

If I were you I'd get 2500 Barton just because it is a "safe" choice, also if you get a nice motherboard you'd be able to overclock that guy past 2.0 GHz for sure, surpassing T-Bred. I mean, Bartons are pretty much guaranteed to o/c well, especially since you gonna be getting a bunch of them.

haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

Post by haysdb » Mon Dec 01, 2003 11:56 pm

By your own logic (supported by AnandTech benchmarks), the additional L2 cache of the Barton core isn't likely to improve performance much, so the higher clocked processor could well be faster. That could well make a 2400+ T-bred faster than a 2500+ Barton. Again, this is borne out by benchmarks, which show that sometimes the Barton is faster, and sometimes the T-bred is faster.

It's irrational, but I have a "need" for a 2.0GHz core clock. For the same reason items are marked $1.99 rather than $2.00 at the grocery store, 2.0GHz just has a nicer ring to it than 1 point anything. I don't "mind" overclocking to achieve that goal, but my #1 priority has to be to Fold reliably. With a 2400+ T-bred, I wouldn't HAVE to OC to reach 2.0GHz. It's also well documented that VIA, in the past, has had trouble reaching high memory speeds. The KT333, for example, did not officially support DDR333 memory. My point is, DDR266 could well prove more stable than DDR333. That's pure FUD, but I have been around PC's long enough to know that the last generation of an old cpu/chipset is likely to be more stable than the first generation of a new cpu/chipset. So, I am wondering to myself, stick with T-bred, 266MHz FSB, and DDR266 memory, at least for the server and first client, then maybe try Barton, 333Mhz FSB, and DDR333 for the next pair of clients, or just take my lumps now?

David

mormakil
Posts: 187
Joined: Sat Oct 04, 2003 1:43 am
Location: Madrid, Spain

Post by mormakil » Tue Dec 02, 2003 2:42 am

I'm quite sure you won't have any problems to make a 2500 Barton work at 2.0Ghz. Even if you memory is not that good and will not oc (as happen with my which starts to lower its bandwith at 169MHz) you just could modify the multiplier. With 166*12 you'll have 1992, and with 167*12 you're above your 2.0 Ghz "psicological limit" :).

Right now I'm testing the stability of my system at 168*11.5 (1932Mhz) and I've got no issues (I've even tested with Prime95+F@H for 3 hours with no problems, not too much time but I didn't want to lower my ppd :D). Next step will be to try 12x multiplier with same FSB = 2016Mhz.

Another thing you should consider is the rumour that Bartons are starting to be sold speed limited (maybe they don't want people buying the same for less money) something about procs built after week 39 being speed limited. You should look at some overclocking forums about this issue.

futureweaver
Friend of SPCR
Posts: 109
Joined: Thu Jul 03, 2003 1:50 am
Location: Reading, UK

Post by futureweaver » Tue Dec 02, 2003 3:36 am

I'm running a 2500 Barton as a 2800 with no problems - 166x12.5 = 2.08GHz. AIDA32 and even AMD's own CPUID say it's a 2800, so I guess it's physically identical. It will go faster, but then it gets a bit warm for my taste.

haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

Post by haysdb » Tue Dec 02, 2003 12:48 pm

As a far as being speed limited, all they can do is fix the multiplier, just like Intel has been doing forever, but that hasn't stopped the overclocking of Intel chips. Alternatively, just yesterday I saw a little doo-dad that you slip over several of the pins before inserting the cpu into the socket, which unlocks the multiplier. Not sure, but I think that was at Directron.

David

mas92264
Patron of SPCR
Posts: 659
Joined: Fri Sep 26, 2003 5:26 pm
Location: Palm Springs, CA, USA

Post by mas92264 » Wed Dec 03, 2003 6:58 pm

David,

Whilst trying not to be overly picky, the 2400 Thorton is actually 2000 ghz - check out the specs for model 10/256 cache. Incidentally, the word "Thorton" does not show up on a search at AMD. Weird. Thorton is apparently the correct spelling, though. And, the only apparent differences between a T'bred and Barton/Thorton are cache size and that Bartons/Thortons have "S2K Bus Disconnect" which is an enhanced cpu cooling/slow down deal. S2K, however, must be implemeted by the mobo mfg from what I just read.

Don't know how much difference cache size or fsb speed makes with respect to folding. Would be nice to know, though. It seems that cache size makes no or little difference?

However, it's hard for me to justify buying a 266 chip for a 333 board. So, to me anyway, it just makes more sense to just get a 2700 t'bred for a few $ more - thinking primarily about folding. For a 266 board, the 2400 Thorton makes the most sense $ wise.

