Introduction

AMD is increasing the speed of their highest performing CPU today. The Athlon FX-57 is a 200MHz bump from the current FX-55, brining the clock speed of the highest performing single core CPU on the market to 2.8GHz. This modest 7.7% increase is not the be all, end all of speed bumps, but AMD is still in a much better position than Intel for extracting performance by tacking on an extra 200MHz. Intel's successive 200MHz increases on Prescott since it's existence have increased performance by smaller and smaller amounts 2.8 GHz to 3.8 GHz is a 35.7% increase in clock speed, which should be an overestimate of performance barring cache size increases. If we look at AMD's performance improvements from 1.8 GHz to 2.8 GHz, the upper bound on our performance increase is greater than 55%, plus any improvement for doubling cache size.

Performance doesn't scale exactly linearly with clock speed in most cases, but the bottom line is that K8 started out faster than Prescott at a lower clock speeds. The result of the speed increases on current generation CPUs has been an increasing performance gap between Intel and AMD in favor of AMD. Even in the benchmarks where AMD's architecture traditionally loses to Intel, the gap is either decreased or the outcome has changed all together. There just isn't any way Intel's current architecture can compete in single threaded performance on the high end.

But as we have mentioned time and time again, steadily increasing clock speed over time is a losing proposition. The future of computing performance must increasingly rely on architectural enhancements. The first incarnation of this outlook has been the introduction dual core processors. This first generation shows some promising numbers in many areas, but a single thread's maximum performance won't increase with the addition of cores. The result of this fact is that those who demand extremely high performance will still demand high end single core processors.

While the focus of the industry is clearly elsewhere, both AMD and Intel still need to cater to the current state of the market. The FX-57 is very expensive and comes in at a whopping $1031 (in quantities of 1000 from AMD). While this is the fastest processor on the market, let's take a look at the benchmarks to see how much we get for the money.

The Test and Business/General Use Performance
Comments Locked

56 Comments

View All Comments

  • cryptonomicon - Monday, June 27, 2005 - link

    what a beast
  • saratoga - Monday, June 27, 2005 - link

    The 533 vs. 400 would make an interesting benchmark. Given the availability of relatively low latency 533 (or at least 500), it might be worthwhile for people who must have the best performance (most FX buyers I'd guess).
  • dougSF30 - Monday, June 27, 2005 - link

    > Who is being realistic here?

    Hello? As you said *in your own post*, anyone splashing out $1000 for an FX-57 can afford to buy top-end memory for it.

    Otherwise, what's the point? DDR-333 is really cheap these days, maybe they should test with that?
  • Viditor - Monday, June 27, 2005 - link

    I have a feeling that the mobo AT used was a poor choice...
    I have checked all of the other reviews I could find, and every one of them that overclocked was able to be stable at 3.0 GHz on air...
  • Goi - Monday, June 27, 2005 - link

    I would've liked a bit more analysis rather than just describing the tests and displaying the results. I notice that this has become more or less an AT template in reviews, with little analysis and a lot of data. I think a bit more analysis on interesting points would be helpful. For example, the 200MHz bump from the FX-55 to the FX-57 causes disproportionate performance increases in certain benchmarks. An detailed analysis on why(going from the 130nm hammer core to the 90nm Venice core with improved memory controller for example) would be helpful I think. There's just too much numbers and not enough analysis IMHO.
  • Viditor - Monday, June 27, 2005 - link

    blckgrffn - I understand your point, but this is a Dream Machine chip anyway, so we should definately see it at it's best (latest mobo, DDR533 LL, etc...).
    JMHO
  • miketheidiot - Monday, June 27, 2005 - link

    this new memory controller has me thinking ddr2....
  • blckgrffn - Monday, June 27, 2005 - link

    For those of you who want DDR500 etc included, I don't think that you understand how/why these articles are written. I can appreciate that you want the whole platform to be about the best performance, but then we would lose all sense of just how much better this CPU is compared to the old ones.

    So, LL DDR500 performs awesome. Great. I suppose if you are buying a $1000 processor you will probably drop $250 on a gig of ram. Super. But for the rest of us, getting ram that runs 2-2-2-10 is hard enough, let alone trying for that super duper ram that runs in how many mobos due to voltage requirements? One or two? Who is being realistic here?
  • suryad - Monday, June 27, 2005 - link

    Looks like if AMD had some sort of HT scheme like Intel did...there would be no benchmark where the Intel would be ahead...but I am most impressed that AMD has now taken over the domain of multimedia and encoding and so on from Intel...cant wait for 3 ghz multicore FX procs!
  • dougSF30 - Monday, June 27, 2005 - link

    Dear Lord! Low latency DDR-500 or DDR-533 will peform better than low latency DDR-400!

Log in

Don't have an account? Sign up now