In the desktop microprocessor market, you can't argue that competition isn't a good thing for us, the consumers. If the market itself were saturated with CPU manufacturers then competition may not be such a desirable affair, but in the case of the desktop x86 market, where AMD and Intel are the only two companies battling it out, competition isn't so bad.

The cost of owning a high performing CPU is not as great as it once was, and the acceleration of the product release roadmaps from both companies is driving the prices of the CPUs we all want to even lower levels in order to make way for faster, more robust solutions from both companies.

When we published our review of the Athlon 700 processor at the beginning of last month, we were told that AMD would have no other processor launches for the rest of the year and the 700MHz chip would be the flagship throughout the remaining portion of 1999. Little did AMD know that Intel was planning the release of a 733MHz Pentium III based on their Coppermine technology that would once again tilt the clock speed battle in favor of Intel.

To most AnandTech readers, clock speed isn't the final determinate of true performance of a system. An example would be the ability of a 500MHz Athlon CPU to outperform a 600MHz Pentium III CPU in 3D Studio MAX. However, to OEMs as well as the population of computer buyers that aren't so informed on architectural differences between processors, clock speed is king.

After Intel released their 733MHz Pentium III, OEMs that supported AMD and featured systems based on Athlon processors desired an Athlon solution that boasted a greater clock speed than Intel's 733MHz for the majority of users that judge performance solely based on clock speed. With the internal yields high enough on the 750MHz Athlon parts to move to full production, AMD fulfilled the wishes of their supportive OEMs and pushed for the release of a 750MHz Athlon processor in 1999.

Unfortunately, there are two problems associated with increasing the clock speed of the Athlon processor. The first problem is making sure that the core can operate reliably at that speed and that AMD can produce the chips at a high enough yield to remain profitable, and the second is making sure that the L2 cache can run at the higher frequency. As we're about to find out, for the first time in quite a while, the first problem isn't one that AMD lost much sleep over.

Pushing the core to 750MHz

From our testing with overclocking the Athlon, we found that hitting the 750MHz mark on the current Athlon core posed a small but navigable obstacle. The way we got around it in our tests was to simply bump up the core voltage from the default 1.60v to around 1.70v or above. The problem with this approach is that it isn't a very elegant way of paving the way for an increase in clock speed, and we didn't expect AMD to do the same with the Athlon.

If you recall from our original Athlon review, the Athlon core is a 22 million transistor monster that is built around a 0.25-micron die fabrication process, referring to the size of the circuit itself. Obviously, pushing the clock speed of the Athlon to 750MHz generates more heat than all of the previous Athlon clock speeds, and in order to gain high enough yields on the CPU without simply bumping up the rated core voltage, AMD felt that making the move down to a smaller process would be the best avenue of approach.

The migration down to a 0.18-micron process helps to shrink the size of the Athlon's die and decrease the generation of heat, as well as the power consumption of the CPU itself. The latter was actually a very large problem with the original Athlon because a large strain was placed on motherboards and power supplies not capable of supplying the system the current it needed to operate properly.

Thus, by looking at AMD's roadmap and the analysis we published right after Comdex, the introduction of the Athlon 750 is actually the introduction of the new K75 core that we talked about in that article. The K75 itself isn't much different from the K7 core; physically, it is a smaller die (102 mm^2 K75 core vs. 184 mm^2 K7 core) as a result of the 0.18-micron fabrication process but other than that the performance and features are identical.

The K75 core put to use on the Athlon 750 will be with the Athlon throughout the first half of 2000. While it will eventually take on an on-die L2 cache, the core itself won't be modified until the second half of 2000 when the enhanced K75 core is introduced in the codenamed Mustang CPU.

In summary, the move to a smaller 0.18-micron process has enabled AMD to hit the 750MHz mark flawlessly. A similar move was made by Intel with their Pentium III E CPUs (Coppermine) just one month ago. In both cases, the move was made seamlessly and there are no signs of keeping yields up on the CPUs.

0.25-micron AMD Athlon

Click to Enlarge

0.18-micron AMD Athlon

Click to Enlarge

Issue #2: L2 Cache Speed

Until AMD moves the L2 cache of the Athlon onto the die itself, they will have to depend on the cache manufacturers to deliver high speed L2 cache chips at a reasonable price for use with their CPUs. Intel got out of this situation after hitting the 600MHz mark with the Pentium III and Pentium III B, and, since then, all further Pentium III CPUs have featured an on-die L2 cache.

