Cache and Memory Performance

I mentioned earlier that cache latencies are higher in order to accommodate the larger caches (8MB L2 + 8MB L3) as well as the high frequency design. We turned to our old friend cachemem to measure these latencies in clocks:

Cache/Memory Latency Comparison
  L1 L2 L3 Main Memory
AMD FX-8150 (3.6GHz) 4 21 65 195
AMD Phenom II X4 975 BE (3.6GHz) 3 15 59 182
AMD Phenom II X6 1100T (3.3GHz) 3 14 55 157
Intel Core i5 2500K (3.3GHz) 4 11 25 148

Cache latencies are up significantly across the board, which is to be expected given the increase in pipeline depth as well as cache size. But is Bulldozer able to overcome the increase through higher clocks? To find out we have to convert latency in clocks to latency in nanoseconds:

Memory Latency

We disable turbo in order to get predictable clock speeds, which lets us accurately calculate memory latency in ns. The FX-8150 at 3.6GHz has a longer trip down memory lane than its predecessor, also at 3.6GHz. The higher latency caches play a role in this as they are necessary to help drive AMD's frequency up. What happens if we turn turbo on and peg the FX-8150 at 3.9GHz? Memory latency goes down. Bulldozer still isn't able to get to main memory as quickly as Sandy Bridge, but thanks to Turbo Core it's able to do so better than the outgoing Phenom II.

L3 Cache Latency

L3 access latency is effectively a wash compared to the Phenom II thanks to the higher clock speeds enabled by Turbo Core. Latencies haven't really improved though, and Bulldozer has a long way to go before it reaches Sandy Bridge access latencies.

The Impact of Bulldozer's Pipeline Windows 7 Application Performance
Comments Locked

430 Comments

View All Comments

  • Iketh - Thursday, October 13, 2011 - link

    I play FSX and Starcraft 2... both require copious processing power.

    And I just built an i7-2600K system with a radeon 6870 and blu ray writer for.... $960
  • paultaylor - Thursday, October 13, 2011 - link

    While the benchmarks are very revealing of the "ahead of its time" nature of Bulldozer, I think AMD should've kicked off by focusing on server applications instead of desktop ones.

    Considering what I've seen so far I think some additional benchmarks on threading/scaling would come in handy – it would actually show the true nature of BD as, right now, it’s behaving like a quad-core processor (due to the shared nature of its architecture, I presume) in most cases, rather than an octacore. Charting that out might be very revealing. The situation now looks like Intel's 2nd (3rd?) generation hyperthreading quad-cores provide more efficient multithreading than 8 physical cores on an AMD FX.

    Don’t get me wrong, we’ve heard from the beginning that BD will be optimised for server roles, but then we’re outside the feedback loop. Shouldn’t someone inside AMD be minding the store and making sure the lower shelves are also stocked with something we want?

    A longer pipeline and the old “we’ll make it up in MHz” line reeks of Netburst, unfortunately, and we all know how that ended. Looking at the tranny count, it’s got almost twice as many as the Gulftown, with 27% bigger die size for the entire CPU… which will mean poorer yields and higher costs for AMD, not to mention that either the fabbing process is really being tweaked or the speed bumps will not come at all, as the TDP is already high-ish. Ironically it reminds me of Fermi. Speaking of which… BD may become the punchline of many jokes like “What do you get when you cross a Pentium 4 and a Fermi?”

    On the other hand it seems AMD has managed one small miracle, their roadmaps will become more predictable (a good thing from a business perspective) and that will exert a positive influence with system integrators. Planning products ahead of the game, in particular in this 12-month cycle, might do some good for AMD, if they survive the overal skepticism that BD is currently "enjoying".

    Other than that, another fine unbiased article.
  • rickcain2320 - Thursday, October 13, 2011 - link

    Bulldozer/Zambezi seems to look more like a server CPU repackaged as a consumer grade one. Excellent in heavily threaded apps, not so hot in single threads.