The mystery is, however, just how much folding performace difference is there between a 2700/333 and a 2400/266? (expanding on your thoughts about 2500 Barton vs. 2400 Thorton)

haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

Post by haysdb » Wed Dec 03, 2003 10:53 pm

mas92264,

Thanks for the correction. I haven't verified it for myself, but it sure makes logical sense that a Thorton and a T-bred, with the same 256MB L2 cache, would be clocked the same for the same "plus" rating. There's really no reason to pick one over the other then, except for the premise that the Barton overclocks better.

Is there a 2700 t'bred? If there was, it would be a 266MHz FSB wouldn't it? Or do you mean Barton? Geez, I can't keep all these overlapping ratings and clock rates and names straight.

For me, if I do go through with this 'Farm' thing, I will go with the Bartons, just because they should be a wee bit faster, and the price difference is just not enough for me to worry about. Even $85 is a hell of a price for that kind o' juice.

David

mas92264
Patron of SPCR
Posts: 659
Joined: Fri Sep 26, 2003 5:26 pm
Location: Palm Springs, CA, USA

Post by mas92264 » Thu Dec 04, 2003 7:30 am

David,

2700/333 T'bred = 2.17 ghz
2600/333 T'bred = 2.08 ghz

Amd specs arehere (page 27 of the pdf doc.) The 2700 is $109 at ZZF and $113 at newegg. The 2600 (if you can find one) is a few $ less. ZZF says it has them for $99. Note that the text product descriptions at newegg can be incorrect.

It's hard to make a case against buying Bartons for a farm - although the recent news about the multiplier being locked on new shipments is not good. I've got one of those Speedstrip things, still in the package. However, at $15 it's, well, just too much for me if I was considering buying several chips. "Course the fsb can be increased, but that requires 3200 memory for a significant increase. It's like chasing your tail. :)

mas92264
Patron of SPCR
Posts: 659
Joined: Fri Sep 26, 2003 5:26 pm
Location: Palm Springs, CA, USA

Post by mas92264 » Thu Dec 04, 2003 7:35 am

Oh, and I'm > 10k points this a.m. :P

mas92264
Patron of SPCR
Posts: 659
Joined: Fri Sep 26, 2003 5:26 pm
Location: Palm Springs, CA, USA

Post by mas92264 » Thu Dec 04, 2003 9:32 am

Thorton specs are here. Page 2 of the pdf doc.

dukla2000
*Lifetime Patron*
Posts: 1465
Joined: Sun Mar 09, 2003 12:27 pm
Location: Reading.England.EU

Post by dukla2000 » Thu Dec 04, 2003 9:46 am

David

AFAIK for Folding with Athlons you need MHz. Memory, fsb, cache et al make no difference. I hope this is verified elsewhere - I know there are some bits in various places. My experience with 2 systems is I have identical points/week potential from
15*133 with SDRAM @ 133
12*166 with DDR 400

(The 15*133 box is currently running with DDR 266 RAM, still Folding as fast as the 12*166. The 12*166 is 'my' system and gets the KT600 mobo and all the other 'best' bits, the 15*133 is hand me downs with KT333 mobo.)

So for a pure Folding CPU, the biggest bang for the buck is more likely to be a high speed Thorton IMHO.

(And I will probably be demonstrated wrong when the Folder with a Barton & a Tbred posts their points/week in that thread. My expectation is a Barton 3200+ @ 2.2GHz clock is similar Folding to a Tbred 2700+ at clock 2.167GHz :oops: )

ColdFlame
Posts: 451
Joined: Wed May 21, 2003 9:39 pm
Location: Somewhere in Time

Post by ColdFlame » Thu Dec 04, 2003 11:42 am

I hope this review might help: http://www.anandtech.com/cpu/showdoc.html?i=1927

One thing to keep in mind is that my Celeron isn't nearly as bad as those results state. In fact, my Celeron 2.9 Ghz is about 2-3 times faster than p3 1 GHz and about 20%-30% faster than AMD 2100+ XP. I mean faster for folding.

mas92264
Patron of SPCR
Posts: 659
Joined: Fri Sep 26, 2003 5:26 pm
Location: Palm Springs, CA, USA

Post by mas92264 » Thu Dec 04, 2003 12:03 pm

I hope this review might help.
Interesting article. The red headed stepchild Duron pretty much whupped the Celeron - in the tests that were run. The only remaining question is whether or not the Model 8 Duron (Applebred?) contains the complete SSE set (for an inexpensive folder.) Have an email to amd re this and no answer yet.

