AMD is a long way from the glory days when it competed clock-for-clock with Intel’s CPUs. For the past few years it has sold its processors as value solutions, and while they are perfectly competent performers, they just haven’t had the grunt that Intel’s Core series has delivered.
Earlier this year AMD finally launched its Fusion APUs, which made the company competitive in the mobile space, smashing Intel’s Atom and nipping at the heels of the Core i3 range. Fusion is awesome, especially for gamers who can take advantage of the powerful on-chip GPU. But for now at least it cannot compete with the CPU performance delivered by Intel’s higher-end Sandy Bridge processors. For that we have been waiting years for AMD’s new CPU architecture, codenamed Bulldozer, which has finally launched in the form of the new FX series of AMD CPUs.
Bulldozer is a curiously different take on multiprocessing design, effectively sporting a hardware implementation of Hyperthreading. In Intel’s Hyperthreaded processors each core can process two threads, taking advantage of the fact that not every process uses every part of the processor. This leads to a situation where you can get a variable performance boost depending on the nature of the code being run.
With the FX, AMD has taken some of the philosophy behind hyperthreading and implemented it in hardware. Billed as an eight-core processor, it actually comprises four ‘modules’.
These modules have single instances of low-use circuits but double up on the high-usage parts; in this case the fetch and decode units are shared, as is the L2 cache. The actual integer and floating point pipelines are doubled up though (although both floating point pipelines can work together to support AVX extensions), so there shouldn’t be as much bottlenecking as there is on hyperthreaded Intel CPUs.
So while the FX isn’t truly an eight core processor, it is much closer to being one than Intel’s Core i7, for example. But this discussion is largely architectural – in reality we have seen that the number of cores doesn’t equal benchmark wins. For this we need to run up the new CPU and see just how it goes against Intel’s competition.
We have tested the top-end FX-8150 CPU. For our benchmarking we used an ASUS Crosshair V motherboard, 8GB of Patriot DDR3, Patriot Wildfire 120GB SSD and a Sapphire Radeon 6850. We used the drivers supplied to us by AMD, which at the time of writing were pre-release versions. For comparison we used both a Core i5 2500K and a Core i7 2600K, each running on a Gigabyte Z68X-UD3H-B3 motherboard, with the same SSD, Graphics card and RAM.
AMD has informed us that its target price for the FX-8150 is US$245, which means it will sit somewhere between the $230ish Core i5 and $340ish Core i7 in pricing.
Our recently revamped benchmarks are normalised around the Core i7 2600K, which scores one in our tests. On that scale the Core i5-2500K comes in at 0.94. When we ran the benchmark on the FX-8150 we got an overall score of 0.89, a touch behind the Core i5. In fact, when we look through the various processors that we have benchmarked, the FX-8150’s score is close to AMD’s previous top of the range CPU, the Phenom II X6 1100T – in fact, the results work out as less performance per core than on the X6.
When we look at gaming tests, traditionally AMD’s strength, there is little to no advantage over Intel’s offerings. Again we find that the choice of graphics card is much more important than the choice of CPU, with no noticeable difference between the FX-8150 and the Core i5-2500K. Whereas the company’s APUs deliver surprisingly good performance, the complete lack of integrated graphics on the FX stops it from having any real influence on gaming performance.
The reality is that, much like the other CPUs mentioned, the FX-8150 is capable of delivering ample across the board performance for the vast majority of people. However from a pure performance standpoint it hasn’t managed to dethrone Intel’s CPU dominance. Remember too that the launch of Intel’s new Socket 2011 enthusiast platform (known as Sandy Bridge-E) will likely push Intel further into the pure performance lead.
This doesn’t necessarily make the FX-8150 a bad CPU. For its pricetag you get an effective eight cores’ worth of high-end processing. Pair that with a discreet graphics card and you get a very robust gaming or high-end editing system for less than that many Intel cores. However this is purely a price-based decision rather than a technologically driven one; Intel’s offerings are just plain faster than AMD’s, everywhere that matters.
Bulldozer’s real day in the sun will likely come next year, when an updated version of the core, codenamed Piledriver, will form part of the new high-end ‘Trinity’ APU. As we saw earlier this year with the A series of APUs, when AMD combines its decent processor cores with its Graphics cores the result is a processor that plays to the company’s strengths, and makes up for lesser CPU performance than Intel with much better graphics performance.
Surprisingly, where the technological differences come in aren’t on the CPU, but on the wider platform. Despite being predominantly focused on the FX processor, AMD’s 990X chipset has been shipping for some months now. This chipset brings to bear some distinct differences from Intel. The first is native support for SATA 6Gbps connections. In Intel’s current chipsets only two ports support the latest SATA technology, whereas AMD is now in its second generation of full support for the tech. The 990X is also certified for USB 3, driving four ports natively.
Click to enlarge...
Our plan is to revisit Bulldozer once AMD actually starts volume shipments. Historically AMD has been about price performance, and we still can’t understand just why they decided to send the high end CPUs out for review. We suspect that somewhere further down the lineup lies a sweet spot, but at the moment there just aren’t the products around to test our theories.
We had hoped that AMD would do something groundbreaking with Bulldozer as it did with the A series APUs, but the reality is that the FX-8150 is an incremental step. It delivers great bang for the buck, but apart from taking a cores-per-dollar approach there’s no clear reason to recommend this over competing products.