The AMD Thoroughbred initially started with the Athlon 2200+. It marked the transition from the Palomino's 0.18 micron process technology, to the Thoroughbred's 0.13 micron process technology. What the process change was supposed to do was lower voltages, and to reduce the die size. Reducing the size of the chip would result in fitting more on each wafer as they're created. More chips mean more CPUs, and given the laws of supply and demand, hopefully cheaper CPUs for the rest of us.
Anyhow, we're not here to discuss Economics 101. Although the first Thoroughbred was based on the 0.13 micron fab process, there was not much to differentiate between the TBred and Palomino. Although the shape of the core changed, both the old and the new carried the same technology. Despite the expected lower voltages, well, it wasn't really lowered all that much and it still ran very hot, dissipating almost 70W of heat. Overclocking had some potential, but a huge cooler was required for any impressive overclocking. Finally, many enthusiasts were probably a little dissapointed that the TBred still rode on a 133FSB, which was of course, bottlenecking the DDR333 and DDR400 ram types that are so popular these days.
The TBred has matured a lot since June 2002. Commonly referred to as the Thoroughbred "B", the latest TBreds run much cooler than the initial batches, and are much more adept at overclocking than before. The current top-of-the-line TBred is rated at 2800+, and runs on a 166FSB (as well as the 2700+). As mentioned in our last paragraph, the previous Athlons bottlenecked the faster ram types, but now it should sync better with DDR333. There is a price to pay though if you want that speed, which makes another TBred "B", the 2400+ provided by our friends at , that we'll be looking at today a very popular choice. Unfortunently, the 2400+ is a 133FSB part, but as we'll see later on, we can fix that. :)
The Athlon Thoroughbred 2400+
Click to Enlarge
The TBred is based on the OPGA, or organic pin grid array, packaging technology. The benefit over ceramics (used in the Thunderbirds) is two-fold as it lowers the impedance and the cost of the CPU. Technology-wise, it carries the same featureset as the Palominos. You can grab the key architectural features on their . Here are sone of the highlights...
QuantiSpeed™ Architecture for enhanced performance: One of QuantiSpeed's key features is the advanced hardware prefetch. What this does is that it allows the CPU to anticipate, or guess, what will be done next and move that information to the cache for faster access. Say if you're working on a word document. The processor follows you as you work, and predicts what you're going to do next. Say that you plan to check for spelling. If the processor guesses correctly, your spell check will open a lot faster. Of course, if it guesses wrong, the cached data is discarded. Thanks to the shorter pipeline, this isn't as noticeable as it would be for the longer pipelined Pentium 4.
3DNow!™ Professional technology for leading-edge 3D operation: The original 3DNow! instructions are present in the Athlon XP, as well as 52 SSE instructions to take advantage of multimedia apps that use Intel's SSE. In reality, little improvement is present in day-to-day work with SSE & 3DNow!, but if you're a Photoshop "Blur" junkie, then you'll appreciate the extra instructions.
333MHz and 266Mhz: As mentioned earlier, the newest TBreds support the 333FSB (166FSB double pumped) natively, though the one we'll be looking at is a 266FSB part.
Performance-enhancing cache memory: There are 64K instruction and 64K data cache for a total of 128K L1 cache. Add 256K of integrated, on-chip L2 cache and you have a total of 384K full-speed, on-chip cache. A little more cache would certainly help, but we'll have to wait for Barton and Hammer for that.
Die size: approximately 37.6 million transistors on 84mm2: The die size is one physical change from the Palomino, which measured 128mm2, and was based on the 0.18 micron fab process. The TBred shrinks things down by about 55%, and is based on the 0.13 micron copper fab process. Like the Palomino, the TBreds come out of their facilities in Dresden, Germany.
Overclocking
As you may know, the TBred "B" CPUs have been overclocking like champs around the web. Not wanting to hold anything back, I slapped the biggest 80mm fan I had, a Delta SHE 68CFM, on our Swiftech MCX462+ and proceeded to testing.