Applebred. Who the heck comes up with these names? Must have either been in a hurry, hungover or both. :?

ColdFlame
Posts: 451
Joined: Wed May 21, 2003 9:39 pm
Location: Somewhere in Time

Post by ColdFlame » Thu Dec 04, 2003 1:32 pm

Well, my Athlon XP runs at 1.73 GHz currently (overclocked 2000+ XP) and my Celeron 2.9 GHz whoops it on folding. Which contradicts the tests in that article.

haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

Post by haysdb » Thu Dec 04, 2003 7:28 pm

Yeah, I definitely think they should have run a few more benchmarks. I don't believe the Intel chips were given a fair chance.

:roll:

The 2400+ held its own quite well. In fact, there wasn't a lot to separate the top three AMD chips. In many (most?) of the tests, the higher clocked Athlon performed better, regardless of cache size. So, which is the better Folder, a 2600+ T'bred or a 2500+ Barton? I'm guessing the T'bred, but as ColdFlame's reply alludes to, those benchmarks don't necessarily apply to FAH.

David

ColdFlame
Posts: 451
Joined: Wed May 21, 2003 9:39 pm
Location: Somewhere in Time

Post by ColdFlame » Thu Dec 04, 2003 8:40 pm

I guess you won't know which is a better folder 'till you try :)

I have however an interesting theory. Let's assume TBred is a better folder. You gonna lose about 4% of performance. I think it is no big deal.

Now, TBred is clocked significantly higher than Barton. Hence Barton has a much higher chance to be overclocked past TBred speeds, potentially reducing or (most likely) eliminating the gap.

Summary: I think Barton is a better buy overall, even if it appears to be a slower folder than TBred at stock speeds, you can always o/c it. Worst case - you lose 4% perf.

Bottom line: It does not matter which one :)

haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

Post by haysdb » Thu Dec 04, 2003 9:03 pm

ColdFlame wrote:Bottom line: It does not matter which one :)
Oh I agree with you. It's not really something I will agonize over for very long, and if I get my Linux server up and going and decide to go forward with the network boot clients, I might just buy one of each and, indeed, find out for myself.

David

haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

Post by haysdb » Mon Dec 08, 2003 9:17 pm

I couldn't stomach a 2600+ T'bred for $100 so went with a 2500+ Barton for $86. The biggest bang/buck is probably a 2400+ T'bred, but I was fixated on getting a 333Mhz FSB, so my choices were one of the above or overclocking a FSB266 part. That option was out though because the mobo I chose (Biostar M7VIZ) has only limited options for overclocking.

David

ColdFlame
Posts: 451
Joined: Wed May 21, 2003 9:39 pm
Location: Somewhere in Time

Post by ColdFlame » Mon Dec 08, 2003 10:29 pm

So how many you got? 4? :)

haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

Post by haysdb » Mon Dec 08, 2003 11:06 pm

How many what? This will be my second Athlon, to go with three P4 systems.

David

ColdFlame
Posts: 451
Joined: Wed May 21, 2003 9:39 pm
Location: Somewhere in Time

Post by ColdFlame » Tue Dec 09, 2003 11:31 am

I thought you were trying to figure out the CPU for the folding farm.

haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

Post by haysdb » Tue Dec 09, 2003 11:18 pm

ColdFlame, remember that when you speak to me you have to use short sentences, small words, and if it has been more than 5 minutes since you said something (or *I* said something), quote it to refresh my memory.

This cpu, the 2500+ Barton, is destined for my first "diskless Folding client". After that's up and running, I will evaluate where to go from there. I should do something "different" for the second one, like maybe go ahead and try a 2600+ T'bred or 2400+ Thornton so that I have something to compare to.

Once the first client is up and running, the second will be ordered straight away.

David

Harry Azol
Posts: 228
Joined: Fri May 02, 2003 9:21 am

Post by Harry Azol » Wed Dec 10, 2003 11:17 pm

FWIW, I have a tbred 1700+ which I overclocked to 2.2ghz (2800+ speeds) at stock voltage.. With the Zalman 7000A, my temps are under 45 C under load

not bad for such a cheap cpu.. :)

haysdb
Patron of SPCR
Posts: 2425
Joined: Fri Aug 29, 2003 11:09 pm
Location: Earth

Post by haysdb » Wed Dec 10, 2003 11:36 pm

Very impressive indeed. And the temps are even lower than my non overclocked P4's.

Unfortunately the Bio-whatever motherboard I bought doesn't offer much in the way of overclocking options.

David

Post Reply