FX series

So far, AMD introduced four different representatives of the new FX series processors. They are all based on the Zambezi codenamed die, an 8 logical core native processor, featuring four Bulldozer modules, described on previous pages. All of the FX models in the near future will be based on Zambezi, those already introduced and those just announced. To optimize and cut down the manufacturing costs, AMD will make only one die, the 8 core one, which will serve as base for not only desktop SKUs but also all of the server SKUs based on Bulldozer architecture. This means fewer core SKUs will have one or two modules disabled, resulting in six and quad core processors, and it would seem the count will only go so low as AMD has yet to announce any plans for dual core models.



Zambezi, with all of its modules, northbridge, improvements and implementations, ended up 315 mm² big, which is less than 346mm² of six core Thuban core, but at the same time it's far more than 216 mm² of Intel's Sandy Bridge die which is its main desktop competitor. It's built by Global Foundries on the new 32 nm SOI High-K metal gate process, same as Intel is using with current Sandy Bridge and Sandy Bridge-E processors. Smaller size means cheaper production, and to compensate smaller volume fabrication capabilities than Intel, AMD chose a very tricky path for Bulldozer architecture. It's clear that Bulldozer's architecture is very, very multithreading-oriented, and it will shine most in that kind of workload. Interestingly, that kind of workload is most often seen in the server market, much less on client-desktop market. From current experiences with games, tools and applications we can say multithreading support has been accepted quite well with application developers by now, but we also know that very few apps receive a performance boost by just adding more cores to tasks, games for one, or 90% of every day tasks for that mater.

So AMD played a double sided card. On the one side, Bulldozer has everything it takes to be a fully fledged server processor, with improved power management, existing server socket compatibility and multithreading workload optimizations; on the other, it's capable of running at very high clock frequencies, featuring Turbo Boost features, unlocked multiplier for easier overclocking, and a few other perks desktop users will hopefully for AMD grow fond of.



And for the end result AMD got a single core die for manufacturing, shipping as is for desktop FX 8000 series processors, or disabling one module with 2 cores for FX 6000 series and 2 modules with 4 cores for FX 4000 series. As for server offerings, Valencia core die is the Zambezi server counterpart featuring 4 modules and 8 cores, together with all of the features Zambezi has to offer. Interlagos on the other hand has two Valencia modules fused together and interlinked with available HyperTransport links, featuring 8 modules and 16 cores.

Currently AMD's new generation processor offerings consist of quad core FX-4100, six core FX-6100 and two eight core models, FX-8120 and reviewed FX-8150. The quad core model features two Bulldozer modules with 4 cores, 4 MB of L2 cache and 8 MB of shared L3 cache memory. Stock frequency is set to 3.60 GHz with Turbo Core reaching just 3.80 GHz, and Northbridge is set to 2.0 GHz. TDP is rated to 95 W.

The FX-6100 has three Buldozer modules enabled, equaling six cores in total, with 6 MB of total L2 cache and 8 MB L3 cache memory. Stock frequency is set to 3.30 GHz and with Turbo Core it can reach up to 3.90 GHz under TDP of 95 W. Its Northbridge speed is also set to 2.0 GHz. The slower of two eight core models, the FX-8120 has all modules enabled featuring eight cores, 8 MB of L2 cache memory and 8 MB of L3 cache. Its stock clock is set to 3.10 GHz and Turbo Core can take it up all the way to 4.0 GHz, while TDP will be rated both 95 W and 125 W, depending from sample to sample. Supposedly, the 125 W model will be somewhat cheaper in stores compared to more efficient 95 W model. Its northbridge speed is clocked to 2.20 GHz.

Previous generation FX processor models always had an unlocked multiplier so the new FX generation continuous that tradition, and all upcoming models will have unlocked multipliers as well. In one way, AMD with this move is carrying on the tradition of FX to be more user friendly, for PC enthusiasts and create a new end user pool - budget overclockers.
Remember, on Intel platforms you'd need to pay a large sum of money to be able to overclock your processor, while with AMD, you can play around even with the FX-4000 series.


And finally, the crown jewel of the new FX series, the FX-8150. For the last few years AMD has been always two steps behind Intel both in top performance and manufacturing technology. After so many years in development, Bulldozer should have changed that, or at least make it a one step behind Intel instead of two. Instead of that, AMD made so many compromises between desktop and server needs, and its desktop flagship model ends up positioned opposing Intel's middle end market offerings - Core i5 2500K and Core i7 2600K. Not something we expected from Bulldozer, but that's what we got.



Based on Zambezi core, featuring four Bulldozer modules with total of eight cores with 8 MB L2 cache (2 MB per module), shared 8 MB L3 cache, and clocked to not that impressive 3.60 GHz. As our theoretical part of the review hinted, Bulldozer based processors have to run at very high operating clocks with least amount voltage used doing so, to be efficient. Since FX-8150 and its 3.60 GHz is as high as AMD can go at this moment, it's safe to say that they had serious problems overcoming the first obstacle. Quick bursts of speed with Turbo Core to 3.90 GHz and 4.20 GHz at its best help the situation somewhat, but for Bulldozer to really achieve what it was designed for, AMD needed a flagship running at base clock of 4.20 GHz and bursting to 4.80 GHz with Turbo Core. Then we could forgive higher power consumption figures, because it would really be a processor worthy of carrying the FX name, and flagship SKU.

But instead, FX-8150 ended up as a upper middle end market product comparing to Intel, and in some points it even has trouble besting previous generation Thuban flagship, Phenom II X6 1100T. Not to be gloomy, FX-8150 has is bright sides, like a new architecture optimized for multithreading, so it's kind of future proof performance wise. The power saving features work great and it does keep its consumption in check while in medium workload, and while loading work on cores it does give away a feeling like it's not breaking a sweat while crunching numbers and it's just asking for more work to do. It's not a slow processor, on the contrary, it's lightning speedy in every way. The unlocked multiplier comes in handy too, as these processors love to overclock, if you have the right tools and knowledge. And for those more of a lazy types, a simple change in multiplier could increase the performance quite a bit.



Operating voltage is set to 1.385 V as base value for base P state, but varies from 1.00 V to 1.40 V depending on the core P state or Turbo Core. TDP rating is set to 125 W and that should be the maximum consumption regardless of Turbo Core active or not, if the settings are left at processor default. The northbridge frequency on this model is set to 2.20 GHz and it should provide some performance boosts via faster L3 cache and more memory bandwidth compared to 2.0 GHz NB speeds.

AMD priced the FX-8150 between Intel's Core i5 2500K and Core i7 2600K, both very competitive and advanced processors, each dominating in its own market segment, so keep an eye on performance ratios between those models.