The Epox 8RDA+ used for testing allows for PCI/AGP locking (which we did) as well as lowering the memory speed, percentage-wise in relation to the FSB. Although this may not be the best environment for maximum performance, our goal is to test the CPU's OC ability... not the ram. This is actually a technique that should be practiced, as once you've determined the maximum overclock of the CPU/motherboard, you can start tweaking the ram to get the right balance.
The first step will to simply see what we can do out of the box. The default multiplier is 15, and the default FSB is 133. Our first attempt was to go for a 166FSB.
Posting at 166FSB wasn't a problem, but booting into windows was a whole other story. We were successful at 165FSB though. Thumbnails are clickable...
15x165
A 487MHz overclock was pretty satisfying, although I was hoping for a 500MHz+ OC. Just in case you're wondering, vCore was at 1.95v. We ran though our usual gamut of benchmarks without any issues, and the system was rock solid. Not wanting to rest at 2.487GHz, it's time to unlock the multipliers, and push the FSB harder.
Unlocking a TBred requires connecting the L3, #5 bridge. This will unlock multipliers 12.5 and under. I should also note that the Epox 8RDA+ unlocks TBreds without modifications. We managed to hit 12.5x201, but Windows simply refused to POST. We started dropping 1MHz at a time before settling on 12.5x198.
12.5x198
The resulting overclock is 486MHz over stock, which is about the same as we got with 15x165. Perhaps an even bigger cooler may have netted us more success, but considering the more expensive 2800+ is clocked at 2.241GHz, we were generally satisfied with our end result.
Test Setup
Athlon XP 2400+, 2.0GHz (Provided by )
Epox 8RDA+ nForce2 Motherboard
2 x 256 Crucial PC2700 Ram
ATi Radeon 9700 Pro
120GB Western Digital SE 8MB Cache
As will all our reviews related to benchmarking, a reformat and reinstall of Windows XP SP1 is done. The OS is a default installation, with no optimizations, in order to easily maintain a consistent test bed, and for you, the readers, to know what our settings are, should you wish to compare your results. The same with drivers, no tweaking is done.
Ed. Note: There seems to be some confusion about the last paragraph. We do run the latest drivers and OS updates with all benchmarks. We also disable all unnecessary background apps. What we don't do is turn off dozens of Windows services, and tweak video drivers to best performance (we leave it at balanced). We don't do this because everyone has their own way of doing it, so your numbers may be higher or lower than ours.
That being said, motherboard settings are configured as optimal as allowed by the BIOS. If the system is unstable, we'll drop a setting back, but we do this mostly to show motherboard/CPUs in their best light.
Drivers used will be the most recent as of this writing (ex: Catalyst 3.0 for Radeon).
Test software will be:
SiSoft Sandra 2003
PC Mark 2002
3D Mark 2001SE
Unreal Tournament 2003
Quake 3: Arena
Jedi Knight 2
Except for 3D Mark, all game tests are done at 640x480 to eliminate the video card as a possible bottleneck (though the Radeon 9700 will not likely have any problems at any resolution). Quake 3 and Jedi Knight 2 were run at "Fastest" settings.
For the benchmarks, we will be demonstrating stock 15x133FSB (2400+), and an overclocked bus @ 12x166 (2400+). We will be adding in some Pentium 4 2.4B (Shuttle XPC SS51) numbers for the heck of it, since I'm sure many of you would like to see comparative scores.
SiSoftware Sandra 2003
Although a synthetic benchmark, it's a popular one, freely available if you wish to make comparison benchmarks. We will be testing the CPU, and MMX speeds.
CPU Arithmetic Benchmark
The scores between both 2400+ ratings were very close. Both beat the Pentium 4 quite decisively, but the Whetstone FPU is owned by the Pentium 4. The FPU has always been one of Intel's strengths, but the victory is a narrow one.
CPU Multimedia Benchmark
Very close again between all 3. The Athlon 2400+ pulls away, by a long shot, in the Integer tests, which is something AMD have always done well on. Between 133FSB, and 166FSB @ 2GHz, there is no clear winner.
PC Mark 2002
As with the SiSoft scores, our PC Mark scores follow a similar trend. 2400+ speeds are close, with the 166FSB giving an extra boost. The Pentium 4 falls a little behind this time around.