    One CPU that is promised but isn't here is the FX-4170. I would have liked to see some benchmarks on it.
  • gvaley - Thursday, October 13, 2011 - link

    We all get that. The problem is, with this power consumption, it can't make it into the server space either.
  • kevith - Thursday, October 13, 2011 - link

    Having waited so long for this, it´s a bit disappointing, when I compare price/performance.

    I went from C2D E 7300 acouple of years ago, and changed setup to Athlon II x2 250, and the performance difference made me regret right away.

    And now, I have to change my MB and memory to DDR3 no matter what I choose, Intel or AMD. So I´ve looked forward to this release.

    And it makes my choice very easy: I´l go back to Intel, no more AMD for me on the CPU side. And Ivy Bridge is coming, and will definetely smoke AMD.

    Which is sad, it would have been nice with some competition.
  • eccl1213 - Thursday, October 13, 2011 - link

    Earlier this week most sites reported that the FX and BD based Opteron 4200 and 6200 where both being released on Oct 12th.

    But I haven't found a single review site with interlagos benchmarks.

    Have those parts been further delayed? We know revenue shipment happened a while back but I'm not seeing any mention of them in the wild yet.
  • xtreme762 - Thursday, October 13, 2011 - link

    I haven't bought an Intel chip since 1997. But with this BS bulldozer launch, that is now going to change! amd should be ashamed of themselves. I for one will now sell all of my amd stock and purchase Intel. I will probably only end up with a few shares, but at this point, I cannot see supporting liars and fakes. And I will NEVER buy an amd product again, not a video card, cpu, mobo, not nothing! What a disappointment amd is.....
    All the amd crap I have will be tossed in the trash. I'm not even going to bother trying to sell it. WooHoo amd made a world record OC with a cpu not worth it's weight in dog poo!
  • connor4312 - Thursday, October 13, 2011 - link

    Very interesting review. I'd be interested to see Bulldozer's benchmarks when it's overclocked, which, if I am correct, is higher than any Intel CPU can go. AMD seems to have made a turnaround in this aspect - Intel CPUs were historically more overclock-able.
  • Suntan - Thursday, October 13, 2011 - link

    As always, a very detailed review. But what about the capability of the "value" chips? Namely, is it worth it to spend around $100 to replace an Anthlon X4 with an FX4100?

    There are a number of us that picked up the X4 a couple years back for its low cost ability to encode and do general NLE editing of video. Is it worthwhile to replace that chip with the FX4100 in our AM3+ mobos? And what kind of improvements will there be?

    As you rightly stated, a lot of us are attracted to AMD for their bang-for-buck. Just because the industry as a whole wants to bump up prices endlessly, there are still a lot of us that like to see good comparisons of the performance of CPUs available for around 1 Benjamin.

    -Suntan
  • Pipperox - Thursday, October 13, 2011 - link

    Frankly, it seems to me the disappointment of AMD fans to be quite excessive.
    Worst CPU ever?
    What was then Barcelona, which couldn't compete with Core 2?

    Bulldozer, set aside old single threaded applications, is slotting between a Core i5 2500 and Core i7 2600K.

    Which other AMD CPU outperforms in any single benchmark a Core i7 2600k?

    A higher clocked Thuban with 2 extra cores would have been hotter and more expensive to produce.

    Setting aside AMD's stupid marketing, the AMD FX-8150 is a very efficient QUAD core.
    The performance per core is almost as good as Sandy Bridge, in properly threaded applications.

    Then they came with the marketing stunt of calling it a 8 core.. it's not, in fact it doesn't have 8 COMPLETE cores; in terms of processing units, an 8 core Bulldozer is very close to a Sandy Bridge QUAD core.

    The only reason why Bulldozer's die is so large is the enormous amount of cache, which i'm sure makes sense only in the server environment, while the low latency / high bandwidth cache of Sandy Bridge is much more efficient for common applications.

    I think with Bulldozer AMD has put a good foundation for the future: today, on the desktop, there is no reason not to buy a Sandy Bridge (however i'm expecting Bulldozer's street price to drop pretty quickly).

    However IF AMD is able to execute the next releases at the planned pace (+10-15% IPC in 2012 and going forward every year) THEN they'll be back in the game.

Log in

Don't have an account? Sign up now