The AMD Atlhon XP was always the most efficent CPU out there, running against Intel's onslaught of Pentium 4s at a lower clock speed - and usually meeting or beating them!

The new Barton series of Athlon XPs, now shoehorned with 512 kb of L2 cache, is clocked a bit lower than the Throughbred B. AMD decided that the extra cache boost is worthy enough to add a higher model number. Ouch!

AMD is was half right, and half wrong. The results were stunning in a few benchmarks (memory and cache intensive), but any thing else that was made for pure clock speed, such as video and audio compression (encoding), choked on the slower clock speed alone. This is where the Pentium 4 took honors.

Alas, this is not were the Athlon XP Bartons succeed. It is in it's high overhead for overclocking! Though the gap closes between the Barton and the P4 with this dark art, it is still a pretty wide gap to close, and therefore, never actually get closed.

The 2500+ runs at 1.83 GHz, where as the T-Bred B 2400+ ran at 2 GHz. But the larger cache does help a bit, though many (including myself) don't feel that it warrented a higher model rating.

As you know, the T-Bred Bs have a lot of overhead, and allows for a huge overclocked speed. Why my 1700+ (1.47 GHz) was able to hit 2 GHz! While that's dandy, the Barton shares the same T-Bred B ways, except for the added cache. This can result in either two ways: it will overclock just as good as the T-Bred Bs, or worse, because of the added complication of cache.

Fortunately, that was not the case, but since the 2500+ Barton (the slowest CPU that sports the Barton die) is already at a relatively high clock already (1.83 GHz), it is not as impressive as the 1700+ overclock. However, it certainly hits 2.2 GHz at 1.85v without any problems, air cooled. :) According to AMD, that should be in the same league as the Athlon XP 3200+! A CPU that was aiming for the Pentium 4 3.2 GHz, but of course, does not fair as well as Intel's true-blue GHz monster.

I have yet to do any benchmarks, but Prime 95 seems to be running flawlessly and without failures. Prime 95 is used by many enthusiest to test the stability of their overclocking work to ensure that what they did would stand
the test of data integrity.

I am currently still in the testing stage, but am very surprised at my findings. Here is the test bed:

Gigabyte GA-7N400 Pro2
AMD Athlon XP 2500+ (Retail Boxed) AQXEA Stepping
Corsair XMS 256 MBx2 DDR 433 MHz (512 MB total)
Running at5-2-2-2-2
ATI Radeon 9600 Pro 128 MB
WD 120 GB /w 2MB cache (non-SE)
Plextor 8x DVD+/-R/W (yes, I am gloating now)

FSB 400 MHz (x11 multiplier) = 2200 MHz
Heatsink: Thermaltake SK-7
Fan: Cheapo 80mm fan
Temps: 47*C idle, 55*C under load

All in all, a $90 CPU that does the work of a $460 CPU? Priceless. Well, not really. :) You are saving $370. Woo hoo!

UPDATE:
Using this config, I found out that the memory and FSB of the motherboard can go as high as 212 MHz (424 Mhz DDR) stable! Prime 95 ran flawlessly.

However, at the lowest mutliplier the CPU offers (x11), at that FSB, I would be running at 2321 MHz. At that speed, Prime 95 kept erroring out, and I had crashes. I slowly upped the voltage and found that the 10% overvolting kept it stable, but temps were soaring into the 70*C range! That's 150*F!!! No can do. But it was stable, though. Good to know for anyone who isn't afraid of other exotic cooling contraptions such as water or vapor.

After getting much frustration, my final config uses a standard (yet still overclocked from 333 MHz) FSB of 400 MHz, stock vCore voltage, and stock 11x multiplier, for a 2.2 GHz clock speed (3200+).

After all is said and done, let's get to the benchmarks, shall we?

SiSoft Sandra 2003
Arithmetic
Dhrystone ALU: 8234 MIPS
Whetstone FPU: 3303 MFLOPS

Well, the closest competitor is the Intel P4-B 2.8 GHz. Wow... really? A P4-2.8 GHz!? Dhrystone ALU at 8196 MIPS and Whetstone FPU at 2408 MFLOPS (non-SSE) and 5250 MFLOPS (with SSE)... AMD is making Intel run for it's money!

However, when compared to their basemodel Athlon XP 2600+ (at 2.13 GHz), their numbers are too close - 8028 MIPS and 3261 MFLOPS. Of course, Sandra also says that the 3000+ is running at 2.4 GHz. I guess if they used the older ratings and overclocked a T-bred B, it may be true. Because of this, the Sandra's 3000+ does in fact beat the pants off a 3200+ Barton. If you can find such a T-bred B that can overclock that high, that is...

SiSoft Sandra 2003
Multi-Media
Integer aEMMX/aSSE: 12225 it/s
Floating-Point aSSE: 13050 it/s

Compared to the 2600+, at Int 12689 it/s, and FP 12689 it/s, there is a strange trade off... the Int of my OCed masterpiece is actually slower than the 2600+ in INT calculations, while FP is slightly faster. Freak of nature?

SiSoft Sandra 2003
Memory Bandwidth
RAM Int Buffered: 3063 MB/s
RAM Float Buffered: 2893 MB/s

YES! I broke the theoretical limit of 2GB/sec of my previous motherboard/CPU combos! And at nearly 3GB/sec, too! Intel's 850E chipset (with RDRAM) running at PC4200 CL2 got Int of 3381 MB/s and Float of 3389 MB/sec. Whoa. It sure whooped me, but not by much.

In comparison with another nForce2 motherboard, which is only running PC2700 memory, their numbers are Int 2566 MB/s, Float 2893 MB/s. Yikes!

At last, a stable and quick machine, with numbers to prove it. :)

UPDATE (July 22, 2004):
It's been a very long time, and the new Bartons have become more refined, and are now have their multipliers bounded and gagged. Locked, sir... they are locked.

Still, the Barton 2500+ still remains the best buy for Athlon XP goers. Why? The default 11x multiplier is perfect for the 200 MHz FSB overclock! You still get the 2.2 GHz out of it (3200+ speeds), and because the new Bartons are more refined, they are now able to run stabily at this overclocked state without changing the core voltage! That's right folks - stock voltage.

In fact, the value leader chipset (Via KT600) is still a pretty good performer when paired up with this CPU. My friend bought an ECS KT600-A motherboard with this chip for only $70 at Fry's (where the chip itself costs $80 still) - both of them retail boxed.

Although it was not my PC (so i didn't push it as hard as I would if it was mine), he was comfortable (didn't try to go any higher) than 204 MHz FSB, with the resultant OC of 2244 MHz. Sandra benchmarked a bit faster than a traditional 3200+ (which is right in line with what we expected - good - the chipset is not hampering performance!). Again, stock voltage.

The hottest the chip got under full Prime95 load was 58*C. Idle was 42*C. This is using the stock HSF that came with the retail packaged CPU.

The 2500+ Bartons (even after being multi-locked) are the best value ever!

Recommended: Yes