Unfortunately, the same can't be said for AMD with the Athlon. The Athlon is already at 750MHz, but it is still running with a 512KB external L2 cache that operates at a fraction of the CPU speed. Prior to the release of the 750, all Athlon CPUs ran their L2 cache at ½ the core clock speed.

The 350MHz L2 cache required for the Athlon 700 was difficult enough to come by because it required L2 cache chips with a rating faster than 3ns. In order to meet the 375MHz L2 cache requirements of a 750MHz Athlon, AMD would have to build the CPUs with 2.6ns L2 cache chips. As you can guess, this isn't too viable of an option for AMD as it would drive the prices of the 750MHz chips to unreasonably high and uncompetitive levels. So what was AMD's solution? Change the L2 cache divider.

Instead of running the L2 cache at ½ the clock speed of the CPU, the Athlon 750 runs the L2 cache at 1/2.5 or at 0.4 times the clock speed of the CPU. This 20% decrease of the L2 cache divider will definitely hurt the performance of the Athlon, in some applications more than others, but the question is how badly will it hurt the overall performance?

Performance Expectations & Compatibility

Business applications will take the largest performance hit as they depend on fast L2 caches, which is why the Celeron with a 128KB full speed L2 cache is capable of performing so well in business applications. Fortunately, business applications don't require too much power in order for them to run more than responsively so the performance hit won't make you cringe every time you fire up an office application. The Athlon 750 will still be faster than the Athlon 700, but not by a huge degree in business applications.

The slower L2 cache of the Athlon 750 versus the Athlon 700 (300MHz vs. 350MHz) will go relatively unnoticed by gamers since most games are heavily FPU dependent and not adversely effected by a 17% slower L2 cache in the case of the 750 vs. 700 comparison.

Professional level applications such as 3D rendering and imaging programs will proceed with a small performance hit, but the higher clock speed of the 750MHz Athlon will keep the negative effects of the slower 300MHz L2 cache from surfacing.

As we discovered in our overclocking tests, changing the L2 cache divider on the Athlon to 1/3 didn't affect the performance of the Athlon too badly at all. Because of those results we expected that the 1/2.5 divider wouldn't adversely affect the overall performance of the Athlon 750 to any noticeable degree. This is primarily because of the fact that the Athlon already has a very large 128KB L1 cache that operates at the full clock speed of the Athlon's core.

As far as compatibility goes, the Athlon 750 should work perfectly fine in all of the currently available Slot-A motherboards. AMD used the Gigabyte GA-7IX as the motherboard in the Athlon 750 Evaluation System they sent out, which takes the place of their reference Fester motherboard that they had previously used. At most, you'll require a BIOS update in order to support the Athlon 750 on your current motherboard.

In spite of the clock speed increase, the Athlon 750 should require less power than the Athlon 700 because of the die shrink. This may ease up the power supply requirements of the Athlon, but it is still recommended that you pay a visit to AMD's Recommended Power Supply list before purchasing a PSU for your Athlon system.

The performance of the Athlon has not changed since our last review, so feel free to visit our other CPU reviews such as the Pentium III 733 Review and the Athlon 700 Review for a full performance comparison of the Athlon to other CPUs. This article is mainly concerned with illustrating the performance advantage the Athlon 750 holds over the "slower" Athlon CPUs. For the sake of comparison, we have included two Intel CPUs in the tests at the bottom of each graph.

Making its first appearance in the AnandTech test setup is the ZD Content Creation Winstone 2000. As the name implies, Content Creation Winstone 2000 focuses on content creation applications such as Dreamweaver and Photoshop and is geared towards the more real world tests rather than running the same old business benchmarks over and over again. AnandTech will soon adopt a new standardized testing configuration, but until then here is a brief introduction to performance under the Content Creation Winstone 2000 tests.

The Athlon 750 enjoys a small but noticeable lead over the Athlon 700, the reason for the small size of the lead is primarily because of the slower L2 cache of the Athlon 750 counteracting the benefits of the faster clock speed. In the end, the Athlon 750 comes out on top of the Athlon 700, but not by a large margin.

A more noticeable performance advantage is held by the Athlon 750 in the SYSMark 98 tests, but the performance is nothing that we didn't already expect from the Athlon.

With the GeForce, the Pentium III seems to hold the advantage over the Athlon seemingly because of SSE driver enhancements in the NVIDIA Detonator drivers but the performance advantage could also be due to the L2 cache dependency of the GeForce. Regardless of the cause, the Athlon 750 isn't penalized much for the slower L2 cache but then again, the performance difference between each of the 6 Athlon CPUs is already very small.

