Tom's CPU Architecture Shootout: 16 CPUs, One Core Each, And 3 GHz


We went back in time for this piece and dug up a number of different processor designs. The idea was to explore how products from Intel and AMD have evolved over the past several years. Only, we needed to normalize the variables we were testing. To do that, we benchmarked each processor using a single core set to run at 3 GHz. The good news is that we're able to generate some interesting data on how successive generations of hardware perform on a per-clock basis. The bad news, if you can really call it that, is that many of the results end up being more theoretical in nature than anything else. Naturally, a threaded app will take advantage of more than one core, and if you can get more cores in an AMD processor than an Intel CPU at the same price, the performance situation is going to change.

Synthetic tests like 3DMark 11 and Sandra 2010 Pro leave no doubt that state-of-the-art workloads require state-of-the-art hardware. Given the results of Sandra's Cryptography suite, we know that anyone who makes use of AES-based encryption or decryption, for encrypting a system partition for example, should consider a processor able to accelerate that task. The same applies to enthusiast gaming. Tessellation and physics stress both the graphics card and the CPU, which means that multi-core processors are more relevant than ever before, even for mainstream users.

It's difficult to generalize about the results we did uncover; they differ from one workload to the next. And again, we're not meaning to suggest that just because one Clarkdale- or Lynnfield-based core might outrun a Phenom II core means a given Core i5 or i7 processor is faster than the AMD chip at a given price point. After all, AMD helps mitigate weaknesses in its performance per clock by adding more processing cores than competing Intel CPUs.

Although Intel winds up on top in most tests, the result ends up positive for AMD, too. Intel processors are generally much faster than AMD chips in terms of performance per clock per core. Responding to that missing performance by adding more physical cores per dollar puts AMD in the fortunate position where its four- or six-core Phenom IIs still serve up ample performance in single-threaded apps, but then take off in more parallelized tasks. Complemented by aggressive pricing, this helps to maintain good enthusiast value.

Yet, it's obviously time for AMD to introduce significant and noticeable improvements in its next desktop platform if it hopes to continue competing with Intel's steady cadence. The 22 nm Ivy Bridge die shrink is on its way, and it'll certainly result in higher clock rates. Power savings will probably be a hallmark of the evolutionary move, too. At this point, we’re afraid to say that AMD’s cores, individually, only beat Intel's old Pentium 4 or maybe even a Core 2 circa 2006/2007. But they don’t stand a chance against anything from Nehalem until now. Again, this refers to performance per clock per core and not to the multi-core products that we can buy. AMD has to get back into the game by not only infusing its processors with plenty of cores, but also by dramatically improving performance per clock.

At the end of the day, competition is what compelled Intel to come up with the Core architecture, first, and then Nehalem. We need competition. And we should know more about AMD's plans to compete within the next month or so when Zambezi hits our labs.