3D Mark 2001SE, 1024x768 - Default
3D Mark is one of those benchmark apps that will eat up as much processing power as it can. Benchmarks were run at default 1024x768, as I felt 640x480 is getting to be fairly pointless. The 166FSB gives an extra 200 3D Marks or so, showing that clock speeds being equal, the faster FSB will net better performance.
Unreal Tournament 2003
I've been playing around with the retail version of UT2K3 for quite sometime now, and have been pretty impressed with the graphics. It's a real system killer, and can bring many pre-2002 killer rigs to its knees. We used the scripts written by , which are excellent tools in testing various resolutions and detail levels. We selected the CPU test, which uses the dm-inferno map.
At the risk of sounding like a broken record... higher FSB owns.
Quake 3 Arena, 640x480 - Fastest
It's getting old, I know, but Quake 3 is still a decent benchmark for almost anything. For motherboards and CPUs, we run at the lowest settings and fire away. We're treading into Intel territory, and Intel holds the fort against AMD here.
Jedi Knight 2, 640x480 - Fastest
Jedi Knight is still a Quake 3 engine game, so naturally, the P4 wins again. The difference is smaller this time around, but nevertheless, the winner is Intel.
Final Words
Though it didn't go all that well for Intel, save for Q3 and JK2, in defence of the P4, the 2.4B is an extremely overclockable CPU, and with the right mobo and cooler, it should be enough to get it at . This isn't an AMD vs Intel article though, so here's our analysis of the TBred 2400+...
At a stock clock of 2GHz, 486-487MHz boost in speed is pretty much a lock if you got a solid air cooler behind it. Peltiers and watercooling will probably allow you to push it higher. Unlocking a TBred the traditional way for overclocking is still a bit of a chore, though much less tedious than unlocking a Palomino. You only have one bridge to connect now, whereas the Palomino required five. There are some motherboards out there that can unlock a TBred (the Epox nForce2 is one that I know of) without any modifications to the CPU.
The TBred "B" also runs much cooler than previous Athlons. At 2.486GHz, our load temperatures were about 49C. The Athlon 1700+, using the same cooling was at 44C, but for an extra 530MHz stock, the TBred was a mere 5C hotter. In fact, OCing the 1700+ (1.47GHz) to 1.66GHz resulted in a 48C load temperature. I think you're getting the picture.
Performance is simply impressive. At stock speeds, the 2400+ did pretty good, and overclocked netted excellent performance. This speed also comes with stability, so long as you got a cooler up to snuff. If performance of a stock 133 bus concerns you in regards to bottlenecking your DDR333 ram, the 2400+ easily does 12x166 anyways, and as our benchmarks have shown, the extra bandwidth made a difference.
If there's one complaint, it's the core is just as fragile as the Palomino, as was the Thunderbird before it. Although we feel pretty confident testing a dozen heatsinks a month, there are a lot of people who sweat over changing their HSF even once. Unfortunently, their stress will continue, as a gaff in installation of a heatsink can still result in a busted CPU. Don't misunderstand though, it's not like a sneeze will crack the core, but do take some care. For the uber-paranoid, consider that if you purchase an Athlon from , and you break it, they will replace your CPU no matter what condition you return it in. Just don't lose that receipt.
When it comes to pricing, it's pretty hard to beat the 2400+. With overclocking performance surpassing a stock 2800+, and at a much lower cost, budget shoppers should definitely give the 2400+ a very close look.
Pros: Very speedy, very overclockable, and a price that won't break the bank.
Cons: Still prone to core cracking if you're careless.
Bottom Line: Although we didn't break a 500MHz+ overclock, it's hard to find a fault with a CPU with this much performance selling at half the price of an Athlon 2800+. If the P4s, and top end AMD TBreds are a little too rich for you, pick the 2400+ up with a nice nForce2, and you'll have money left over towards a GFFX or an ATi Radeon.
We'd like to thank for making this review possible.
If you have any comments, be sure to hit us up in our forums.
HOME
AMD Athlon XP 2400+: The 2800+ may be faster at stock speeds, but you'll be surprised what this "budget" CPU can do at a fraction of the price. Oh, it also overclocks like a champ.