UTBench.dem can be considered to be the equivalent of a crusher demo for Unreal Tournament. While it isn't nearly as long and involved as crusher was, UTBench does provide a good idea of how poorly a system can perform under stress in UT.

The Pentium III seems to hold a serious advantage over the Athlon in Unreal Tournament, one reason could be UT's ability to stress the L2 cache performance of a system. Other explanations could include heavy SSE optimizations in the UT code or in the GeForce drivers themselves.

The reason we were leaning towards an L2 cache performance explanation is because the performance of the Athlon 750 was identical to that of the Athlon 700, indicating that the slower L2 cache of the Athlon 750 was bringing down its performance. Speaking from the perspective of real world gameplay, the difference between the 6 processors is minuscule and any performance advantage the Athlon 750 would hold over the 700 wouldn't justify the price. In this case, the 750 has a very good purpose, that is, pushing down the prices of all of the other CPUs in the Athlon line.

The Business Application performance of the Athlon 750 suffers because of its slower L2 cache and thus gives it a meager 0.2 Winstone points advantage over the Athlon 700. As we already know, however, all of these processors are more than capable of running even the most complex business applications so there shouldn't be any performance concerns here.

HE Winstone's dependency on a strong FPU keeps the Athlon 750 ahead of the pack although it does take a small hit for the slower L2 cache in comparison to the Athlon 700. Overall, the difference in performance between the Athlon 700 and the Athlon 750 isn't enough to justify the increase in price. The physical difference between the two chips (smaller die, less heat generated, greater overclocking potential, etc...) does make the 750 much more appetizing.

Final Words

The conclusion still hasn't changed all that much from the original Athlon review. The 750MHz part is just another step in the Athlon ladder, although, this time, the jump from the previous clock speed champ, the 700, wasn't as great due to the somewhat slower L2 cache.

The chip is still quite a strong performer. The business application and gaming performances of the Athlon 750 are respectable, but it truly shines in the professional 3D rendering and imaging environment where it can't be touched by any single processor Intel system because professional applications take full advantage of the Athlon's superior FPU.

The 0.18-micron K75 core should be quite an overclocker. We've seen reports and had first hand experience with Athlon 500s that have been able to hit 650MHz flawlessly, and some even hit the 700 and 750MHz marks with nothing more than a regular heatsink/fan combo and a bump in the voltage to 1.75v. The 0.18-micron 750 should be able to hit some pretty high clock speeds using conventional air-cooling methods. The 750 is already the heart and soul of Kryotech's SuperG system which, by cooling the core down to -50C, manages to run the CPU at 1000MHz. It won't be long before more Athlon overclocking solutions become available on the market, and when they do, in conjunction with the Athlon 750, expect to be able to push the limits of the 0.18-micron K75 quite far. A 850MHz+ overclock shouldn't be far from reality with this CPU, especially as yields improve.

Looking towards the future, AMD can't keep the L2 cache off of the Athlon's core for much longer. Their roadmap calls for a move to an on-die L2 cache in the first half of 2000, but, currently, we do not know at what clock speed. While at Comdex, we were given a look at an air-cooled 800 and a 900MHz Athlon, so the potential to hit higher clock speeds using the 0.18-micron K75 core is there. Both of those CPUs will be released during the first half of 2000; hopefully, at least one will come with an on-die L2 cache which would help to rid AMD of their dependency on L2 cache manufacturers.

The Athlon motherboard issue does not seem to be improving and it will not improve in the remaining month of 1999, but, in 2000, things will change. From what we've seen, there is quite a bit of support building up for the Athlon in the motherboard industry, and the spark that we're waiting for is the release of VIA's KX133 chipset to ignite the fury of Athlon motherboards. The chipset itself is completed, and, right now, VIA is working with motherboard manufacturers to make sure that their implementations of the chipset are solid and are worthy of release. When the KX133 hits, it will be very good for Athlon supporters.

Going back to the point of competition, will Intel release a faster Pentium III in order to once again gain the clock speed advantage over AMD before the year is over? Most likely not, considering that there is only about a month left before the start of the new year. But then again, Intel has a recent habit of starting off the year with an onslaught of processor releases, so don't be surprised if Intel pushes the clock speed battle to the next level before AMD early on in Q1-2000.